Image conversion flow

• Select the Image tab at the top of the right column.
• Choose the Preset, the output Format (WEBP, AVIF, JPG, PNG, ICO or TIFF) and the Scale.
• Adjust the format-specific advanced controls if you need to (the details of each one are in the Formats tab).
• Click Convert. You can cancel during the process.
• The generated files appear in the right-hand column.

Supported input formats

You can load image files in the following formats: JPG PNG GIF WebP BMP TIFF HEIC HEIF AVIF SVG ICO and camera RAW. HEIC/HEIF files are decoded with heic-to (libheif) and TIFF with UTIF.js (for a multi-page TIFF the first one is used) before going to the selected encoder. SVGs are rasterised to the requested size.

Camera RAW (DNG CR2 CR3 NEF NRW ARW ORF RW2 RAF PEF SRW and others LibRaw knows): they are developed on load with LibRaw (LibRaw-Wasm, ~1.5 MB downloaded with the first RAW) using the camera white balance, in 8-bit sRGB and with the camera orientation applied. After that they are handled like any other image. Developing a 24 MP RAW takes a few seconds; the RAWs of one load are developed one at a time. A RAW that LibRaw cannot open does not enter the queue and a warning is shown. LibRaw uses threads, so it needs the page to be isolated (COOP/COEP), like FFmpeg.

Scale and dimensions

The Scale slider adjusts the size as a percentage of the original. Click the Scale label to open the advanced modal with the following options:

• Numeric percentage: Enter the exact value or use the /2 (half) and x2 (double) buttons for quick adjustments.
• px mode: Defines how the target width and height are interpreted. There are five modes, in two families.

The three single-factor modes use a single field (percentage, width or height) and deduce the other side from the original ratio, so they never distort, but they also do not let you set width and height at the same time:

• Maximum: it will not exceed the given dimension; images that are already smaller are left intact.
• Minimum: it will cover the given dimension; images that are already larger are left intact.
• Force: applies the dimension strictly, in both directions.

In those three the active field priority applies: the last field you edit becomes the active rule and the inactive ones are shown struck through.

The two fit modes use width and height at the same time to reach an exact dimension without distorting:

• Centred crop: scales until it covers the requested width and height and crops the excess from the centre. Exact size, no bars and no distortion, at the cost of losing part of the image. Ideal for square thumbnails and grid galleries.
• Letterbox: scales until the whole image fits and fills the gap with bars. Exact size, no distortion and nothing lost, at the cost of the bars. Ideal for fitting mixed-ratio material into a common canvas. The Bars control lets you choose their colour or mark them transparent (in JPG, which has no alpha, they are filled with white).

In the fit modes the percentage does not come into play and its slider is disabled. If you only fill in one of the two fields, both behave as Force. The width and height always refer to the final image: with the image rotated 90° or 270°, the crop and the bars are calculated on the already rotated result.

Interpolation

The Interpolation drop-down in the scale modal decides how the new pixels are calculated when resizing:

• Smooth (photography) — averages the neighbouring pixels. It is the default behaviour and the right choice for photographs and any image with gradients: the edges come out clean and without jaggies.
• Nearest neighbour (pixel art) — copies the closest pixel without blending colours. The edges are kept absolutely hard and no new colour is introduced. It is what is needed in pixel art, screenshots, graphics with small text, single-pixel lines or diagrams: with smooth interpolation, enlarging pixel art blurs it and shrinking a screenshot smudges the text.

Note: when enlarging, Nearest neighbour produces the classic square-pixel effect. If what you want is to enlarge a photograph without it looking blurry, the right tool is not interpolation but the Upscale described below.

Upscale (enlargement)

The Upscale checkbox enlarges the image before fitting it into the target dimensions, with a 2x or 4x factor. It only acts when the target is larger than the source, and on the already cropped area. Two engines:

• Lanczos3 (fast): Lanczos resampling with radius 3 in its own worker. No downloads, works offline and takes seconds. It gives much cleaner edges than the browser's scaling, but it does not invent detail that was not already in the image. It is the default engine.
• Real-ESRGAN (AI): super-resolution neural network with onnxruntime-web, by overlapping tiles. It does reconstruct texture and edges, but it downloads about 105 MB the first time (then cached) and needs WebGPU to be practical: without it, it falls back to a single-threaded CPU, in the order of tens of minutes per megapixel, and you are warned. Your images still do not leave the machine: the only thing that travels is the model download.

The modal warns of the expected cost according to engine and factor. During conversion the file's progress bar reflects the phases (download, initialisation and inference) and Cancel aborts instantly.

Remove background

The Background cell opens the modal to remove the background of each image with a segmentation neural network (onnxruntime-web) that runs on your machine: it leaves the subject with transparency. Two engines:

• Fast (U²-Netp): downloads 4.6 MB the first time and takes a couple of seconds per image. It computes the mask at 320 px, so on hair or fine details the edge is somewhat soft.
• Quality (ISNet): downloads 178 MB the first time and takes in the order of 10–20 s per image. Mask at 1024 px, with much more precise edges, hair included.

Edge contracts (to remove the background halo) or expands the mask, and Smoothing blurs the cut; both in image pixels. Fill places the subject over a colour instead of transparency (JPG, which does not support transparency, comes out with a white background if you do not choose a colour). Crop to subject fits the frame to the subject before scaling. To keep the transparency, convert to PNG, WebP, AVIF or TIFF.

The network is applied after the edits to each image (erasing, straightening, cropping) and before scaling. Only the first conversion of each image runs it: changing the edge, the fill or the format reuses the mask. The models are cached and are deleted from Settings.

Max size

The Max size cell sets a size ceiling per image (in KB, MB or % of the original). When converting, the highest quality that fits is sought, without going above the chosen one or below 40; if that is not enough and it is allowed (by default), the dimensions are reduced just enough. Lossless PNG and WebP can only reduce dimensions. The output row states the quality it ended up with, or warns in amber if it did not fit.

Image adjustments

The Adjustments cell opens a modal with a selector, nine sliders and a preview of the reference image, already edited and resized, reduced to the size of the modal, and below it its histogram (red, green and blue, with the luminance as a line):

• Automatic — Levels stretches each channel of each image from black to white and corrects colour casts; Contrast uses the same points for the three channels and keeps the colour.
• Temperature and Tint (−100 to +100) — white balance: warmer or cooler, towards magenta or towards green.
• Brightness (−100 to +100) — lightens or darkens the whole image equally, shadows included.
• Contrast (−100 to +100) — spreads out or brings together the light and dark tones around mid-grey; −100 leaves a flat grey.
• Highlights and Shadows (−100 to +100) — lighten or darken the light or the dark tones without moving white or black: Shadows + brings out detail in a backlit shot; Highlights − recovers a sky.
• Saturation (−100 to +100) — colour intensity; −100 converts to greyscale and +100 doubles it.
• Gamma (0.20 to 3.00) — lightens (> 1) or darkens (< 1) the midtones without moving black or white. Useful for underexposed photos.
• Sharpness — positive enhances the edges (unsharp mask); negative blurs. It is calculated at the final size, so it is worth checking in the comparator.

They are applied to all the images when converting, after the edits to each image and the resizing, and they do not touch the letterbox bars. A double click on a slider returns it to its neutral value; Reset returns them all.

Compose images

The + button next to Convert puts the selected images together into one, in queue order, and adds it as a new row. The Compose (+) cell sets the Type (grid, contact sheet with the name or number under each cell, or sprite sheet with cells the size of the largest image), the columns (0: automatic), the cell and its fit, the spacing and the background. Each image comes in with its edits and the colour adjustments, but the cell sets the size; the output uses the tab's format and quality.

Optimize SVG

With the Optimize SVG checkbox, the SVGs in the queue come out as SVG optimized with SVGO instead of being rasterized; the rest follow the chosen format. Comments, metadata and editor leftovers are removed and numbers and paths are shortened, without touching the ids, the viewBox or the titles. Crops, rotations, scale, adjustments, background and max size are not applied to those SVGs.

Processing order

Each image goes through the stages in this order: decoding (HEIC, TIFF, SVG…) → areas erased in the viewer → straightening → viewer crop → remove background (and crop to subject) → upscale, if enlarging → scale and px mode (centred crop or letterbox), with the rotation and flip → colour and sharpness adjustments → encoding in the chosen format → max size, if active (repeats the encoding with another quality or size) → output metadata.

Individual and batch editing

Each image can be edited in the viewer (the eye icon in its row): rotate, flip, crop with a fixed or free ratio, straighten at a free angle and erase areas filling them with colour. The edits are saved in the file and are applied when converting; the details of each tool and its shortcuts are in the Advanced tab. For several images at once, the batch bar under “Imported” rotates, flips and edits the metadata of all the selected ones (see General).

After converting, the output viewer lets you compare with the original with a sliding divider and calculate the SSIM (see Advanced).

ICO / Favicon

The ICO / Favicon format generates a .ico with the sizes 16, 32, 48, 64, 128 and 256 px, all inside the same file. With the Favicon pack (.zip) checkbox enabled, the output is a web-ready ZIP with favicon.ico, apple-touch-icon.png, icon-192.png, icon-512.png and a site.webmanifest. It is best to start from a square image of at least 512 px.

TIFF

The output TIFF is always lossless. Controls:

• Compression — LZW opens in any print and layout program; Deflate (ZIP) weighs somewhat less and is read by Photoshop, GIMP, Affinity and libtiff; Uncompressed is the most compatible and weighs 3 bytes per pixel (4 with transparency).
• Resolution (72, 150, 300 or 600 dpi) — the one noted in the file, which the layout program or print shop will take for the printed size. It does not change the pixels; that is what Scale is for.

Browsers (except Safari) do not show TIFF: TIFF outputs have no preview or clipboard copy, but they download normally.

JPEG XL

JPEG XL weighs considerably less than JPG at the same quality and supports transparency and a lossless mode. It is encoded by libjxl (@jsquash/jxl, ~1.4 MB downloaded when you choose the format). Controls: Quality (cjxl scale: 90 is visually lossless), Effort (1–9; 7 by default), Lossless and Progressive.

Few browsers display JPEG XL (Safari does). For the thumbnail, the preview, the comparison and Copy to clipboard, the output is decoded again with the libjxl decoder and shown as PNG: what you see is what was encoded. The download, the ZIP and Share deliver the .jxl. No metadata is written to the JPEG XL.

Per-format controls

Each format shows its own advanced controls panel when selected. The exhaustive detail of each parameter is in the Formats tab. Quick summary:

• WEBP: Quality, Method, Near-LL, Alpha, Filter and Lossless mode. Encoder libwebp.
• AVIF: Quality, Speed, Denoise, Sharpness, Subsample and Alpha. Encoder libavif.
• JPG: Quality, Subsampling, Smoothing, Q Table, Trellis and Grayscale. Encoder MozJPEG.
• PNG: Optimisation (1–6), Colours and Interlaced. Encoder OxiPNG; lossless unless the colours are reduced.
• ICO / Favicon: Pack checkbox (ZIP with favicon and web icons).
• TIFF: Compression (LZW, Deflate or uncompressed) and Resolution in dpi. Always lossless.

Image presets

Optimised

Quality/size balance intended for general web use. Recommended starting point for most images.

Max Quality

Prioritises visual fidelity over file size. Suitable for images that are going to be viewed at high resolution or that require minimal degradation.

Min Size

Maximises size reduction with aggressive compression. Ideal for thumbnails, background images or contexts with limited bandwidth where size is critical.

Lossless

Lossless encoding: reproduces the original pixels exactly. In WEBP and PNG it guarantees perfect quality; in AVIF and JPG it gets as close as possible with their encoders.

— Custom —

It activates automatically when any control is modified manually. Selecting a preset again restores its default values for the active format.

Video conversion flow

• Select the Video tab at the top of the right column.
• Choose the Preset, the output Format (MP4, MP4 HEVC, WebM, MKV, animated GIF or animated WebP) and the global controls (Resolution, FPS, Audio, Filters).
• Adjust the format-specific advanced controls if you need to (the details of each one are in the Formats tab).
• If you want only a segment, open it in the viewer (eye icon) and set the start and end of the trim, or use the percentage trim in the batch bar.
• Click Convert. You can cancel during the process.
• The FFmpeg.wasm engine (~33 MB) is served from the site itself; the first video conversion downloads it and the following ones are immediate.

Trim

In a video's viewer, under the player, a bar appears with two handles to set the start and the end, and two time fields in mm:ss.ms format. The playback line marks where you are and the label shows the duration of the segment. The reset button removes the trim and returns to the full clip. The conversion generates only the chosen segment (also in GIF and animated WebP). For several videos at once, the batch bar accepts the trim as a percentage of each one's duration (see General).

Video filters

The Filters cell opens a modal with filters that are applied to all the videos when converting, also in GIF and animated WebP. The preview only shows the colour, approximately. A double click on a slider returns it to its neutral value; Reset returns them all.

• Brightness, Contrast, Saturation and Gamma — the same scale as the image adjustments: −100 to +100 (saturation −100 leaves the video in black and white) and gamma from 0.20 to 3.00.
• Noise (None, Light, Medium, Strong) — noise reduction with hqdn3d. It smooths the grain of low-light recordings and usually lowers the size; at high levels fine detail is lost.
• Deinterlace — removes the horizontal lines from interlaced videos (DVB, old cameras). Automatic only processes the frames the file marks as interlaced; Always processes them all, for mislabelled files.
• Stabilise (slow) — compensates for handheld camera shake (deshake). It is a basic stabiliser that makes conversion considerably slower; the edges left exposed are filled with the reflected image.
• Speed (0.25× to 4×) — speeds up or slows down the video (setpts) and its audio (atempo, which keeps the pitch). The output duration and the fades are worked out on the new timing; in slow motion each frame lasts longer, no new ones are invented.
• Reverse (short clips) — plays the video and its audio backwards (reverse, areverse). FFmpeg holds the whole segment in memory before writing the first frame: if it does not fit (about 10 s at 720p and 30 fps), the conversion stops before starting and says how many seconds fit.
• LUT (.cube) — loads a 3D colour table (2–65 points per axis, normalised domain) and applies it with lut3d. The strength (0–100) is baked into the entries when loading: 0 leaves the video unchanged and 100 applies the LUT as-is. It is validated on load (1D LUTs, out-of-range sizes and files over 6 MB are rejected) and the file is not stored in the browser. WebCodecs does not apply it, so its presence forces the FFmpeg path (or the intermediate copy for VP9/AV1) and the conversion is slower.

With Speed or Reverse the audio is re-encoded even when set to Keep (at 320k). When joining videos they apply to the whole, after concatenating.

Extracting audio from a video

With a video selected, switch to the Audio tab and convert: only its audio track comes out, in the chosen audio format. The output row is presented as audio.

Supported input formats

You can load video files in the following formats: MP4 WebM MOV MKV AVI OGV. In general FFmpeg.wasm can demultiplex most common containers; the ones listed are those explicitly filtered from the selector and recognised when dragging.

Global video controls

• Resolution: maximum output height in pixels. It keeps the original aspect ratio. Original keeps the input resolution; preset options: 1080p, 720p, 480p.
• FPS: frames per second of the output video. Lowering the FPS noticeably reduces size. Original keeps the source frame rate; preset options: 60, 30, 24, 15.
• Audio: audio track handling. Four modes:
  • Encode audio 128k (default): re-encodes the track at 128 kbps (AAC in MP4 and MKV, Opus in WebM).
  • Encode audio 320k: the same but at 320 kbps, higher quality and size.
  • Remove audio: discards the track. Reduces size and avoids incompatibilities.
  • Keep audio: copies the original track without re-encoding (-c:a copy). It saves time and preserves quality, but it fails if the source codec is not compatible with the container (MP4 accepts AAC; MKV accepts AAC/Opus; WebM accepts Opus/Vorbis).
  For GIF and animated WebP the control is disabled: the audio is always discarded (GIF does not support audio). The settings in the Audio tab (Adjustments: high-pass, low-pass, gain and fades) are also applied to the audio of videos when it is re-encoded; with Keep or Remove there is no audio to adjust.

Per-format controls

Each format shows its own advanced controls panel when selected. The exhaustive detail of each parameter is in the Formats tab. Quick summary:

• MP4 (H.264): Quality (mapped to CRF 18–28), Encoder speed (ultrafast → slow), Profile (Baseline / Main / High) and Accelerate with WebCodecs. Encoding with x264 via FFmpeg or, with the checkbox ticked, with the browser (WebCodecs): several times faster, 15–20% larger. The checkbox only appears when the browser encodes H.264 at constant quality (Chrome and Edge); if a video can't go that way (deinterlacing, denoising, stabilization, inputs the browser can't decode) or it fails, x264 is used.
• MP4 (HEVC): Quality (16–26 quantizer). Always available: the browser encodes it in hardware when it can; otherwise x265 in FFmpeg (slow, with a warning). CFR output with the hvc1 tag, so Safari and QuickTime play it; Windows needs the system HEVC codec. A variable-frame-rate source loses its timing.
• WebM (VP9) and MP4 (AV1): Quality (24–63 quantizer). The browser encodes them with WebCodecs, and they only appear when it can. With deinterlacing, denoising, stabilization or an input the browser can't decode, FFmpeg first prepares an intermediate copy and the conversion takes longer.
• MKV (H.264): the same controls and the same H.264 video as MP4, with AAC audio, but in a Matroska container. Useful as a master file or for desktop players; for the web, MP4 or WebM is better.
• WebM (VP8): Quality (mapped to CRF 15–35) and CPU used (encoder effort). Encoding with libvpx (VP8) via FFmpeg: the VP9 encoder of the current FFmpeg.wasm engine does not work.
• Animated GIF: Colours (palette max. 256) and Dithering (Bayer / Floyd-Steinberg / None). The audio is always discarded.
• Animated WebP: Quality, Method (encoder effort) and Lossless mode.
• Container only (no re-encoding): changes the container (MP4, MOV, MKV or WebM) by copying the streams, losslessly. If a codec does not fit the target, only that stream is re-encoded and you are warned. Resolution, FPS, filters and rotation are not applied; trimming is, from the previous keyframe.

Video presets

Optimised

Quality/size balance intended for general web use. It keeps the original resolution and FPS with medium-high quality. Recommended starting point for most videos.

Max Quality

Prioritises visual fidelity over file size. It keeps the original resolution and FPS, raises the encoder quality to the maximum (slower encoder, larger files) and keeps the original audio without re-encoding (-c:a copy). Suitable for masters, archiving or playback on a large screen.

Min Size

Maximises size reduction with aggressive compression. Reduces to 720p and 30 FPS to minimise size. Ideal for quick previewing, sending by messaging or contexts with limited bandwidth.

No Audio (Social)

Configuration optimised for social media: it removes the audio, reduces to 720p and 30 FPS, and applies balanced compression. Suitable for clips intended for Instagram, TikTok, X and the like, where the audio is replaced or plays muted.

Lossless (animated WebP only)

Available through the Lossless checkbox in animated WebP: lossless encoding that reproduces the pixels exactly. It generates much larger files than lossy mode. MP4, MKV, WebM (VP8) and GIF do not offer a dedicated lossless mode.

— Custom —

It activates automatically when any control is modified manually. Selecting a preset again restores its default values for the active format.

Audio conversion flow

• Select the Audio tab at the top of the right column.
• Choose the Preset, the output Format (MP3, AAC/M4A, Opus, WebM/Opus, OGG Vorbis, FLAC, WAV, AIFF, Original, which extracts the track as is in the container of its codec, or Container only, which copies the track into M4A, AAC or MKA without re-encoding it) and the global controls (Sample rate, Channels, Track for files with several audio tracks, Normalise volume).
• Adjust the format-specific advanced controls if you need to (the details of each one are in the Formats tab).
• Click Convert. You can cancel during the process.
• The conversion is done with FFmpeg.wasm: if it has not yet been loaded, it is downloaded the first time (~33 MB) and cached for later use.

Supported input formats

You can load audio files in the following formats: MP3 WAV AIFF AAC M4A OGG Opus FLAC. Extracting audio from video files loaded in the queue is also accepted by selecting the Audio tab before converting: only the audio track comes out, in the chosen format.

AIFF is lossless big-endian PCM, the native audio of macOS, DAWs and hardware samplers. It is written at 16 or 24 bits and, unlike WAV, no browser plays it: the preview of the result will not sound; the downloaded file is correct.

Global audio controls

• S. rate: output sample rate in Hz. Original keeps the input file's. For voice, 16,000–22,050 Hz is usually enough; for music, 44,100 or 48,000 Hz.
• Channels: Original keeps the file's ones, Stereo forces two channels and Mono reduces to one (roughly half the size in lossy formats).
• Normalise volume: matches the level of the files; it increases encoding time. Mode: Loudness (EBU R128) brings the perceived volume to −16 LUFS with loudnorm and limits the true peak to Peak; Peak applies a fixed gain so that the sample peak sits at Peak (dBFS), without touching the dynamics; True peak does the same with the inter-sample peak, the one that matters for MP3 and AAC. The peak modes analyse the audio first. Peak (dB) goes from −9 to 0 (−1.5 by default). The measured result appears in the converted file's Properties (Normalisation row).

Audio adjustments

The Adjustments cell opens a modal with filters that are applied to all the audios when converting, before normalisation, and also to the audio of videos when it is re-encoded (with “Keep” or “Remove” there is no audio to adjust):

• High-pass (40 Hz, 80 Hz for voice, 120 Hz, 200 Hz) — removes what is below that frequency: hum, microphone thumps and wind noise.
• Low-pass (16 kHz, 12 kHz, 8 kHz, 4 kHz telephone-style) — removes what is above: it softens hiss and strident highs.
• Gain — raises or lowers the volume in decibels. +6 dB is roughly double the amplitude; raising it a lot can clip. A double click returns it to 0.
• Fade in and Fade out — duration in seconds of the fade from silence at the beginning and towards silence at the end. The fade out is counted over the trimmed segment. 0 disables them.

Reset returns them all to their neutral value.

Trim

When you move the cursor over an audio file in the left column a playback bar appears with two handles to set the start and the end of the trim; the same in the viewer (eye icon). The conversion will generate only the selected fragment. The time fields accept the mm:ss.ms format; the reset button removes the trim and returns to the full clip.

Processing order

Trim → high-pass → low-pass → gain → fades → normalisation (loudnorm, or the gain measured with ebur128 in the peak modes) → output sample rate and channels → encoding in the chosen format. The audio output always discards the image: neither the video nor the embedded artwork goes into the file.

Audio max size

The Max size cell sets a cap per file (KB, MB or % of the original). From the duration (the trimmed one, if any) the bitrate that fits is calculated, never above the one the format already has; MP3 switches to CBR and Vorbis to average bitrate. If the file is still too big, it is redone with a lower one; if not even the codec minimum fits, a warning is shown. The output row shows the final bitrate. It does not apply to FLAC, WAV, Original or Container only.

Audio presets

Optimised

Quality/size balance intended for general use. Recommended starting point for most files. It keeps the current format and applies intermediate values (e.g. MP3 VBR V4, AAC 128 kbps, Opus 96 kbps, OGG q5, FLAC level 5, WAV 16-bit).

Max Quality

Prioritises fidelity over size. It keeps the current format and raises the parameters to the maximum reasonable (MP3 VBR V0 ≈ 245 kbps, AAC 256 kbps, Opus 192 kbps, OGG q8, FLAC compression 12, WAV 24-bit).

Min Size

Maximises size reduction with aggressive compression. If the current format is MP3, AAC or OGG, it keeps the format but lowers the bitrate; in any other lossy format it keeps the choice, and from lossless formats (FLAC/WAV) it automatically switches to Opus 48 kbps for its superior efficiency at low bitrates.

Lossless

Lossless encoding. If the current format is not already FLAC or WAV, it automatically switches to FLAC (compression 8). It reproduces the input samples losslessly, at the cost of files significantly larger than any lossy format.

Roland SP-404 MK2

Configuration for the Roland SP-404 MK2 sampler: it forces WAV 48 kHz, 16-bit PCM output and keeps the original channels. Hardware limit: 16 minutes per sample (not enforced in encoding; it remains as a warning).

Roland TR-8S

Configuration for the Roland TR-8S drum machine: it forces WAV 44.1 kHz, 16-bit PCM, mono output. Hardware limit: 600 ms per sample and 256 MB in total (not enforced in encoding; it remains as a warning).

Polyend Play

Configuration for the Polyend Play groovebox: it forces WAV 44.1 kHz, 16-bit PCM, mono output. Hardware limit: 32 MB per sample (not enforced in encoding; it remains as a warning).

— Custom —

It activates automatically when any control is modified manually. Selecting a preset again restores its default values for the active format.

Data transformation flow

• Select the Data tab at the top of the right-hand column.
• Choose the Operation you want (HTML↔MD, JSON↔CSV, JSON↔YAML, spreadsheet↔CSV/JSON, DOCX→HTML/MD/TXT, Image→Text (OCR), Beautify or Minify).
• Adjust the operation-specific controls if you need to, or choose a variant under Settings.
• Click Convert. The result is generated instantly in the browser.

Operation selector

The operations are grouped by intent: Text (HTML↔Markdown), Documents (DOCX, EPUB and image OCR), Structured (JSON, CSV, YAML and spreadsheets) and Code (Beautify and Minify).

By default the selector shows only the applicable operations: those that have a file in the queue they can read. The last option in the dropdown, Show all, shows the full catalogue (and Show only applicable returns to the filter). If the queue has nothing any operation can read, all of them are shown. The chosen operation decides which input files can be selected: the rest are dimmed.

Supported input formats

You can load text and data files in the following formats, which are accepted depending on the active operation:

• HTML / Markdown: .html .htm .md .markdown
• JSON / CSV: .json .csv .tsv
• YAML: .yaml .yml
• Spreadsheets: .xlsx .xls .ods (processed with SheetJS)
• DOCX: .docx (processed with Mammoth.js)
• EPUB: .epub (EPUB 2 and 3 without DRM, read with JSZip)
• Beautify / Minify: .html .htm .css .js .mjs .json
• Image → Text (OCR): any image in the queue.

Each operation produces one output file per input file, with the target extension (for example report.docx → report.html).

Available operations

HTML → Markdown

Converts HTML documents to Markdown using Turndown (v7.2.4) + GFM plugin. The available controls are:

• Headings: ATX (#) or Setext (===) style.
• List: unordered list marker (-, *, +).
• Code: fenced (```) or indented.
• Italic: emphasis delimiter (_ or *).
• GFM: enables GitHub Flavored Markdown (tables, strikethrough, task lists).

Markdown → HTML

Converts Markdown to HTML using Marked (v12.0.2). The available controls are:

• GFM: enables GitHub Flavored Markdown.
• Line breaks <br>: converts single line breaks into <br>.
• Sanitize: sanitises the HTML with DOMPurify to prevent script injection.
• Full doc.: wraps the output in a complete HTML document with basic embedded CSS.

JSON → CSV

Converts JSON to CSV using PapaParse (v5.7.0). The available controls are:

• Delimiter: separator character for the CSV (comma, semicolon, tab, pipe).
• Flatten deeply: flattens nested objects and arrays with dot-separated keys (e.g. dir.ciudad). Reversible when doing CSV → JSON.
• Header: includes a first row with the column names.
• Line ending: line break style (CRLF or LF).
• Always quote: quotes every cell value.

CSV → JSON

Converts CSV to JSON using PapaParse (v5.7.0). The available controls are:

• Delimiter: separator character of the input CSV (auto, comma, semicolon, tab, pipe).
• First row = header: uses the first row as field names in each JSON object.
• Automatic typing: automatically converts numeric and boolean values.
• Unflatten: rebuilds nested objects/arrays from dotted keys (the inverse of Flatten deeply).
• Formatted JSON: output with readable indentation.

JSON → YAML

Converts JSON to YAML using js-yaml (v4.1.0). The available controls are:

• Indent: spaces per indentation level (2 or 4).
• Wrap: maximum line width before wrapping (-1 = no limit).
• Sort keys: sorts the keys alphabetically.
• Quotes: quote style for strings (single or double).
• No anchors: disables YAML anchors/aliases (&ref / *ref).

YAML → JSON

Converts YAML to JSON using js-yaml (v4.1.0). The available controls are:

• Formatted JSON: output with readable indentation vs. compact on one line.
• Indent: spaces per indentation level.

Spreadsheet → CSV

Exports a sheet from an XLSX, XLS or ODS workbook to CSV using SheetJS (v0.20.3). The library (about 950 KB) is only downloaded when you choose a spreadsheet operation. The available controls are:

• Sheet: number (1 is the first) or sheet name; empty, the first. If a workbook in the batch does not have it, that file fails and the notice says which sheets it has.
• Delimiter: comma, semicolon (the one Excel expects in Spanish), tab or pipe.
• Formatted values: each cell as it appears in the sheet (dates, currency, percentages); disabled, the raw numbers.

Spreadsheet → JSON

Exports one sheet, or all of them, to JSON with SheetJS. The available controls are:

• Sheet: as in Spreadsheet → CSV.
• All sheets: an object { "sheet name": rows } with all the sheets in the workbook.
• First row = header: each row is an object with the names from the first row; disabled, an array of cells. Empty cells come out as null.
• Formatted values: each cell as text, just as it appears; disabled, numbers as numbers and dates in ISO 8601.
• Formatted JSON: output with readable indentation.

CSV/JSON → Spreadsheet

Creates an XLSX or ODS workbook per file with SheetJS; the sheet is named after the file. A JSON that is an array gives one sheet (nested objects are flattened with dotted keys, as in JSON → CSV); one shaped like { "sheet": [rows], … }, the output of All sheets, gives one sheet per key. The available controls are:

• Format: XLSX or ODS.
• Delimiter: separator of the input CSV (auto detects it).
• Detect numbers: saves CSV cells that look like numbers as numbers. Codes with leading zeros stay as text.

DOCX → HTML

Converts Word documents to HTML using Mammoth.js (v1.13.0). The available controls are:

• Include images: embeds the DOCX images as data: URIs.
• Ignore empty paragraphs: skips paragraphs with no content from the original document.
• Full document: wraps the output in a complete HTML document with basic CSS.

DOCX → Markdown

Converts Word documents to Markdown using Mammoth.js. The available controls are:

• Include images: includes the images as data: URIs in Markdown format.
• Ignore empty paragraphs: skips paragraphs with no content.

DOCX → Text

Extracts the plain text from a Word document using Mammoth.js. No additional controls: it returns the document's plain text.

EPUB → HTML, Markdown and Text

One file per book with the chapters in reading order, separated by a rule (--- in Markdown). Links to another chapter become anchors of the joined document, so footnotes keep working. An EPUB with DRM is not read: the row warns and the rest of the batch goes on. The available controls are:

• Include images (HTML and Markdown): embeds the book images as data: URIs; turned off, they are removed. Text never carries them.
• Full document (HTML): wraps the output with the book title, author and language and basic CSS.

Markdown / HTML → EPUB

Builds one EPUB 3 book per file, with a chapter and a contents entry for each level 1 heading (or level 2 too, with Chapters: By H1 and H2). Title and author come from the front matter (title, author) or the HTML <title>; otherwise, from the file name. The language is the interface one. Scripts, videos and audios are removed, and images are removed keeping their alt text (the .md arrives without its local images). Old HTML (<center>, <font>, align, width…) becomes CSS, forms are removed and task checkboxes become ☑ / ☐, so the book passes epubcheck. It carries both EPUB 3 and EPUB 2 contents, so it also works with Send to Kindle.

Image → Text (OCR)

Recognises the text of each selected image with Tesseract (tesseract.js 7.0.0) and returns one .txt per image, read with its viewer edits (crop, rotation). It is the only Data operation that reads images. The available controls are:

• Language: Spanish + English, Spanish only or English only.
• Confidence: removes words recognised with low confidence (50, 70 or 85 %) to clear up noise.
• Straighten: corrects a slight tilt before recognising.

The engine (up to about 9 MB, depending on the languages) is not downloaded until you convert for the first time; it stays stored, works offline and can be deleted from Settings. For a scanned PDF, use Searchable PDF (OCR) in the PDF tab.

Beautify

Formats HTML, CSS, JS and JSON code using js-beautify (v1.15.1). The available controls are:

• Indent: spaces per indentation level (2, 4 or 8).
• Wrap: maximum line length before wrapping (0 = no limit).
• Preserve line breaks: keeps existing blank lines.
• Final newline: adds a line break at the end of the file.

The operation automatically detects the type of code from the file extension and applies the corresponding rules.

Minify

Minifies HTML, CSS, JS and JSON code using Terser (v5.51.2) and CSSO (v5.0.5). The available controls are:

• Mangle JS: renames variables and functions to short names (JS only).
• Drop console: removes console.* calls (JS only).
• No comments: removes comments from the code.
• Minify embedded: in HTML, also minifies the CSS and JS inside <style> and <script>.

The output file keeps its original extension, adding .min before the extension (e.g. script.js → script.min.js).

Library loading

Each operation loads its JavaScript libraries on demand, served from the site itself (with a fallback to a CDN verified by SRI if the local file failed). The pilot in the right panel shows Loading libraries… during the download and Ready when they are available. The libraries are cached in the browser for later use.

Operation presets

Each operation offers one or more presets with predefined configurations. The preset varies by operation: for example, in HTML→MD you have Standard (GFM on, ATX), GFM strict and Compact; in JSON→YAML you have Standard, No line wrap (no wrap limit) and Sorted keys (keys sorted). The — Custom — preset switches on automatically when you change any control by hand.

PDF workflow

• Select the PDF tab at the top of the right-hand column (or click the PDF button in the output bar, which opens it with Images → PDF).
• Choose the Operation. Unlike Image, Video or Audio, here you do not choose an output format but what you want to do; the operation decides which controls are shown and which input files can be selected.
• Choose a variant in the second selector, whose name changes depending on the operation (Type, Result, Position, Level, Language, Format, Mode…), and adjust the controls that appear. Below, a note explains what will happen with the chosen operation.
• If needed, edit the pages of each PDF in the viewer (the eye icon on its row): rotate, delete, duplicate, reorder, insert blank pages and crop margins. All operations start from the already edited document.
• Click Convert. You can cancel during the process.

The engine is pdf-lib (creating and modifying documents) together with pdf.js (drawing pages and reading their text), both served from the site itself and loaded the first time they are used. Operations that do not need to draw run in a worker, in the background, so the interface does not freeze on large documents.

Operation selector

The operations are grouped into Create PDF (Images → PDF, Text → PDF and Text and images → PDF) and Operate on PDF (the rest). As in Data, the selector shows only the operations applicable to what is in the queue by default; Show all, at the end of the dropdown, shows the full catalogue, and Show only applicable returns to the filter.

The Pages field

Several operations act on a range of pages. The field accepts:

• Single numbers — 3.
• Ranges — 2-6 (also with an en dash).
• Open ranges — 8- (from 8 to the end) and -3 (from the start to 3).
• Several parts separated by commas or semicolons — 1-5, 12, 20-.
• all (or everything), or the field empty — the whole document.

Pages are counted from 1 and over the already edited document (if you deleted or moved pages in the editor, the numbering is the new one). The order you type is the order of the result: 5, 1, 3 puts those pages out in that order. A repeated page is used only the first time. A backwards range (6-2), a page that does not exist or text that cannot be understood give an error shown under the controls, checked against the first selected PDF; with a batch, a range that goes out of bounds in one document does not stop the rest.

Create PDF

Images → PDF

Assembles a PDF with the already converted images from the right-hand column, in the order of the list (drag the rows to change it). It is the only operation that does not read the input selection: it works on the output, so the images arrive with their edits, adjustments and metadata already applied. In simple mode there is no need to convert them first: the images in the list are assembled as they were loaded (or their output, if already converted); only those with unapplied edits go through Image first. Controls:

• Page — Fit to image gives each page the exact size of its image; A4, Letter and Custom fit the image inside, centred and without distorting it. With Custom the Dimensions appear in centimetres (short side and long side, in any order: the orientation decides which one goes horizontally).
• Orientation — Automatic makes each page portrait or landscape depending on whether its image is taller or wider; Portrait and Landscape fix it. Not used with Fit to image.
• Margin (cm) — blank margin around the image; 0 leaves it edge to edge. If the margin leaves no room on the page, a warning is shown and nothing is generated.
• Files — A single PDF with one page per image (images.pdf), or One PDF per image, each with the name of its image.
• Quality — JPEG quality with which the images are re-encoded before embedding them: it is what decides how much the PDF weighs. High (92 %), Medium (80 %) or Low (60 %). Images with transparency go to PNG, losslessly. Original (no re-encoding) embeds the JPEGs and PNGs as they are: the PDF weighs what they weigh.
• Resolution — pixels per inch at most (96, 150, 200 or 300 dpi), according to the size at which each image is drawn on its page. A 12 Mpx photo on an A4 runs at about 360 dpi: lowering it to 150 leaves it the same on screen and removes most of the weight; 300 dpi is the useful ceiling for printing. It never upscales, and does not affect Fit to image. Original does not reduce.
• PDF/A-2b — see below.

The note under the controls says how many images are ready and estimates the weight of the PDF. Variants of Type: Fit to image (page the size of each image), A4 and Letter (with a 1 cm margin), A4 full bleed (no margin) and 10×15 cm photo.

If the queue has unconverted images, the note offers a guided flow: Next takes you to the Image tab with all the images selected so you can adjust the conversion; another Next converts them and returns to PDF; Convert assembles the PDF. The link Use current settings and assemble now does the three steps in a row with the settings Image already has.

Text → PDF

One PDF per selected text file (.txt .md .markdown .html .htm), with selectable and searchable text, written with the Exo 2 font embedded (only with the characters used). Markdown and HTML become plain text: no bold, drawn tables or images (for that there is Text and images → PDF). Controls: Page, Orientation and Margin (as above; with Fit to image, A4 is used), Font (8 to 18 pt) and PDF/A-2b. Variants: A4, Letter and A4, large type.

Text and images → PDF

Like Text → PDF, but laid out: it keeps headings, bold, italics, lists, quotes, code blocks and images, with selectable text. One PDF per file. The images are taken from the file itself if embedded (data:) or from the queue by name (if the HTML or Markdown references photo.jpg and there is a photo.jpg loaded, it is used); those from the internet are not downloaded. Same controls and variants as Text → PDF.

It also accepts DOCX, but not directly: a DOCX is first turned into HTML in the Data tab (with its images embedded) and what comes out is laid out. If you select DOCX without converting, the note offers a guided flow: Next takes you to Data with DOCX → HTML chosen and the DOCX files selected; another Next converts them and returns to PDF; Convert lays out one PDF per document. Use current settings and lay out now does it all in a row. The converted text is stored separately, so you can lay it out again with other settings without repeating Data.

PDF/A-2b

The PDF/A-2b checkbox saves the PDF in the long-term archival format required by administrations and archives: it adds XMP metadata, the sRGB colour profile and an identifier. It is only available in the PDFs the app creates (the three Create PDF operations), not when modifying an existing PDF.

Operate on PDF

Merge

Merges the selected PDFs into one (merged.pdf), in the order of the input list: drag the rows to change it. It needs at least two. Each document goes in with its edits. The note estimates the weight of the result.

If in addition to PDFs there are unconverted images in the queue, the note offers the guided flow Merge with images (scans or photos that are joined with documents): 1/3 takes you to Image and converts the images; 2/3 assembles them into a PDF with Images → PDF; 3/3 returns to Merge with that PDF at the end of the others, and you can reorder the list before clicking Convert. The PDFs you had selected at the start are merged or, if there were none, all the loaded ones.

Merge pages

Like Merge, but choosing which pages go in from each document. In Pages, separate with | one range per document, in the order of the list: 1-3 | 5 | all takes pages 1 to 3 from the first, page 5 from the second and the whole third one. A range without | applies to all.

Insert a PDF into another

It needs exactly two selected PDFs: the first in the list is the base document and the second, the one inserted (drag the rows to swap them). Pages chooses which pages of the second go in. The Position variant decides where: At the end, At the beginning or After a page; with the latter After page appears, the page of the base document after which they are inserted (0 puts them before everything). The result is named after the two documents joined by + (contract+annex.pdf).

Split / extract

Puts out pages from each selected PDF. The Result variant decides how:

• One PDF with those pages — a single document with the pages from the Pages field, in the order written (name_p2-6.pdf).
• One PDF per range — one document per comma-separated part: 1-3, 4-8, 9- gives three files.
• One page per file — splits the whole document, one page per PDF (name_p01.pdf, name_p02.pdf…); the Pages field is not used.

With the field empty and One PDF with those pages you get the document as it was left in the editor: it is the way to download a PDF to which you have only rotated, deleted or reordered pages.

PDF → images

One image per page in the Pages field (empty: all), from the already edited document, drawn with pdf.js. They enter the queue as converted images (name_p01.jpg…), with the Generated label. Controls: Format (JPG is light and works well with photos and scans; PNG is lossless, better for text and crisp drawings, but heavier; WebP sits between the two) and Resolution (72–96 dpi for screen, 150 for comfortable reading, 300 for printing; an A4 at 150 dpi is 1240 × 1754 pixels). Variants: JPG, 150 dpi, PNG, 150 dpi and JPG, 300 dpi (print). If a page is enormous, it is drawn at a lower resolution than requested and a warning is shown.

PDF → Text

One .txt per PDF with the text of the pages in the Pages field. It only reads selectable text: a scanned PDF has none, and then a warning is shown and nothing is generated (run it through Searchable PDF (OCR) first). With several columns or tables, the order may not be the reading order. Variants: Continuous paragraphs (joins the lines of each paragraph) and Lines as on the page. The Mark pages checkbox puts === Page N === before the text of each page, with its number in the PDF.

PDF → DOCX

Approximate conversion: one .docx per PDF with the text of the pages in the Pages field, rebuilt into paragraphs and headings (by font size), with bold and italics. Images, tables and layout are not carried over. A scanned PDF has no text: run it through OCR first. Variants: Continuous text and Keep page breaks (a Word page break where each PDF page ended).

Compress

Recompresses the photos and scans inside the PDF to JPEG; the text and vector drawings stay the same. Variants of Level: Light (85 %, 200 dpi), Balanced (75 %, 150 dpi) and Strong (60 %, 110 dpi). An image is only reduced in resolution if it clearly exceeds the target, and is only replaced if the result weighs less; very small ones (icons, bullets) and whatever cannot be handled are left intact. The final notice says how much it went down and how many images were recompressed. If the document does not weigh less, no file is generated. Output: name_compressed.pdf.

Searchable PDF (OCR)

Each scanned page in the Pages field is drawn, goes through OCR (Tesseract) and gets an invisible text layer on top: the PDF looks and weighs almost the same, but it can be searched, selected and copied. Variants of Language: Spanish + English, Spanish or English. Controls: Reading (300 dpi recognises small print better; 200 dpi is faster and enough for normal text; it does not change the appearance or the weight of the PDF) and Scanned only (leaves pages that already have selectable text as they are, to avoid duplicating the layer). The engine (up to about 9 MB, depending on the languages) is downloaded the first time and stays stored; each page takes a few seconds. Output: name_ocr.pdf.

Equalise size

Each page is fitted, without distorting it and centred, onto a sheet of the chosen size: useful for documents with pages of mixed sizes. Page, Orientation and Margin controls as in Images → PDF. Variants: A4, Letter and A4 with margin. Links and form fields are lost.

Number pages

Writes the page number on each page in the Pages field (empty: all), with the numbering of the edited document and upright even if the page is rotated. Variants of Format: 1, Page 1 of 5 and 1 / 5. Controls: Position (bottom or top; centre, right or left, as the page is seen) and Font (8 to 18 pt, in Exo 2). Output: name_numbered.pdf.

Metadata

Title, author, subject and keywords of each selected PDF. The ones from the first appear in grey. Variants of Mode: Change what is written (only the fields you fill in are touched) and Replace all (empty clears) (empty fields are cleared). The document XMP metadata is removed, so it does not contradict the new values.

Encrypt with password

Saves the PDF encrypted with AES-256. Open password is needed to open it; owner password is the one that allows removing the restrictions. Permissions choose what can be done without the owner password: print, copy text and images, and modify. Removing a permission does not make the content unreadable: it prevents normal use, not malicious use. The same block appears in Images → PDF and Text → PDF, where it is optional: with both fields empty the PDF is created unencrypted. Output: name_encrypted.pdf. The engine (2.2 MB) is downloaded the first time something is encrypted or a protected PDF is opened.

A password-protected PDF asks for its password when it is loaded and is decrypted in memory: the password is not stored and the original file is not touched. If the dialog is cancelled, the file stays in the queue marked as password-protected and out of the operations.

Optimise for web (linearize)

Reorders the PDF so the first page can be painted with the first bytes (fast web view): what a viewer using HTTP range requests takes advantage of. The content and the pages do not change. It is done by qpdf, the same engine as encryption.

It can increase the weight: qpdf rewrites the streams and in text or vector PDFs they may end up worse compressed (measured: +86 % on a 20-page text PDF), while image PDFs grow only ~0.1 % (the hint tables). The notice reports the before and after, and warns when it grows more than 10 %. Output: name_web.pdf.

Page editor

The eye icon of a PDF in the input opens a grid with its pages in thumbnail (they are drawn as they come on screen, so a document of hundreds of pages opens quickly). From the top bar:

• Mark — a click marks only that page; Ctrl+click or Space adds or removes it; Shift+click marks the stretch from the last pressed; the arrows move the focus. Ctrl+A marks or unmarks all.
• Rotate the marked ones to the left or right (Shift+R / R).
• Move the marked ones one place backwards or forwards (Alt+← / Alt+→), or drag them to their new place.
• Insert blank page after the last marked one (or at the end), the size of the neighbouring page.
• Duplicate and Delete (Del) the marked ones.
• Crop margins of the marked ones (or all of them): opens a bar with Top, Right, Bottom and Left in centimetres, measured over the page as it is seen. Detect suggests the margins by finding the content box. Apply or Enter saves them; Esc cancels; the remove button deletes the crop of the affected pages.
• Undo and Redo (Ctrl+Z / Ctrl+Shift+Z), up to 200 steps, and Reset to return to the original document.

The editing is non-destructive: the file is not rewritten; what pages remain, in what order, with what rotation and what crop is stored, and each operation starts from the resulting document. Close and save keeps the changes; Esc or Close and discard throws them away. The row indicates how many pages the edited document has. To download the edited PDF as it is, use Split / extract with the Pages field empty.

Problematic PDFs and large documents

• Password-protected — it cannot be opened (pdf-lib does not decrypt). Remove the password with another program and load it again.
• Damaged — its internal structure cannot be read; it usually happens with incomplete downloads. Opening it in another viewer and saving it again sometimes fixes it.
• Not a PDF — it does not start with the %PDF header: probably another kind of file with the extension changed.
• No pages — there is nothing to work on.

These PDFs are marked in their row with the reason and are not selectable. Before working with very large documents (from about 150 MB, or 400 MB on machines with 8 GB of memory or more; with Merge, Merge pages and Insert the sum of all of them counts), confirmation is asked: the browser has to load them entirely into memory and the tab could close. If memory runs out halfway, the notice says so; close other tabs or split the document.

With a batch, a PDF that fails is warned about and does not stop the others.

Recipes

A recipe is a named sequence of up to 12 steps, plus a finishing action. Each step is a tab (Image, Video, Audio, Data, PDF or Rename) with its settings. When you apply it, the app does exactly what you would do by hand: it puts the step's settings in its tab, goes to it and clicks Convert (or Rename), step after step. You see it go through the tabs and each file shows its usual progress.

What a step stores

When you click Add step (or Update step), the recipe copies all the controls of that tab as they are at that moment: format or operation, preset or variant, the format's controls and those of its modals (Scale, Settings, Background, Max size, Upscale, Filters, Audio adjustments…). In Image it also stores which field drives the scale (percentage, width or height). What it does not store is what belongs to each file: crops, rotations, straightenings, erased areas, time trims, PDF page edits or hand-edited metadata. Those stay with each file and the recipe respects them when converting.

When a step is applied, whatever it does not carry (the example recipes only carry what changes) takes the factory value, not the one that was on screen: a recipe gives the same result wherever you use it.

Composing and editing

• New recipe opens the editor: type a name (up to 80 characters; it cannot repeat that of another recipe), choose the tab in Add step and click Add. Set up that tab as usual beforehand.
• Each step shows a readable summary of its settings (for example WEBP · quality 80 · 1920 px wide (max.), or the name of the Data or PDF operation and its variant).
• The pencil of a step closes the window, loads its settings in its tab and shows a strip: change whatever you want and click Update step (or Cancel). In both cases, the tab then returns to the settings it had before editing.
• The arrows or the ⋮⋮ handle change the order; the bin removes the step.
• When finished: Do nothing, Download ZIP (with a single output file that file is downloaded, without a ZIP) or Download one by one.
• Save saves it; Back returns to the list (if there are unsaved changes, it asks before discarding them).

From the list, each recipe has Apply (disabled if it has no steps, if a recipe is already running or if there are no selected files; the tooltip says why), Duplicate, Edit (also by clicking its name) and Delete. Its steps are summarised under the name.

Warnings when composing

The editor marks the steps that will probably not do what you expect (they do not prevent saving or applying):

• Two Image, Video or Audio steps — the second converts the same files again from the original and replaces the result of the first: they do not chain. To combine settings, put them in a single step.
• Two Rename steps — the second renames the same files again.
• Images → PDF without an Image step before it — that operation assembles already converted images, so it needs an Image step before it.
• Recipe without steps.

In Data and PDF there can be several steps, because they usually touch different files (different operations).

What files each step receives

• The recipe is applied to the files selected in the input when you click Apply. With no selection, it warns and does not start.
• Each step works on those files plus what the previous steps of the same recipe have generated (the PDF from Images → PDF, the pages from PDF → images…). So “Scans to PDF” converts the photos in the Image step and the PDF step assembles exactly those.
• Within that selection, each tab only processes what it knows how to process: an Image step ignores audio, an Audio step ignores images. That is why a recipe can handle a mixed queue in one go (“Everything for web”).
• If the recipe has a Video step, its Audio step does not touch the videos: it would extract their audio and replace the video conversion.
• In an Images → PDF step, only the images converted by the recipe itself are assembled, not those that were already converted before.
• A step with nothing to do (no file of its type, no applicable operation) is skipped and it is said at the end.

During and after

• A strip indicates the step in progress (Recipe “…” · Step 2 of 3 · Image) and offers Cancel, which asks for confirmation: what has already been converted stays in the list.
• While it lasts, the pending renaming of the Rename tab is left blank (Convert would apply it): names only change in the Rename steps. The automatic download in Settings does not act either.
• If a guided flow was in progress, it is abandoned.
• When it finishes, all the tabs return to the settings they had, the selection returns to the previous one and you return to the tab you were on.
• A notice summarises the result: output files, errors and steps with nothing to do.
• The finishing action (ZIP or downloads) only includes what the recipe has produced: what it generated and what it converted or renamed. A previous output of a file that no step touched does not go in.

Try with 1 file

From the editor, apply the recipe as it is (even if it is not saved) only to the first selected file and without a finishing action. The window closes so you can see the result; the recipe being edited is parked and you come back to it when you open My recipes.

Apply from the context menu

Right-click in the input list: under Apply recipe your saved recipes appear (up to six; with more, More recipes… opens the window). They are applied to the selection like the Apply button. They are disabled while a conversion or a recipe is running.

Example recipes

Grouped into Destinations, Image, PDF, Audio and video, Data and Mixed types. The destinations fit format, size, quality and weight to what a platform asks for; the maximum weight is per file, not for the batch:

• WordPress — WEBP quality 80, 2048 px wide and 200 KB at most, lowercase names, ZIP.
• Shopify (product) — JPG quality 85, 2048 px wide at most (Shopify then generates the WEBP and thumbnails), ZIP.
• Instagram (feed) — JPG quality 90, 1080 px wide, ZIP.
• Link sharing image (Open Graph) — JPG quality 82 at 1200 × 630 px cropped to the centre and 300 KB at most, ZIP.
• Photos for email — JPG quality 82, 1280 px wide at most and 500 KB per photo (about 50 fit in 25 MB), ZIP.
• Podcast — MP3 CBR 128 kbps at 44.1 kHz, loudness at −16 LUFS and the peak at −1 dB, ZIP.
• Lightweight web — WEBP quality 80, 1920 px wide at most, lowercase names, ZIP.
• Scans to PDF — JPG quality 85 and assembled into a single A4 PDF.
• Merge PDF — the selected PDFs into one, in the order of the list.
• PDF to images — each page as a JPG at 150 dpi, ZIP.
• Audio to MP3 — MP3 VBR V0 with normalised volume, ZIP.
• Extract audio from videos — the audio track of the videos as MP3, ZIP.
• Text from screenshots (OCR) — one .txt per image (Spanish and English), ZIP.
• Everything for web — images to WEBP at 1920 px, videos to WebM 720p and audio to Opus 96 kbps, ZIP.

They are applied as they are or copied to your recipes (with a free name, e.g. “Lightweight web (2)”) to change them.

Importing and exporting recipes

Export downloads your recipes as a JSON; Import adds them to the ones you have. Those that already exist (same identifier) are not duplicated, and invalid entries are discarded and counted in the notice. A file from a newer version of the app is rejected. Recipes also travel in the export of the complete configuration.

Guided flows

Some tasks need to go through two tabs, for example converting images in Image before assembling them into a PDF. Instead of sending you to a tab and having you come back, a guided flow takes you: the Convert button is called Next in the intermediate steps, shows an n/N indicator and its tooltip says what it will do. Everything can be adjusted at each step before continuing. Cancelling a conversion leaves you at the same step; choosing another operation abandons the flow. The state lives only in memory: on reload there is no flow.

Available flows (detailed in the PDF tab): Images → PDF with unconverted images, Merge with images and DOCX → PDF. Each offers a shortcut in its note to do it all in a row with the current settings.

With Step-by-step guided flows disabled in Settings, when a flow starts all its steps run in sequence. The After converting → Download automatically preference only acts on the last step of the flow.

My presets

My presets (the header icon) saves under a name a complete snapshot of the settings of all the tabs at once (Image, Video, Audio, Data, PDF and Rename). When you apply it, the six tabs recover those values in one go. Differences from recipes: a preset does not run anything, it only changes the controls; a recipe chains steps and converts.

• Save — type a name (it cannot be repeated) and save the current settings.
• Apply — click the name.
• Delete — asks for confirmation.
• Export / Import — as JSON; on import, those that already exist are not duplicated.

Do not confuse it with the tab presets (Optimised, Max Quality…), which only change the controls of their tab.

Image viewer and editor

The eye icon of an image in the input opens the full-screen viewer, which is also the editor. The edits are stored in the file (they do not rewrite it) and are applied when converting, before scaling.

• Rotate 90° to the left or right (Shift+R / R) and flip horizontally or vertically (F / Shift+F).
• Crop (C) — drag the frame or its corners. Aspect ratio: Free, Original, 1:1, 4:3, 3:2, 16:9 or 4:5; the rotate button sets the frame to landscape or portrait. You can also type the width and height in pixels. Remove crop takes the frame to the whole image.
• Straighten (S) — rotates by a free angle from −45° to +45°, in tenths (arrows ← →; with Shift, degree by degree). The corners left exposed are filled with a colour, in Auto (the colour of the image's edge) or Transparent (in JPG they come out white). No corners crops to the largest rectangle with the original's proportions that does not touch the corners, and replaces any previous crop.
• Erase areas (E) — drag over the image to cover an area (a number plate, a piece of personal data, a stain) with a colour, or in Auto with the colour around it. Click an area to select it and remove it with Del; another button removes them all.
• In crop, straighten and areas, Apply or Enter saves and Esc cancels.
• Undo and Redo (Ctrl+Z / Ctrl+Shift+Z) and Reset (undoes crop, rotation, straightening, flipping and areas).
• 1:1 toggles actual size (1) and fitted (0); also with a double click. The wheel zooms and you drag to move.
• Previous / next file (Page Up / Page Down) — discards what is not saved.
• Close and save, or Close and discard (Esc).

In video and audio, the viewer is the player with the trim bar (see the Video and Audio tabs); in PDF, the page editor (see the PDF tab).

Before/after comparator

In the viewer of a converted image, Compare with the original (C) overlays the two versions with a divider you drag (or move with ← →). On the left, the original with the same edits and the same resizing but unencoded; on the right, the converted one. So what you see is only the effect of the compression. Zooming and panning affects both layers at once; Fit and 1:1 change the zoom. The bar indicates the change in weight.

Compute SSIM measures the structural similarity between the two (on reduced copies, from 0 to 1) and rates it: indistinguishable (≥ 0.995), good (≥ 0.98), acceptable (≥ 0.95) or visible loss. It is useful for lowering the quality to just before it shows.

Settings

The gear icon in the header opens the global preferences. Only what you change from the default value is saved.

Remember between visits

• Remember the session (loaded files), Save the conversion history and Remember the tab settings. When you switch one off, you are offered the option to delete what has already been saved of that type.

Session

• When opening with a saved session — Ask (a banner at the bottom offers Restore or Discard for a few seconds; if it is hidden without choosing, the session is kept for the next visit), Restore it or Discard it without asking.
• Space to save files (50 MB, 200 MB, 500 MB or 1 GB) — files that do not fit are remembered by name only, without the content, and appear as placeholders when restoring. Lowering the limit does not delete what is already saved.

The session is saved in IndexedDB, on your machine, with the order of the list; generated files come back as generated. As soon as you load new files, the previous session stops counting.

History

• Maximum entries (100, 500 or 2000) — once the maximum is reached, the oldest ones are discarded.

The Conversion history modal (the clock icon in the header) lists the original file, operations, date, output files and downloads. Download exports it and Clear deletes it. With the history disabled, this session's conversions are shown, but they are lost when the page is closed.

Output

• ZIP name — accepts %date%, %time% and %n%; empty, transformer_YYMMDD.
• After converting — Do nothing or Download automatically (in a guided flow, only on the last step; in a recipe, never: its finishing action rules).
• Pause between downloads (no pause, 250 ms, 500 ms or 1 s) — so the browser does not block multiple downloads.
• Suggest ZIP from (5, 10, 20 or 50 files).
• Output metadata — Clean (recommended), Copy from the original without GPS or Copy all. It only applies to JPEG, PNG and WebP when converting; manual per-file editing always rules.

Interaction

• Interface mode — Automatic (Simple on narrow screens and when arriving from a task page; Advanced everywhere else), Advanced (two queues) or Simple (one list). See Simple mode in the General tab.
• Preview on hover and Panel speed (fast, normal, slow). It only affects the mouse: the keyboard and clicking still open the panel.
• When loading files — Select them, Do not select them or Select only the new ones (deselects the previous ones).
• Ask for confirmation when deleting.
• Step-by-step guided flows — disabled, it runs all the steps in sequence with the current settings.

Performance

• Warn if more than — memory threshold for the warning before converting large video, audio or PDF: automatic based on the device, 1, 2 or 4 GB, or do not warn.
• Files at a time when loading — automatic, 2, 4 or 8.
• Estimate the output size — disabled, the reference row does not show the estimated size.

Data in this browser

It shows how much each saved thing takes up and lets you delete it separately: saved session, history, user presets, recipes, tab settings, preferences, upscale models (AI), background removal models (AI) and the OCR engine and languages. The models and the engine will be downloaded again the next time they are used.

Reset everything deletes all of the above at once, removes the files from the list (your originals are not touched) and reloads the page.

Exporting and importing configuration

Export configuration downloads a JSON (transformer_config_YYYYMMDD.json) with the tab settings, the user presets, the recipes and the preferences; the history and the session are left out on purpose. It is useful for copying your configuration to another browser or machine. Import configuration reads that file and lets you choose which blocks to apply (checkboxes Tab settings, User presets, Recipes and Preferences). Note: unlike importing from My recipes or My presets, which add, here the imported presets and recipes replace the ones you have.

Image formats

WEBP — encoder: libwebp

• Quality (0–100) — in lossy mode it controls visual fidelity; in lossless mode it controls the compression effort (0=fast/larger, 100=slow/smaller).
• Method (0–6) — encoder effort. 6 produces the smallest file at the cost of more time.
• Near-LL (0–100) — near-lossless preprocessing before lossless encoding. 100=faithful to the original; low values reduce artifacts with minimal perceptual loss.
• Alpha (0–100) — quality of the transparency channel, compressed independently of colour. 100=perfect alpha.
• Filter (0–100) — strength of the spatial anti-blocking filter. High values smooth the image; low values preserve detail.
• Lossless — enables lossless encoding. The Quality label becomes Effort.

AVIF — encoder: libavif

• Quality (1–100) — passed as-is to libavif. 60–75 is a good range for the web; with 100, alpha 100 and 4:4:4 it is lossless.
• Speed (0–10) — 0=best quality/slowest, 10=fastest/lowest quality. 4–6 offers a good balance.
• Denoise (0–50) — pre-encoding noise reduction. Useful with grainy images; improves compression at the cost of some detail.
• Sharpness (0–7) — post-processing edge sharpness. High values can create halos on high-contrast edges.
• Subsample — chroma sampling: 4:4:4=maximum colour fidelity, 4:2:0=web standard (lighter files).
• Alpha (1–100) — quality of the transparency channel; 100 = lossless alpha.

JPG — encoder: MozJPEG

• Quality (1–100) — 80–85 offers a good quality/size ratio. Above 95 the size grows with minimal gain.
• Chroma subsampling — chroma sampling: 4:4:4=maximum fidelity, 4:2:0=web standard, 4:1:0=more aggressive (a quarter of chroma on each axis).
• Smoothing (0–100) — Gaussian smoothing before encoding. Reduces artifacts in gradients; recommended only at low values (5–15).
• Quantisation table — quantisation table that distributes the loss across frequencies: 0=JPEG standard, 3=Photoshop style, 8=IJG optimised for detail.
• Trellis — optimisation of the DCT coefficients after quantisation. Off=disabled; Basic=multipass (several passes over all the coefficients); Full=multipass + opt_zero + opt_table (adapts tables to the image). Higher compression, more time.
• Grayscale — encodes the JPEG in a single luminance channel (color_space=JCS_GRAYSCALE). Discards the colour information, producing smaller files.

PNG — encoder: OxiPNG (+ libimagequant)

• Optimisation (1–6) — lossless encoder effort level: how many filter and compression combinations it tries. It only affects the file size, never the visual quality. The gain runs out quickly: between level 3 and 6 the difference usually stays at 2–3% of size in exchange for quadrupling the processing time.
• Colours — reduces the image to an indexed palette with the given number of colours, with libimagequant (the engine of pngquant). It is the only lossy setting PNG has. In flat graphics, screenshots, logos and illustrations it saves between 40% and 70% with a visual difference that is hard to notice; in photographs with smooth gradients it can produce banding. No reduction keeps the full colour and the conversion is 100% lossless.
• Interlaced — Adam7 interlacing: the image loads in successive passes, showing a low-resolution version before completing. Useful on slow connections. Unlike progressive JPEG, it increases the file size, typically by between 20% and 40%.

ICO / Favicon — encoder: OxiPNG per size

• Sizes — a single .ico with 16, 32, 48, 64, 128 and 256 px; each size is encoded as PNG with the PNG settings.
• Favicon pack (.zip) — instead of the loose .ico, a ZIP with favicon.ico, apple-touch-icon.png, icon-192.png, icon-512.png and site.webmanifest.

TIFF — own encoder (lossless)

• Compression — LZW (the most compatible with printing and layout programs), Deflate (ZIP) (somewhat lighter; Photoshop, GIMP, Affinity, libtiff) or Uncompressed (3 bytes per pixel, 4 with transparency).
• Resolution (72, 150, 300 or 600 dpi) — a value stored in the file for the printed size; it does not change the pixels.
• Input: loaded TIFFs are read with UTIF.js; from a multi-page TIFF the first one is used.

JPEG XL — encoder: libjxl

• Quality (1–100) — on the cjxl scale: 90 is visually lossless; 75–85 is a good range for the web. Ignored with Lossless.
• Effort (1–9) — more effort, slightly smaller files and more time. 7 is libjxl's default; 9 can take several times longer.
• Lossless — reproduces the pixels exactly; usually lighter than PNG.
• Progressive — lets a low-resolution version show before the download completes; adds some weight.
• Presets: Optimised 80, Max Quality 92 (effort 8), Min Size 60 and Lossless.

Video formats

Video formats are encoded with FFmpeg.wasm, which is downloaded the first time a video is processed and stays in the browser cache for later use. WebM (VP9), MP4 (AV1) and, with Accelerate with WebCodecs, MP4 and MKV (H.264) are encoded by the browser (WebCodecs); MP4 (HEVC) uses the hardware encoder when available and x265 in FFmpeg otherwise. FFmpeg still reads the input and packs the output.

MP4 (H.264) — encoder: x264

• Quality (0–100) — mapped to CRF (Constant Rate Factor): 100=CRF 18 (high quality, large file), 0=CRF 28 (more compressed). 70–80 offers a good balance for the web.
• Speed — x264 encoder preset (ultrafast, superfast, fast, medium, slow). Slower presets produce smaller files at the same visual quality; faster ones are heavier but finish sooner. medium offers a good balance.
• Profile — H.264 profile. High=better quality/compression, compatible with most modern devices. Main=middle ground. Baseline=maximum compatibility with old hardware, at the cost of a larger size.
• Audio: re-encoded to AAC 128k by default. The Audio dropdown lets you choose 128k, 320k, remove the track or keep it without re-encoding (-c:a copy; requires an AAC source).

MP4 (HEVC) — encoder: WebCodecs or x265

• Availability — always in the selector. It uses the browser's hardware HEVC encoder (WebCodecs) when available; otherwise x265 in FFmpeg.wasm, with a warning that it takes much longer (at 1080p, several times the clip duration).
• Quality (0–100) — 16–26 quantizer in WebCodecs (100=16, 0=26), calibrated with SSIM against x264 CRF 18–28. The x265 fallback uses CRF 23–33.
• Audio: re-encoded to AAC 128k by default, as in MP4 (H.264).
• Compatibility — the MP4 carries the hvc1 tag, the one Safari and QuickTime play. On Windows the system HEVC codec is required; without it the file is valid but will not play.
• Frame rate — the output is CFR: the HEVC stream FFmpeg reads carries no timestamps, so the chosen frame rate (or the source's) is applied. A VFR source loses its variable timing.
• Presets — Optimised 55, Max Quality 90, Min Size 15 (720p/30) and No Audio 35 (720p/30).

MKV (H.264) — encoder: x264

• Controls — exactly those of MP4: Quality (CRF 18–28), Speed and Profile.
• Container — Matroska (.mkv): it accepts almost any codec and multiple tracks, and is common as a master file and in desktop players (VLC, mpv, Plex). Its playback in browsers and mobile devices is not reliable: for the web, MP4 or WebM.
• Audio: re-encoded to AAC 128k by default. With Keep (-c:a copy) it accepts source AAC and Opus.

WebM (VP9) — encoder: WebCodecs

• Availability — only when the browser encodes VP9 with WebCodecs (Chromium-based browsers, in software).
• Quality (0–100) — mapped to a fixed quantizer: 100=24 (high quality, large file), 0=63. It compresses better than VP8; browser encoders are built for real time, are somewhat larger than x264 at the same quality, and their lowest quality is still high. No lossless mode.
• Audio: re-encoded to Opus 128k by default, as in WebM (VP8).

MP4 (AV1) — encoder: WebCodecs

• Availability — only when the browser encodes AV1 with WebCodecs (Chromium-based browsers, in software).
• Quality (0–100) — mapped to a fixed quantizer: 100=24, 0=63. It is the slowest of the WebCodecs paths, still faster than x264 in FFmpeg.wasm.
• Audio: re-encoded to AAC 128k by default, as in MP4 (H.264).

Animated AVIF — encoder: WebCodecs

• Availability — the same as MP4 (AV1): the AV1 sequence is packaged as an animated AVIF image, without re-encoding.
• Quality (0–100) — the same quantizer as AV1. Far lighter than GIF, and usually lighter than animated WebP.
• No audio, like GIF and animated WebP.

WebM (VP8) — encoder: libvpx

• Quality (0–100) — mapped to CRF: 100=CRF 15 (high quality), 0=CRF 35 (more compressed), with a generous bitrate ceiling (25 Mbit/s). VP8 is somewhat larger than H.264 at the same quality. Until the migration to FFmpeg.wasm 0.12 WebM used VP9, whose encoder does not work in that engine.
• CPU used (0–5) — VP8 encoder effort. 0=maximum compression and slower; 5=fast but larger file size. 2 offers a good balance for web use.
• Audio: re-encoded to Opus 128k by default. The Audio dropdown lets you choose 128k, 320k, remove the track or keep it without re-encoding (-c:a copy; requires an Opus or Vorbis source).

Animated GIF — via FFmpeg

• Colours — size of the colour palette. GIF is limited to 256 colours at most. Reducing the palette noticeably lowers the size, but can cause banding or loss of nuance.
• Dithering — dithering algorithm to simulate more colours than the palette allows. Bayer=ordered pattern, fast and with regular grain. Floyd-Steinberg=diffuses by error diffusion, more natural but heavier. None=no dithering, produces visible banding in gradients.
• Audio: the GIF format does not support audio; it is always discarded.
• Recommendation: reduce the resolution (480p) and FPS (12–15) to keep a reasonable size. GIF is inefficient compared with animated WebP or a short MP4.

Animated WebP — via FFmpeg

• Quality (0–100) — WebP encoder quality. 75 offers a good quality/size balance for animated clips. High values preserve detail but increase the size.
• Method (0/4/6) — encoder effort. 0=fast and larger file, 6=slow and maximum compression.
• Lossless — lossless encoding: reproduces the pixels exactly. Produces larger files than lossy mode.
• Audio: animated WebP does not support audio; it is always discarded.
• Comparison: at the same visual quality, animated WebP usually weighs between 30% and 60% less than GIF. Compatible with all modern browsers.

Audio formats

All audio formats are encoded with FFmpeg.wasm using the same engine as for video. The global controls Sample rate, Channels and Normalise volume are applied before the specific encoder.

MP3 — encoder: LAME (libmp3lame)

• Mode — VBR (Variable Bit Rate) adjusts the bitrate according to the complexity of the audio: better quality/size ratio. CBR (Constant Bit Rate) keeps a fixed bitrate: predictable and more compatible with streaming.
• Bitrate (V0–V9) — only in VBR mode. 0 = best quality (~245 kbps average), 9 = lowest quality (~65 kbps). V4 (~165 kbps) is a reasonable balance; V2 (~190 kbps) is common for music.
• Bitrate (64–320 kbps) — only in CBR mode. 128 kbps is usually the minimum acceptable for music; 192–320 kbps for high quality.
• Compatibility: universal — it plays on practically any device, browser and hardware player.

AAC (M4A) — encoder: aac (native FFmpeg)

• Bitrate (64–320 kbps, CBR) — 128 kbps is usually the minimum for acceptable music; 192–256 kbps for high quality. AAC offers better quality than MP3 at the same bitrate.
• Profile — only LC (Low Complexity) available. HE-AAC requires libfdk_aac, which is not included in this FFmpeg.wasm build.
• Container: the files are generated with the .m4a extension (MP4 Audio). Native compatibility on iOS/macOS and most modern browsers.

Opus — encoder: libopus

• Bitrate (24–256 kbps) — Opus offers perceptually good quality from ~48 kbps for voice and ~96 kbps for music. It is the most efficient codec at present: at 96 kbps it beats MP3 at 192 kbps in quality.
• Application — encoder mode: audio optimises music; voip optimises voice at low bitrates; lowdelay minimises latency (real-time streaming).
• VBR — Variable Bit Rate (enabled by default): the encoder adjusts the bitrate according to the signal to keep the quality constant. Disabling it forces CBR (fixed bitrate).
• Container: .opus files inside OGG. Compatible with Chrome, Firefox and Edge, but not with Safari, which has never supported the OGG container. If you need Safari, use WebM (Opus). Hardware support is still patchy.

WebM (Opus) — encoder: libopus

• Exactly the same Opus codec with the same controls (Bitrate, Application, VBR); the only thing that changes is the container: Matroska/WebM instead of OGG. At the same settings the size and quality are identical.
• When to use it: when the target is the web. WebM/Opus is played by Chrome, Firefox, Edge and also Safari from version 18.4 (March 2025), whereas .opus in OGG does not work in Safari in any version. It is also the native container of MediaRecorder, so it fits with audio recorded from the browser.
• Extension: .weba is generated, the conventional extension of audio-only WebM. It also stops TRANSFORMER itself from taking the file for a video if you load it into the queue again.

OGG Vorbis — encoder: libvorbis

• Quality (0–10) — Vorbis VBR quality scale: 0 ≈ 64 kbps (more compressed), 10 ≈ 500 kbps (best quality). 5 (~160 kbps) is usually a good balance.
• Compatibility: a free and open format, widely supported in software but not universally in hardware. A good substitute for MP3 in environments where Opus has not reached yet.

FLAC — encoder: flac

• Compression (0–12) — encoder effort level. 0 = fast, larger file; 12 = slow, smaller file. Always lossless: the level only affects the size and the encoding time, never the quality. 5 is FFmpeg's default value; 8 is the level of the Lossless preset.
• Typical size: between 50% and 70% of the equivalent WAV, with no loss at all.
• Recommended use: archiving, masters, later editing. For distribution to the end user, Opus or AAC is usually preferable.

WAV — encoder: PCM

• Depth — linear PCM sample format:
  • 16-bit PCM (pcm_s16le): CD standard. Enough for most uses.
  • 24-bit PCM (pcm_s24le): greater dynamic range, common in production and modern hardware (samplers, grooveboxes).
  • 32-bit float (pcm_f32le): practically unlimited dynamic range, common in editing workflows that apply chained processing.
• Compression: none — the files are large (≈10 MB/min in stereo 44.1 kHz 16-bit). Universal compatibility on music hardware.