Format compatibility

These are engine capabilities with explicit limits. A file is only successful when the engine produces an output. No claim of 204 verified conversions is made.

Tool	Accepted input	Output	Limits
Images	JPG / PNG / WebP / GIF / BMP / AVIF / SVG / TIFF / ICO	PNG / JPEG / WebP / GIF / AVIF / BMP / TIFF / ICO / raster SVG	Browser decoder required. Static first frame / TIFF page only; EXIF and profiles are not preserved. JPEG uses white background.
Video	MP4 / WebM / AVI / MOV / MKV / FLV / WMV / MPEG / GIF	MP4 / MOV / MKV (H.264/AAC), WebM (VP8/Opus), AVI (MPEG4/MP3), FLV (FLV1/MP3), WMV (WMV2/WMA2), MPEG (MPEG2/MP2), GIF	FFmpeg decoder availability varies by codec, regardless of filename. One video and first optional audio track.
Audio	MP3 / WAV / OGG / M4A / AAC / FLAC or video audio	MP3 / WAV / OGG / M4A / FLAC	FFmpeg build must contain the input decoder and output encoder.
Documents	TXT / MD / HTML / RTF / DOCX / text PDF / Word97+ DOC	TXT / MD / HTML / RTF / PDF / DOCX	PDF extracts text only; scanned PDFs require unsupported OCR. MD↔HTML and DOCX→HTML/MD preserve basic structure. Other document outputs use plain text. Word97+ DOC extracts main body text only; older/encrypted/corrupt variants rejected. No DOC export.
Subtitles	Browser-decodable audio/video audio	SRT	English Whisper tiny.en, local model download required. Browser audio decoding required.

Fidelity and verification

Image output is rendered through canvas, which loses metadata and animation. Most document outputs use a plain text intermediate. MD↔HTML and DOCX→HTML/MD preserve basic headings and lists, with active content and embedded images removed. PDF input extracts text, not page layout or OCR. Legacy Word97+ DOC reads the bounded main document piece table (32 MB / 5 million character limits). It does not execute macros or preserve fields, headers, footnotes or layout; encrypted, pre-Word97 and malformed DOC variants are rejected. DOC export is unsupported. Video output keeps the first video and optional first audio stream. Fast cuts depend on keyframes; precise cuts re-encode. Compression is not guaranteed to reduce size.

Automated unit checks cover time range validation, FFmpeg command generation and valid subtitle timing. Browser engine and model downloads, real codecs, color fidelity and device compatibility require device acceptance testing; the matrix is not a record of every pair having passed on every browser.

Engine sources and licenses

ffmpeg.wasm wrapper 0.12.15 (MIT), core 0.12.10 (GPL-2.0-or-later build; FFmpeg and included codec licenses apply). Uses a single-thread WASM engine; source and build instructions are linked in that repository.
Transformers.js 3.8.1 (Apache-2.0) and Whisper tiny.en ONNX model (underlying OpenAI Whisper model: MIT; see model card for the ONNX conversion source). Initial model download requires internet. Cache storage can be cleared or evicted.
pdf-lib 1.17.1 (MIT), Mammoth 1.10.0 (BSD-2-Clause), docx 9.5.1 (MIT).
PDF.js 5.4.394 (Apache-2.0), UTIF.js 3.1.0 (MIT), Turndown 7.2.1 (MIT) and Marked 15.0.12 (MIT).
CFB 1.2.2 (Apache-2.0), used for OLE container parsing. The text extractor follows Microsoft MS-DOC.
gifenc 1.0.3 (MIT) and jSquash AVIF 2.1.1 (Apache-2.0 wrapper; AVIF/AOM codec notices apply). Pinned package license and source manifest; available package license texts are distributed alongside it.

Review engine source licenses and codec patent obligations for your distribution and jurisdiction. The browser tools do not grant rights to uploaded content.