Changelog
Every new tool, improvement and fix, newest first. To hear about changes as they ship, subscribe to the RSS feed.
October 2026
- 24 October · Fixed: Model downloads now survive a dropped or switching connection on a phone. A retry waits for the network to come back and picks up where the download stopped, instead of starting over or giving up within a second. Reported by YD-DKPZ. Thank you.
- 24 October · Fixed: Voice cloning now evens out the loudness of your sample before learning from it. A quiet phone recording was learned as a noticeably different voice from the same words recorded louder. Reported by YD-N3V9. Thank you.
- 23 October · Fixed: Long recordings no longer repeat or drop words where one 30-second stretch meets the next: the audio is now cut only in pauses, found by a small voice-activity model. On three minutes of real speech, word errors fell from 27% to 2%.
- 22 October · Improved: Norwegian transcription is much more accurate: choose Norwegian and it uses NB-Whisper from the National Library of Norway, same size and speed. On a public Norwegian test set it made about a third of the errors on short clips, and a little over half on five minutes of continuous speech.
- 22 October · Fixed: A voice recorded for cloning is now taken from the microphone as it is, without the browser’s noise suppression and auto-gain, which the model learned and repeated as a processed, robotic sound. Reported by YD-2M10. Thank you.
- 22 October · Fixed: Voice cloning no longer fails at “learn voice” with a long recording. It now learns from the first ten seconds, which also used to run phones and 4 GB laptops out of memory on shorter clips. Reported by YD-T5W2, YD-HR5H, YD-W7D0, YD-TEZ4, YD-M9BF. Thank you.
- 21 October · Improved: Swedish transcription is much more accurate: choose Swedish and it uses KB-Whisper from the National Library of Sweden, same size and speed. On a public Swedish test set it made about a third of the errors.
- 21 October · Fixed: Transcripts of recordings longer than 30 seconds no longer repeat a sentence where one 30-second stretch meets the next. On a five-minute test it cut the word errors from 16% to under 7%, in every language.
- 20 October · Improved: Danish transcription is noticeably better: choose Danish with the Quality model and it uses a Whisper model trained on Danish speech, same size and speed. On a real two-hour meeting it made about a fifth fewer errors.
- 20 October · Fixed: Transcribe now reads long recordings on phones. A two-hour MP3 needs about a quarter of the memory it did, and a big file is no longer given up on after 30 seconds of reading. Reported by YD-EVA6. Thank you.
- 20 October · Fixed: Subtitle translation now translates a sentence whole when it runs over several subtitles, then splits it back across them. Verbs and grammar no longer break at the subtitle edges. Reported by YD-W681. Thank you.
- 20 October · Fixed: Text to speech now checks the graphics card before using it. On a card that produces noise instead of speech, the voice runs without it, so it comes out clear. Reported by YD-66AR. Thank you.
- 20 October · Fixed: Transcribe no longer loses a long run when the graphics card fails part-way. It keeps what was already transcribed and carries on without the graphics card, and Try again now really starts afresh. Reported by YD-EWC5, YD-K3NW, YD-GY71, YD-EJQA. Thank you.
- 19 October · Fixed: Transcribe now says why a file could not be read: no audio track, an empty or unfinished file, or a damaged container. Empty files no longer leave Safari stuck on the reading step. Reported by YD-12BV, YD-Q41N, YD-JY1J, YD-HPBV, YD-1EJF, YD-A2CA, YD-YV9K, YD-1Q6S, YD-4C3H, YD-7ZW8, YD-DHJW, YD-QYE1, YD-QJMY, YD-H1S4, YD-R4E5, YD-EBQH. Thank you.
- 18 October · Improved: Upscale images has a Best detail mode for sharper text and fine texture, on a GPU, and shows the result as a before/after you can zoom to 100%.
- 9 October · New tool: Make PDF fillable
- 9 October · Improved: The PDF form filler can now find the boxes and checkboxes on a PDF that has no fields, ask you about the few it is unsure of, and save a PDF anyone can type into.
- 9 October · Improved: Identify speakers now uses NVIDIA’s Nemotron 3 Diarization model. It tells up to 8 people apart and found the right number of people in every test meeting we ran, where the old method often reported 10 or more. The first run downloads about 120 MB.
- 6 October · New tool: Word to PDF
- 6 October · New tool: PowerPoint to PDF
- 6 October · New tool: Excel to PDF
- 6 October · Improved: The licences page now credits LibreOffice, which the Word, PowerPoint and Excel to PDF tools run in your browser, and shows the one change we made to it.
- 5 October · Improved: Translation now uses the models Firefox uses, from Mozilla: more accurate, a 14–38 MB download per language instead of about 630 MB, and much faster.
- 5 October · Improved: Translation now runs on the M2M-100 model from Meta, which is licensed for any use, and its download is about 30% smaller.
- 5 October · Improved: The licences page now credits every model the tools run, with its licence and notice.
- 5 October · Fixed: Text to speech no longer offers the voices Cosette and Jean, whose recordings are licensed for non-commercial use only.
- 4 October · Improved: A new home page, with search at the top, popular tools one tap away, and your starred and recent tools first.
- 4 October · Improved: Fern answers faster in long chats, because it no longer rereads the whole conversation every turn.
- 4 October · Improved: Fern writes long answers in full when you ask for them, and stops if it starts repeating itself.
- 4 October · Improved: Text to speech and voice cloning now download 48 kHz WAV files, ready to drop into a video editor.
- 4 October · Improved: The SDR to HDR page explains more clearly what an HDR photo with a gain map is, in every language.
- 4 October · Fixed: Fern recovers when the graphics card runs out of memory, and tells you that memory was the problem.
- 4 October · Fixed: Long spoken conversations with Fern no longer run out of room in the middle of a reply.
- 4 October · Fixed: SDR to HDR only says you are seeing the brighter highlights when your browser can actually show them.
- 3 October · New tool: Clone your voice
- 3 October · Improved: Text to speech speaks six languages, with a library of 26 voices.
- 3 October · Improved: New guide: how voice cloning works, and when it is legal.
- 2 October · Improved: The home page sorts tools into ten groups, finds them by what you want to do, and keeps your favourites under Your tools.
- 1 October · New tool: Blur faces & plates
- 1 October · Fixed: The New badge only marks tools launched in the last two weeks.
September 2026
- 27 September · Improved: When two tools could do what you searched for, the tool switcher says which one to pick.
- 21 September · Improved: A tool you have used once keeps working with the network switched off.
- 20 September · Fixed: The transcription page no longer says it works offline. It does not yet.
- 19 September · New tool: Compress video
- 19 September · Improved: Our most visited pages explain how to check in your browser that your files are never uploaded.
- 15 September · New tool: Convert HEIC to PDF
- 15 September · New tool: Rotate an image
- 15 September · New tool: Flip an image
- 15 September · New tool: Audio joiner
- 15 September · New tool: Convert FLAC to WAV
- 15 September · New tool: Convert AIFF to MP3
- 14 September · New tool: Convert M4A to WAV
- 14 September · New tool: Convert AAC to MP3
- 14 September · New tool: Convert Opus to MP3
- 14 September · New tool: Metronome
- 14 September · New tool: Tap BPM
- 14 September · Improved: The metronome has eight meters, and compound ones like 6/8 accent their groups.
- 13 September · New tool: Convert MP4 to MP3
- 13 September · New tool: Video to MP3
- 13 September · New tool: Convert WAV to MP3
- 13 September · New tool: Convert MP3 to WAV
- 13 September · New tool: Convert M4A to MP3
- 13 September · New tool: Convert OGG to MP3
- 13 September · New tool: Convert FLAC to MP3
- 13 September · New tool: Convert WebM to MP3
- 13 September · New tool: Convert JFIF to JPG
- 13 September · New tool: Convert JFIF to PNG
- 13 September · New tool: Convert BMP to PNG
- 13 September · New tool: Convert BMP to JPG
- 13 September · New tool: Convert GIF to WebP
- 13 September · New tool: Convert WebP to PNG
- 13 September · New tool: Convert WebP to JPG
- 12 September · Fixed: Fern’s voice mode no longer cuts off the first syllable you say.
- 11 September · New tool: Convert HEIC to PNG
- 11 September · New tool: Convert SVG to PNG
- 11 September · New tool: Convert SVG to JPG
- 11 September · New tool: Convert AVIF to JPG
- 11 September · New tool: Convert AVIF to PNG
- 11 September · New tool: Convert GIF to PNG
- 11 September · Fixed: Change speed & pitch plays sound again. The live preview was silent.
- 10 September · New tool: Change speed & pitch
- 10 September · New tool: Resize to a KB limit
- 10 September · Improved: Resize to a KB limit has presets for the IBPS application photo, signature, thumb impression and declaration.
- 9 September · Improved: All tools on one page, listed by what goes in and what comes out.
- 8 September · Improved: The tool name in the header opens a switcher with your recent and pinned tools. Press / to open it from anywhere.
- 8 September · Improved: Send a finished transcript straight to the translator or the summarizer.
- 8 September · Improved: Larger body text across the site, easier to read on a phone.
- 1 September · Fixed: The transcriber never loses a recording, and uses less memory while it works.
August 2026
- 30 August · Fixed: Restore old photos is more reliable on phones, and says plainly when a device cannot run it.
- 28 August · New tool: Voice enhancer
- 27 August · New tool: Restore old photos
- 26 August · New tool: Private AI chat
- 26 August · Improved: Private AI chat typesets maths, and its voice conversation mode was rebuilt.
- 26 August · Fixed: Blurred photo backgrounds no longer show bands in smooth areas like skies.
- 24 August · Improved: Big model downloads check you have room before they start, and the size is shown before you press the button.
- 24 August · Improved: EPUB to PDF sets tables as tables, footnotes at the foot of the page, and Greek and Cyrillic text in a real book font.
- 24 August · Improved: New guides: what OCR is, what a HEIC file is, what an EPUB file is, and how to transcribe audio to text.
- 24 August · Fixed: The transcription page no longer claims to detect the spoken language by itself.
- 23 August · New tool: EPUB to PDF
- 23 August · Improved: Crop PDF pages by dragging a box over the page.
- 22 August · Improved: The PDF editor can extract images and make scanned PDFs searchable.
- 21 August · New tool: Convert HEIC to JPG
- 21 August · Improved: The PDF editor can flatten a PDF, baking form fields into the page.
- 21 August · Fixed: Text to speech reads web addresses and file names properly, like “dot com”.
- 19 August · New tool: View & remove photo metadata
- 19 August · Improved: The metadata remover also removes identifiers stored outside EXIF, and shows on a map where a photo was taken.
- 18 August · New tool: Unlock PDF
- 15 August · New tool: Instrument tuner
- 14 August · New tool: Audio to MIDI
- 14 August · Improved: Audio to MIDI plays back what it heard on a sampled grand piano, and records straight from your microphone.
- 14 August · Improved: Upscale images runs on Real-ESRGAN, with a Detail control.
- 14 August · Fixed: Audio to MIDI places notes on time. They used to land late.
- 14 August · Fixed: Upscale images no longer shrinks large photos.
- 13 August · New tool: Video to GIF
- 13 August · Improved: Every tool page says which AI model it runs, or that it runs none.
- 11 August · New tool: Handwriting to text
- 11 August · Improved: The background remover can give you the cut-out on solid white as well as transparent.
- 11 August · Improved: Blur photo backgrounds focuses on the face by default.
- 11 August · Improved: Remove background noise accepts video and cleans its soundtrack.
- 10 August · Improved: Every tool walks you through the same numbered steps, from file to download.
- 9 August · Improved: Remove vocals runs on a stronger separation model, BS-PolarFormer, using your graphics card.
- 9 August · Improved: Click a result preview to see it full size.
- 8 August · New tool: Remove vocals from a song
- 8 August · New tool: SDR to HDR
- 8 August · New tool: Meeting notes
- 8 August · Improved: Pages for audio, image, PDF and text tools, each listing the whole category.
- 7 August · New tool: Text to speech
- 7 August · New tool: Redact PDFs
- 7 August · New tool: Fill PDF forms
- 7 August · Improved: Install YourDevice as an app from your browser.
- 5 August · New tool: Summarize text & transcripts
- 4 August · Improved: The main tools are available in French, Portuguese and Italian.
- 2 August · New tool: Blur photo backgrounds
- 2 August · Improved: Model downloads show real progress, in megabytes.
July 2026
- 31 July · New tool: Translate subtitles & text
- 30 July · Fixed: The transcriber no longer offers an auto-detect that could guess the language wrong. It starts in English.
- 29 July · Improved: PDF to Markdown reads scanned pages and keeps nested lists indented.
- 28 July · New tool: PDF to Markdown
- 28 July · New tool: Upscale images
- 28 July · Improved: Upscale images works in tiles, so large photos no longer run out of memory, and offers 2× or 4×.
- 27 July · New tool: Transparent background maker
- 27 July · New tool: Image to text (OCR)
- 26 July · New tool: Remove background noise
- 26 July · New tool: Edit PDFs
- 26 July · Improved: Choose the spoken language before you transcribe.
- 22 July · Improved: Transcribe iPhone .mov videos directly.
- 21 July · Improved: Transcripts mark when a different person starts speaking.
- 17 July · Improved: Download a transcript as a Word document.
- 16 July · New tool: Transcribe audio & video