What's New

v1.15.0 — Relight, Voices & Subtitles (2026-10-02)

New Relight page: change the lighting of a photo. Describe the light ("warm sunset light, beach") and pick where it comes from. The subject keeps its pose and face, and the background is redrawn to match.
Change voice on Audio Tools: sing or speak in another voice (RVC), with a pitch control. Tick "keep the music" to turn a song into a cover in the new voice. "Find voices…" searches voice-models.com, and voices you add can be used by everyone.
Speech to text on Audio Tools: get a transcript and subtitles (.srt) from a recording. You can also have both translated into another language, and the subtitle timings stay the same.
New Gallery page: every image you've made in one big grid that keeps loading as you scroll. Make the tiles bigger or smaller, search and filter like History, and click any image to view it full screen.
Qwen Image Edit 2509: combine up to three pictures in one edit. "Replace background" puts the subject of your photo into the scene of another.

v1.14.0 — Polish (2026-10-01)

New LoRAs page: add LoRAs from Civitai by pasting a link, ID or file hash, or search Civitai right on the site. Once added, everyone can use them. LoRAs nobody uses for 2 days are removed, and you can add them again any time.
Dark theme: pick Auto, Light or Dark with the button in the top bar. Auto follows your device.
Simple and Advanced settings on every tool. Simple shows the essentials and a Quality choice (Fast, Balanced, High) instead of step counts.
Every setting has an ⓘ button that says what it does, shows its default and can reset it.
Inpaint: undo and redo for the mask (buttons, or Ctrl+Z and Ctrl+Shift+Z).
A banner tells you when the connection drops and when it's back. Your jobs keep running meanwhile.
Install the site as an app from your browser's menu. Job notifications now also work there.
Send feedback from the link at the bottom of every page. Failed jobs have a "Report a problem" button that includes the job.
Announcements from the admin, and a one-time "see what's new" note after an update.
Easier to use with a keyboard or screen reader.
API keys can be limited to reading or writing and set to expire. The account page shows each key's webhook deliveries, and the API docs have a "Copy curl" button on every endpoint.

v1.13.0 — Better Images (2026-10-01)

Text to Image (SD / SDXL) can start from an image: add a Start image and choose how much to repaint.
Reference image: make the result follow another image's style or subject.
Hires fix: upscale up to 2× and repaint for sharper detail.
Detailer: finds faces and hands and redraws each one at full size.
To fix an image you already have, send it to SD / SDXL, set Repaint to 0 and turn on the detailer or hires fix.

v1.12.0 — Workflows & Safety (2026-10-01)

Workflows: run several tools in a row with one click, like "generate → cut out as a sticker" or "photo → vector logo". Use the built-in ones or build and save your own.
Batch mode: run a prompt several times at once, compare the same settings across models on Text to Image, and process many images at once in Image Tools and Photo Restore.
New Terms of Service. On your first visit you'll be asked to confirm you're 18 or older and agree to them.
Editing a photo of real people now asks you to confirm they agreed to it.
Shared pages have a Report link, and you can appeal a safety flag you think was a mistake.
Uploaded photos have their location and camera data (EXIF) removed.
The privacy policy explains the new 90-day job log and how reports are handled.

v1.11.0 — Find Your Work (2026-10-01)

Search your history by prompt or model, and filter it by tool, by kind (images, 3D, audio, text, failed) and by date.
Star jobs as favorites, and show only your favorites.
Run again: one click repeats a job with a new seed, from its card or its History page.
Your recent prompts and input images on each tool are one click away, next to Saved prompts and Saved images (guests too).
Drop or paste an image you made here onto any tool page to load the settings it was made with.

v1.10.0 — Fair Queue & New Tools (2026-10-01)

Job cards show roughly how long until a job starts ("Queue position 3 of 5 · ~2 min"), and the Generate bar shows the wait and run time before you submit.
Clearer job states: "Loading Flux…", "Up next — waiting for the GPU", "Retrying (attempt 2 of 3)".
A badge in the menu shows your running and queued jobs on every page, with the GPU's status. The home page shows it too.
The queue now takes turns between people, so one person's big batch doesn't hold up everyone else's first job.
Failed jobs explain what went wrong in plain words and have Retry and Retry smaller buttons (also on the job's History page). Failed jobs no longer count against your hourly or daily limit.
Your Account page shows how much of your limits you've used. Guest jobs show when they'll be deleted.
Sending the same job twice within a few seconds asks first.
New: Erase on the Inpaint page removes what you paint over, no prompt needed.
New in Image Tools: Sticker (cut out with a white outline), Read text (OCR) and Image → SVG.
New page: Audio Tools. Split a song into vocals, drums, bass and the rest, make a karaoke track, or clean background noise out of speech. Results are lossless FLAC files. Guests can use it too.
Results now have labels where it helps (Vocals, Drums, …), and speech results can be sent to Audio Tools.
REST API: jobs show an estimated wait and what kind of failure happened (errorKind), and POST /jobs accepts an Idempotency-Key header for safe retries.

v1.9.0 — Quality of Life (2026-10-01)

Get a browser notification (and a "(1) Done" tab title) when a job finishes while you're in another tab.
Click any result to view it full screen: zoom in, flip through results with the arrow keys or a swipe, and download or send it from there.
On image tools, a Compare slider shows the original next to the result.
Paste an image with Ctrl+V, or drop one anywhere on the page, to use it as the input.
Every tool page now remembers your prompt and settings if you leave or refresh. A Reset button puts the defaults back.
Each finished job has a Settings panel with buttons to copy its prompt, seed or settings.
Ctrl+Enter now generates from anywhere on a tool page.
The menu is split into smaller groups: Generate, Edit, Photo, 3D and Speech.

v1.8.0 — Safety (2026-10-01)

Prompts and images are now checked by a safety filter. If something gets flagged, the job is stopped or the result is blurred, and you'll see a message explaining why.
Flagged material is held for an admin to review, even if you use "Delete my data". The privacy policy has been updated to say so.
REST API: jobs now include a flagged field, and a flagged prompt is rejected with an error.

v1.7.0 — REST API & New Look (2026-09-30)

A fresh, cleaner design across the whole site, including a new home page.
Tool pages have a new layout: settings on the left, images and results on the right, and the Generate button always within reach. Ctrl+Enter also generates.
Aspect-ratio buttons on Text to Image, Flux and Qwen, plus a "Custom" option for your own size.
The site now works properly on phones.
New REST API at /api/v1: submit jobs, upload images, check on or cancel jobs, and download results. Create personal API keys on your Account page. Docs are at /api/v1/docs.
Get notified when a job finishes by polling, a live event stream, or a webhook.

v1.6.0 — Your Work (2026-09-27)

Run several jobs at once. Each tool page shows your jobs with progress and previews, and keeps them after a refresh.
Remix: reopen any past job with its prompt, settings and images already filled in.
Manage your history: delete jobs, or download everything (or just a selection) as a zip.
Share a job's results with a private link (off unless you turn it on).
Save your favourite prompts, styles and negative prompts on the Presets page.
"Delete my data" on your Account page removes all your jobs, files, links, presets and API keys.

v1.5.0 — Reliability (2026-09-13)

Your jobs are no longer lost when the server restarts or you refresh the page.
Guests keep their jobs between visits.
Large images (up to 50 MB) now upload reliably.
Fixed guests not being able to see their results after a job finished.

v1.4.0 — Photo Restore & Voices (2026-08-02)

New Photo Restore page: colorize black-and-white photos, fix faces, and upscale 2× or 4×.
Text-to-speech now speaks 9 languages with 54 voices (added Spanish, French, Hindi, Italian, Portuguese, Japanese and Chinese).

v1.3.1 (2026-07-12)

Fixed cancelling a job sometimes stopping the wrong one.
Stability fixes for text-to-speech.

v1.3.0 — Inpaint & Image Quality (2026-07-05)

New Inpaint page: paint over part of an image and only that area gets redrawn, at full resolution.
"Send to…" buttons pass a result straight into Edit, Inpaint or Restore.
Aspect-ratio presets on the Flux and Qwen pages.
Optional automatic quality tags for Pony and Illustrious models.
Faster 8-step option for Qwen Edit.
LoRAs now show their trigger words.
Fixed SDXL images sometimes coming out completely black.

v1.2.0 — Security & Fixes (2026-06-14)

Security improvements and sensible limits on image size, steps and image count.
Fixed some LoRAs loading the wrong version on Pony models.
Fixed jobs occasionally getting stuck in the queue.
Fixed long Flux jobs sometimes failing partway through.
Fixed broken links around the site.

v1.1.0 — Flux, Qwen & Speech (2026-05-24)

Edit a photo by describing the change, with Flux Kontext and Qwen Image Edit.
New text-to-image models: Flux.1-dev and Qwen Image.
Text-to-speech.

v1.0.0 — First Release (2026-05-03)

Text-to-image and inpainting.