InvokeAI 7.0.0 (alpha 1)

Okay. This is InvokeAI 7.

It's the biggest release we've ever made: about 320 pull requests, 2,600 commits and more late nights than anyone involved will admit to. If we listed everything that changed you'd be scrolling until next Tuesday, so here are the highlights. The full documentation for v7 lives at invoke-ai.github.io/InvokeAI-7.

Before you start, two practical notes:

• v7 needs Python 3.12. Python 3.11 is no longer supported. The Launcher handles this for you.
• The new interface is the default. If you need the old one, start Invoke with --web-legacy, but be aware that the legacy front end does not support all of v7's new features, and some of the old features may no longer work as expected.

And now here's a rundown of what's new:

🗂️ Projects: wander off and come back

You can now start something new without the nagging feeling that you're about to lose what you were working on. Everything lives in a project: your canvas, your generation settings, your workflows, your queue history, and a board of its own for everything the project produces.

Keep several projects open as tabs. Close one and go do something else. Come back next week from the new Launchpad and it's right where you left it.

Projects autosave. Edits go to your browser first and then sync to the server, so a dropped wifi connection, a crashed tab, or the same project open in two windows won't eat your recent work. You can also export a whole project as an .invk file to move it to another machine or just keep it somewhere safe.

📖 Projects guide

⚡ One workbench, your way

The separate tabs with a loading screen between them are gone. v7 is a single workbench built from widgets: Generate, Canvas, Gallery, Preview, Workflows, Video, Upscale, Image Map and more. Drag them wherever you like, switch between layout presets, or pop a widget out into its own floating window. Floating windows behave like real desktop windows: they resize from any edge, maximise on a double-click, and remember where you left them. Lost? Press Mod+K and type what you want.

It's also fast, and we intend to keep it that way. Widgets don't load until you open them, and every page has a strict budget for how much it downloads. CI flatly refuses any change that goes over. We're a little smug about this one.

📖 The Workbench · Hotkeys

🎨 A canvas you can actually paint on

The old canvas was great at what it was built for, which was inpainting. We wanted real, fast image editing with generation built in, so we wrote a new canvas engine from scratch.

It has layer groups (nest away), 16 blend modes, non-destructive adjustments, and a layers panel that stays responsive at 2,000+ layers and works entirely from the keyboard. There's a pressure-sensitive brush, marquee and lasso selections, gradients, shapes, a text tool that accepts your own fonts, a magnifying colour picker, and click-to-select objects. There's also a history panel you can scrub through, and PSD export for when your art needs to go and meet Photoshop.

Inpainting, regional guidance, control layers and staging are all still there, in the same document as everything else.

📖 Canvas introduction · Canvas guide · Layers panel

🎬 Video, now with sound

There's a whole Video panel with its own prompt, a built-in Video layout (Alt+3) and a tile on the Launchpad. It reshapes itself around whichever model you pick, and if the Invoke button is greyed out it tells you why.

• MiniMax H3 makes video with audio. Its Ref2VA mode accepts up to three video and nine image references, and you can refer to each one in your prompt (<Video 1>, <Picture 1>). Turbo LoRAs get it done in 4 to 8 steps.
• LTX-2.5 also does stereo audio, plus two-stage upscale-and-refine up to 1536p, audio-to-video and video-to-audio, first and last keyframes, interpolation and extend. Its prompt encoder now runs on 16 GB cards, and there's an NVFP4 distilled transformer in the starter models.
• Wan 2.2 (T2V, I2V, TI2V-5B and Lightning) gets the same panel treatment.
• Your videos remember how they were made. Generation metadata is embedded in the MP4, so you can download a clip, upload it to another install, and still recall its settings.
• Uploads just work. .mov, HEVC and ProRes files are converted to H.264 MP4 on the way in, and audio files become waveform videos.

📖 Video generation · MiniMax H3 · LTX-2.5 · Wan 2.2

🖼️ A gallery that knows what's in it

• Rebuilt from the ground up, with reliable ordering, a starred strip above the grid, drag-to-board, and live updates when new media arrives from another tab or a script.
• Semantic search. Type what you're looking for ("red car at night"), or drop in an image to find similar ones. Ever wanted to find everything you made from one particular reference image? Now you can.
• The Image Map. Every image and video you've made, laid out by visual similarity on a zoomable map. It finds clusters by itself and names them ("portraits", "foggy forests"...). Click a point to jump to it in the gallery, or select a whole cluster at once. It stays quick on galleries of 170,000+ images.
• An intermediates manager in Settings shows how much disk space each project's temporary files use, and cleans them up without touching anything an open project still needs.
• Touch support everywhere: hold to drag, swipe between images, pinch to zoom.

📖 Gallery · Image Map · Intermediates

🧩 Workflows, with loops

• A library you can browse. It's a card grid with thumbnails and model badges. Each workflow shows which models it needs, and you can install all the missing ones with a single click.
• Workflows live in your project. Opening a library template gives you your own copy to mess with. The shared library only changes when you choose Save to library.
• For / ForReturn loops. Loop over a collection while carrying state between iterations, stop early, nest loops, and resume after a restart. New Concat, Zip and Cartesian collection nodes go with them.
• Media fields accept files directly. Drop, upload or pick from the gallery, and choose a frame for video inputs.
• Outdated nodes offer Update node or Update all, and you can export a workflow as a PNG.

📖 Workflows · Loop nodes

🧠 Models, quantisation and GPUs

Besides MiniMax H3 and LTX-2.5, a lot of work went into making the models you already have smaller, faster and less likely to fall over:

• Smarter VRAM handling. Model loads no longer starve each other, freshly loaded models aren't evicted straight away, and installing a pile of models no longer reads entire checkpoints into RAM. When you switch between model architectures, models the new generation doesn't need move out of VRAM automatically.
• No more mysterious VAE hangs on Windows. Decodes that could take minutes on NVIDIA cards after a few generations are fixed.
• Multi-GPU. Cancel stops mid-step and frees VRAM, and prompt tools and image indexing hop onto whichever GPU is idle.
• ROCm moves to torch 2.13 + ROCm 7.2, with fixes for black images and crashes on long video clips.
• Tiled FLUX.1 VAE. Decoding stays at around 1.2 GB however big the image is.
• Anima now supports the 2.9B and 3.8B community finetunes.
• Works offline. Tokenizers and encoder configs ship with Invoke instead of being downloaded from Hugging Face.

📖 Model families · FP8 storage · Low-VRAM mode

✨ And a bunch more

• Dynamic prompts, wildcards (with search and autocomplete) and prompt templates. 📖 Advanced prompt syntax
• A redesigned Generate panel, scrubbable number fields, and seed modes (random, fixed, increment, decrement). 📖 Generate panel
• An Upscale widget, now with FLUX.1 support. 📖 Upscale
• Recall All and Remix now restore the LoRAs (concepts) an image was made with, at their original weights.
• Paste media anywhere, an install queue panel, and a Diagnostics panel with JSON export (very handy for bug reports 😉).
• More translations, a high-contrast mode, searchable settings, feature hints, a What's New panel, and a Catppuccin Mocha theme 🐱.
• Full tablet support.

💜 Contributors

v7 was built by a small, slightly sleep-deprived crew. Huge thanks to everyone who wrote code for this release:

@joshistoast

@lstein

@Pfannkuchensack

@JPPhoto

@primorLee

@unmarco

And an equally big thank-you to everyone who tested early builds and told us, kindly and otherwise, what was broken.

🛟 Addendum: Restoring your v6 database

If you ran v7 against your existing root directory and now want to go back to v6, you can restore the backup that Invoke made just before it upgraded your database.

What you'll lose: anything you did while running v7. Images and videos you generated in v7 stay on disk in your outputs folder, but they won't appear in the v6 gallery. Models you installed in v7 stay in your models folder but won't be listed in the Model Manager until you install them again (the Scan Folder option makes this quick).

1. Quit InvokeAI completely. Make sure no Invoke process is still running.

2. Open the databases folder inside your InvokeAI root directory. (If you set db_dir in invokeai.yaml, look there instead.)

3. Find the backup. You'll see files named like this:

   invokeai.db
   invokeai_backup_20261005-143012.db
   

   The timestamp is YYYYMMDD-HHMMSS. Choose the backup made when you first launched v7: it holds your database exactly as v6 left it. Earlier upgrades may have left older backups in the folder too, so pick by date. If you aren't sure which one it is, it's usually the oldest backup created on or after the day you installed v7.

4. Move the upgraded database out of the way. Don't delete it yet. Create a folder such as v7-db and move these files into it (the last two may not exist, which is fine):

   invokeai.db
   invokeai.db-wal
   invokeai.db-shm
   

   [!IMPORTANT]
   Don't skip the -wal and -shm files. If they're left behind, SQLite may try to apply v7's leftover changes to the restored database.

5. Copy the backup into place and rename the copy to invokeai.db. Copy rather than move, so the original backup stays as a safety net.

   • Windows (PowerShell): Copy-Item invokeai_backup_20261005-143012.db invokeai.db
   • macOS / Linux: cp invokeai_backup_20261005-143012.db invokeai.db

6. Check invokeai.yaml. v7 adds a few new settings (db_synchronous, fp8_compute, image_index_*, fonts_*). v6 won't start if any of these are in the file, so remove any you added.

7. Start InvokeAI v6 (6.14.2). Your boards, images and models should be back where you left them.

If v6 still refuses to start, or you can't find a backup, ask in the Discord or open an issue before you try anything else, and keep every .db file you have.