Skip to content

v7.0.0 (beta 1)

Pre-release
Pre-release

Choose a tag to compare

View all tags
@lstein lstein released this 09 Nov 04:40
· 4 commits to main since this release
v7.0.0-beta.1
be31872

Caution

⚠️ Please read this before you install

If you point v7 at your existing InvokeAI root directory, it will upgrade your database, and that database will no longer work with InvokeAI v6. The migrations only go one way, and v6 refuses to open a database that contains migrations it doesn't recognise.

To try v7, create a brand-new InvokeAI root directory and experiment there. In the Invoke Launcher, choose a new, empty install location. When prompted for the version, select Manual and enter version v7.0.0-beta.1. For a manual install, pass --root /path/to/new/folder or set INVOKEAI_ROOT. Keep your v6 install and its root directory exactly where they are.

Upgraded your v6 database by accident? Don't panic. Invoke makes a backup before it migrates anything, and the startup log names that backup before the upgrade begins. See Restoring your v6 database at the bottom of these notes, or the Upgrading to v7 guide.

Warning

This is beta software. All the major features of 7.0 are in place, and we've spent four alphas shaking out bugs, but you will still find rough edges. Please keep backups of work you care about. When you hit a bug, please open an issue on GitHub and put v7 in the title, or tell us about it on our Discord. Screenshots, logs and the output of the Diagnostics panel all help a lot.

Note

Coming from a v7 alpha? You can install beta 1 over alpha 4 and keep using the same root directory. Beta 1 adds no database migrations of its own. If you're upgrading from alpha 3 or earlier, remember that alpha 4's migration is one-way, so once the beta has opened your root directory the older alphas won't start on it.

InvokeAI 7.0.0 (beta 1)

Okay. This is InvokeAI 7, and it's now in beta.

It's the biggest release we've ever made. Alpha 1 alone was about 320 pull requests and 2,600 commits, and four alphas later the count has grown a good deal more. Many of you are meeting v7 for the first time with this beta, so these notes start with the tour of what v7 is, then cover the major features added during the alphas, and finish with what's new in beta 1. The full documentation for v7 lives at v7.invoke.ai, which has itself been redesigned for this release (more on that below).

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.

Contents: What's new in InvokeAI 7 · Added during the alphas · New in beta 1 · Contributors · Restoring your v6 database · Full changelog

What's new in InvokeAI 7

🗂️ 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 boards 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 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 lose your work. (However, this is not a full "offline mode.") 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. Beyond the brush and eraser there are 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 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, starred items in their own strip above the grid, 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.
  • Intermediates, accounted per project. Settings shows how much disk space each project's temporary files use, and cleanup now leaves alone 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.
  • Video inputs. Workflow fields accept videos the same way image fields accept images (drop, upload or pick from the gallery), and you choose which frame to use.

📖 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:

Format What it buys you
FP8 Storage Enabled automatically when you install an fp8 file. FLUX.2 Klein 4B goes from 7.4 GB to 3.9 GB with identical output
FP8 Compute (opt-in, RTX 40-series and newer) Runs fp8 checkpoints on the tensor cores
int8_convrot One shared int8 scheme for FLUX.1, FLUX.2 Klein, Z-Image, Krea 2, Ideogram 4, Qwen3 encoders, PiD, MiniMax H3 and LTX-2.5 (and, since then, more families: see below)
NVFP4 / MXFP8 Comfy NVFP4 checkpoints stay packed, and MXFP8 files load
GGUF Qwen3-VL A quantised text encoder for Krea 2 and Ideogram 4
  • 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 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 · Which memory option do I need? · FP8 storage · Low-VRAM mode

✨ And a bunch more

  • Wildcards get a manager with search, and the prompt box autocompletes wildcard names. 📖 Advanced prompt syntax
  • A redesigned Generate panel, scrubbable number fields, and seed modes (random, fixed, increment, decrement). 📖 Generate panel
  • Upscaling is now a widget, so it can sit beside the Canvas or Gallery in any layout. 📖 Upscale
  • Recall All and Remix find an image's LoRAs (concepts) by install key, content hash or name, restore them at their original weights, and tell you about any they couldn't restore.
  • A Diagnostics panel with JSON export (very handy for bug reports 😉).
  • More translations, a high-contrast mode, searchable settings, and a Catppuccin Mocha theme 🐱.
  • Full tablet support.

Added during the alphas

If you followed the alphas you've seen these already. If not, here are the major features that landed after alpha 1, grouped by the release that introduced them.

Alpha 2

  • Multi-board projects. A project can have as many project-specific boards as you like, not just one. (@joshistoast)
  • MySQL and MariaDB. You can swap the default SQLite database for MySQL or MariaDB, which is particularly useful for multi-user installations. 📖 MySQL and MariaDB (@Pfannkuchensack)
  • More GGUF transformers: Krea-2 (including Q8_CR), Ideogram 4 and ERNIE-Image. (@Pfannkuchensack)
  • Atlas Cloud joins the externally hosted (API) image providers. (@binyangzhu000-sudo)
  • Boolean logic node (AND, OR, XOR, NOT, NAND, NOR, XNOR). (@JPPhoto)
  • reserve_vram_gb, a new invokeai.yaml option that holds back VRAM from the model cache for use during generation, to help with out-of-memory errors. (@JPPhoto)
  • The remote control API (used by, for example, PhotoMapAI) can send reference videos and first/last frames to the Video panel, and MiniMax H3 Ref2VA runs with audio-only references. (@lstein)
  • ROCm 10 libraries for better performance on AMD GPUs. (@Pfannkuchensack)

Alpha 3

  • Board sharing by role. Each board has an owner, and everyone else gets a role on it: Viewer, Contributor, Editor or Owner. Boards shared with you show up under Shared with me in the Gallery. Projects also keep each member's own settings (layout, generation settings, prompt history and so on) separate from the project's shared content, which is the groundwork for project collaboration later in 7.x. 📖 Multi-User Mode user guide (@joshistoast)
  • Qwen-Image-2.1, Base and the 8-step Turbo, for text-to-image and for Canvas img2img, inpaint and outpaint, as Diffusers folders, single files (bf16, scaled fp8, int8_convrot) and GGUF. 📖 Qwen-Image-2.1 (@Pfannkuchensack)
  • SageAttention (attention_backend: sage), an opt-in 8-bit attention kernel for RTX 30-series and newer. Whole generations run 15–35% faster at 2048² on an RTX 4090, and video gets 12–35% faster. 📖 SageAttention (@Pfannkuchensack)
  • More models in the Upscale widget: Z-Image, Anima and Krea-2 join SD1.5, SDXL and FLUX.1, with Creativity recalibrated to behave the same across models. FLUX.2 VAE nodes can tile. (@Pfannkuchensack)
  • Old favourites from the legacy interface: SDXL Refiner, per-family tuning controls (FLUX.1 DyPE, Z-Image and Qwen-Image shift overrides, Krea-2 style reference), the NSFW checker and invisible watermark, PiD decoding on Canvas, pressure-sensitive eraser, retry for failed queue items, and Load Workflow on any generated image. (@joshistoast)
  • Gallery: a Both / Images only / Videos only filter in the Gallery and the Image Map, and live updates for images made by external REST API clients. (@lstein)
  • Video extensions that don't degrade. Extend and Trim copy existing footage and audio without re-encoding wherever they can, and video_concat only re-encodes the transitions. (@lstein, @JPPhoto)
  • ComfyUI single-file CLIP-L and CLIP-G encoders, Krea-2 LoKr LoRAs, and kohya MiniMax H3 LoRAs. (@Pfannkuchensack, @Balrog994, @lstein)

Alpha 4

  • New model families: FLUX.3 Klein and MOSS-Image. FLUX.3 Image Klein from Black Forest Labs (distilled Klein and undistilled Klein Base) generates from text and edits or combines up to ten reference images. MOSS-Image from OpenMOSS (Base and SFT, Apache 2.0) works in Generate, on the Canvas and in workflows, and comes with a prompt enhancer. 📖 FLUX.3 · MOSS-Image (@Pfannkuchensack)
  • Reference images and edits: IP-Adapter character references for Anima (@lstein), up to three edit references for Krea-2 (which makes ai-toolkit pose and depth edit LoRAs work), and Qwen-Image-2.1 edits and LoRAs (@Pfannkuchensack).
  • New ControlNets and type-first control layers. FLUX.2 [dev] Fun-ControlNet-Union and InstantX Qwen-Image ControlNet Union are supported (@Pfannkuchensack). On the Canvas you now say what a control layer should control (Pose, Depth, Edges, Line art...) and Invoke picks the detector and the model (@joshistoast).
  • Tiny AutoEncoder previews. Set tiny_autoencoder_previews: true and install the matching starter model, and the in-progress preview becomes a close likeness of the final image. 📖 Tiny AutoEncoder Previews (@Pfannkuchensack)
  • int8_compute, an opt-in setting that runs int8 convrot models on the tensor cores of RTX 30-series and newer cards: 1.3–2× faster denoising on an RTX 4090, and up to almost 4× when the model only partly fits in VRAM. Plus fp8 and int8 single-file starter models, and Krea-2 int4, Anima GGUF, LTX-2.5 GGUF and Gemma-4 GGUF formats. 📖 int8 Compute · Model Format Support (@Pfannkuchensack)
  • Upscaling with FLUX.2 and Qwen-Image, a Model only mode that runs an upscaler like 4x UltraSharp on its own, and an Upscale submenu on every image. (@Pfannkuchensack, @joshistoast)
  • Canvas adjustments: Noise (film grain), Clarity, Texture and Dehaze on layers and groups, also available as workflow nodes. 📖 Image Detail and Contrast Nodes (@joshistoast, @JPPhoto)
  • Shared workflows. In multi-user mode, a Library workflow can be made Public for everyone on the server. (@JPPhoto)
  • Windows on ARM64, including NVIDIA's RTX Spark developer preview. (@lstein)
  • Working memory that fits the model. VRAM reserved for denoising and VAE decoding now comes from measurements of each model rather than rough estimates, so fewer out-of-memory errors without holding back more VRAM than needed. (@Pfannkuchensack)
  • Undo instead of confirm. Moves, removals, archives, stars and layout resets happen straight away and offer Undo. The Invoke button opens a card that shows where the run is going and anything blocking it, along with the fix. (@joshistoast)
  • Hosted-provider options, Canvas modes, model record editing, PBR map generation and image storage administration are back from the legacy interface. (@joshistoast)

New in beta 1

Beta 1 is mostly about filling the gaps: LoRA support for the families that didn't have it, the quantised files people actually download, and a lot of fixes from your alpha reports.

🧬 LoRAs for more families

📦 More model formats, and starters to go with them

  • FLUX.3: quantised single-file checkpoints load (#10093), and the FLUX.3 prompt upsampler installs as a Text LLM that you can use from Expand Prompt and the Text LLM nodes, including Comfy's int8 build at about 10.7 GiB (#10092). (@Pfannkuchensack)
  • MOSS-Image: fused, int8 convrot and GGUF transformers now load, as does Comfy-Org's int8 build of its Qwen3.5 encoder. (@Pfannkuchensack, #10176, #10177)
  • Qwen-Image: ComfyUI int8 convrot checkpoints (#10186) and GGUF builds of the Qwen2.5-VL text encoder (#9969). Qwen-Image-2.1 reference images now work with single-file models and GGUF encoders (#10153). (@Pfannkuchensack)
  • Wan: ComfyUI int8 convrot checkpoints (#10185) and the UMT5 text encoder as GGUF (#9967). (@Pfannkuchensack)
  • Text encoders: NVFP4 Qwen3-VL encoders for Krea-2 and Ideogram 4 (#9804) and ComfyUI single-file T5-XXL encoders (#9799). (@Pfannkuchensack)
  • 27 new starter models for formats that had none, including MOSS-Image in bf16, int8 and GGUF, Krea-2 Raw in fp8, int8, NVFP4 and GGUF, Qwen3-VL encoders, and fp8 builds of Anima, FLUX.1 dev, schnell and Kontext, and Z-Image Turbo. (@Pfannkuchensack, #10190)
  • Multi-part and unreadable GGUF files are refused at install time with a clear message instead of failing later. (@Pfannkuchensack, #10199)

👀 Tiny AutoEncoder previews for video and more

The sharp live previews from alpha 4 now cover Wan 2.2, Qwen-Image, Krea-2, Anima and LTX-2.5, using madebyollin's TAEHV decoders. Install the matching TAEW2.1, TAEW2.2 or TAELTX2.3 starter model and turn on tiny_autoencoder_previews. The Model Manager lists each preview decoder under every family it serves, so it's easy to see which one you need. (@Pfannkuchensack, #10041; @lstein, #10125)

📖 Tiny AutoEncoder Previews

📐 Saved aspect ratios

A new ratio picker replaces the aspect-ratio dropdown in Generate, the Canvas Frame tool and the Video panel. Built-in ratios appear as shape tiles, and you can name and save up to 50 of your own ratios ("Cinemascope 2.39:1", "Instagram 4:5"...). Because they're ratios rather than fixed sizes, they work across models with different native resolutions: picking one keeps the current pixel count. (@joshistoast, #10101; @lstein, #10130)

🎨 Canvas

  • Regional negative prompts and Auto-Negative for Anima, Z-Image and Krea-2. (@joshistoast, #10110)
  • Fuller Canvas recall. Canvas generations now also record their layers, geometry, processing settings and controls, so Recall All and Remix bring back much more of a Canvas generation than before, from the Gallery, Preview, staging and the queue. ControlNet, T2I-Adapter and IP-Adapter settings from older images recall as well. (@joshistoast, #10127)
  • Shape tool drawing and properties fixed, merge-down shortcut in the Layers panel, colour compensation applied when encoding images, masks hidden under previews again, and PSD export that no longer runs out of memory on big documents. (@joshistoast, @DustyShoe)

🎬 Video

  • A new video player in Preview and the workflow frame picker, the same in every browser, with a loop toggle, frame stepping, playback speed and keyboard shortcuts. Trim spans from the Video panel loop in the player. (@joshistoast, #10005)
  • Dynamic prompts now expand in the Video panel, so {a|b} and wildcards queue one video per combination. (@lstein, #10128)
  • The Preview menu and recall shortcuts work for videos. (@lstein, #10114)

🖼️ Gallery and workflows

  • Delete with confidence. Before deleting media or a board, the Gallery lists where that media is still used: canvas layers, reference images and workflow inputs in any of your projects, library workflows, and queued jobs. You can cancel or delete anyway. (@joshistoast, #10126)
  • File inputs for custom nodes. Node authors can declare a FileField, and users attach documents and data files straight from the node in the UI instead of typing a server path. The server manages the upload, with extension and size limits. (@Gohankaiju, #9302)
  • The workflow library renders and loads faster on large libraries. (@joshistoast, #10167)
  • The Uncategorized board can now be downloaded. (@chiliec, #10148)

📖 New documentation site

The docs at v7.invoke.ai have been rebuilt to look like the app, with topic-scoped search (also available from the app's Mod+K command palette), German, Spanish and Hindi translations, an OS/GPU picker on the install guides, a searchable community-nodes catalogue, and rewritten interactive guides, including a new Which memory option do I need? picker. (@joshistoast, #10044)

🐧 Linux on ARM64 with CUDA

Manual installs on Linux ARM64 machines with NVIDIA GPUs, such as the DGX Spark, now get CUDA builds of PyTorch with --extra cuda, without the old post-install workaround. (@JPPhoto, #10188, #10183)

🐛 Major bug fixes

  • The web client no longer resubmits a failing enqueue forever. After three server errors the run fails and shows the reason, and waiting runs now appear in the Queue panel. (@joshistoast, #10124)
  • Autosave recovers stalled project saves, and a stale model list no longer overwrites a change you just made. (@joshistoast, #10161)
  • GGUF and SDNQ weights move fully onto the GPU when a model is fully loaded, instead of leaving pieces behind in RAM. (@Pfannkuchensack, #10096)
  • Tiled VAE decodes and encodes reserve measured working memory, so fewer out-of-memory errors on large images. (@Pfannkuchensack, #9946)
  • Cached images no longer change when a node modifies the image object it was handed. (@Pfannkuchensack, #10138)
  • The FLUX.2 starter bundle now contains only the ungated, Apache 2.0 Klein 4B, and FLUX.2 [dev] takes its components only from Diffusers pipelines. (@joshistoast, #10120, #10121)
  • HiDiffusion ratio presets and denoising transitions are corrected. (@DustyShoe, #10105)
  • Tile seam blending starts at the lowest-energy column. (@MohammadHijjawi97, #9602)
  • Wan's serialized UMT5 tokenizer loads correctly. (@guishiw, #10049)
  • CLIP hidden states are preserved after the first skipped forward pass, and textual inversion embeddings are built one at a time to avoid races. (@wunianze666-netizen, #9806, #9844)
  • VAEs without a decoder.conv_in weight are no longer filed as SD1. (@stanleys12, #10029)
  • Krea-2 GGUF dequantisation uses the right dtype on Apple MPS. (@grunblatt-git, #9573)
  • Run on CPU sticks for the MiniMax H3 text encoder. (@joshistoast, #10123)
  • Wildcard autocomplete stays open while you scroll. (@mvanhorn, #9999)

💜 Contributors

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

Core development team


@joshistoast
@lstein
@Pfannkuchensack
@JPPhoto
@dunkeroni
@DustyShoe

Major community contributors


@primorLee
@unmarco
@wunianze666-netizen
@guishiw

@kevin9327
@Totara-thib
@MohammadHijjawi97
@chiliec

Many more people contributed fixes and features during the alphas. They're credited throughout these notes and in each alpha's changelog. And an equally big thank-you to everyone who tested the alphas 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. The startup log names the backup when v7 upgrades your database. 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 new settings (for example db_synchronous, fp8_compute, int8_compute, attention_backend, tiny_autoencoder_previews, reserve_vram_gb, image_index_* and fonts_*). v6 won't start if any of these are in the file, so remove any you added.

  7. Start InvokeAI v6 (6.14.4). 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.

What's Changed Since Alpha.4

New Contributors

Full Changelog: v7.0.0-alpha.4...v7.0.0-beta.1

Contributors

lstein, Balrog994, and 18 other contributors
Assets 2
Loading