Skip to content

Releases: invoke-ai/InvokeAI

v4.2.0b1

03 May 03:47
@psychedelicious psychedelicious
v4.2.0b1
0787c6c
Compare
Choose a tag to compare
v4.2.0b1 Pre-release
Pre-release

This is a beta release. There may be some hiccups, but overally, it is purring along nicely.

Changes since v4.2.0a4

  • Initial image support in Control Layers, no more dedicated tab
  • Tabs renamed to Generation, Canvas, Workflows, Models and Queue
  • Refactored internal handling of control layers, which fixes all reported UI errors
  • T2I Adapter support in Control Layers
  • FF v125 bug fixed
  • Add TCD scheduler @l0stl0rd
  • Image viewer updates
  • Fixed Control Adapter processors' image size constraints
  • Metadata recall for Control Layers
  • Many small and not-particularly-memorable bugfixes

🧪 Control Layers

Control Layers give you control over specific areas of the image. Draw a mask and set a positive prompt, negative prompt, or any number of IP Adapters to be applied to the masked region. Control Adapters (ControlNet & T2I Adapters) and the Initial Image are visualized on the canvas.

Full documentation to be included with the full release.

Your feedback is greatly appreciated as we continue to iterate on Control Layers.

🐛 Known Issues

  • The viewer stays open after using Send to Image to Image, Send to Unified Canvas, and doing anything that adds a layer.

📈 Patch Nodes for v4.2.0

Enhancements

  • Control Layers
  • Add TCD scheduler @l0stl0rd
  • Image viewer updates

Fixes

  • Fixed inpainting models on canvas
  • IP Adapter starter models fixed
  • Fixed bug where temp files (tensors, conditioning) aren't cleaned up properly
  • FIxed trigger phrase form submit @joshistoast
  • Fixed SDXL Checkpoint Inpainting models
  • Fixed installing models on external SSDs on macOS
  • Fixed Control Adapter processors' image size constraints

💾 Installation and Updating

To install or update to v4.2.0b1, download the installer and follow the installation instructions.

To update, select the same installation location. Your user data will not be touched.

Missing models after updating from v3 to v4

See this FAQ.

What's Changed

New Contributors

Full Changelog: v4.2.0a4...v4.2.0b1

Contributors

psychedelicious, joshistoast, and l0stl0rd
Assets 5

v4.2.0a4

30 Apr 20:46
@psychedelicious psychedelicious
v4.2.0a4
abb3bb9
Compare
Choose a tag to compare
v4.2.0a4 Pre-release
Pre-release

This is an alpha release. We suggest backing up your database in case there are any issues and you need to roll back.

Changes since v4.2.0a3

  • "Regional Control" -> "Control Layers"
  • Control Adapters supported in Control Layers
  • Updates to Control Layers UI
  • Fixed SDXL Checkpoint Inpainting models
  • Fixed installing models on external SSDs on macOS

🧪 Control Layers

Control Layers give you control over specific areas of the image. Draw a mask and set a positive prompt, negative prompt, or any number of IP Adapters to be applied to the masked region. Control Adapters (ControlNet & T2I Adapters) are visualized on the canvas.

Full documentation to be included with the full release.

Your feedback is greatly appreciated as we continue to iterate on Control Layers.

🐛 Known Issues

These issues will be fixed for the full release.

  • If you undo after deleting a Regional Guidance layer's IP Adapter, Global IP Adapter layer, or Global Control Adapter Layer, the UI may error. You'll just need to click Reset if this happens.
  • There is a bug on Firefox v125 that causes very poor performance in the Control Layers canvas. This is not an issue for Firefox v126 Developer Edition. Other versions of Firefox have not been tested.
  • T2I Adapters are not supported in Control Layers.

📈 Patch Nodes for v4.2.0

Enhancements

  • Control Layers

Fixes

  • Fixed inpainting models on canvas
  • IP Adapter starter models fixed
  • Fixed bug where temp files (tensors, conditioning) aren't cleaned up properly
  • FIxed trigger phrase form submit @joshistoast
  • Fixed SDXL Checkpoint Inpainting models
  • Fixed installing models on external SSDs on macOS

💾 Installation and Updating

To install or update to v4.2.0a4, download the installer and follow the installation instructions.

To update, select the same installation location. Your user data will not be touched.

Missing models after updating from v3 to v4

See this FAQ.

What's Changed

Full Changelog: v4.2.0a3...v4.2.0a4

Contributors

dunkeroni, psychedelicious, and 2 other contributors
Assets 5

v4.2.0a3

25 Apr 03:28
@psychedelicious psychedelicious
v4.2.0a3
3207822
Compare
Choose a tag to compare
v4.2.0a3 Pre-release
Pre-release

This is an alpha release. We suggest backing up your database in case there are any issues and you need to roll back.

Changes since v4.2.0a2

  • Updates to the Regional Control UI
  • Fixed inpainting models on canvas
  • IP Adapter starter models fixed
  • Fixed bug where temp files (tensors, conditioning) aren't cleaned up properly
  • Fixed trigger phrase form submit @joshistoast

🧪 Regional Control

Regional Control (name may change) gives you control over specific areas of the image. Draw a mask and set a positive prompt, negative prompt, or any number of IP Adapters to be applied to the masked region.

To support this powerful feature, we are introducing a new canvas editor. Here's a brief demo:

My.Movie.12.mp4

Full documentation to be included with the full release.

Your feedback is greatly appreciated as we continue to iterate on Regional Control.

🐛 Known Issues

These issues will be fixed for the full release.

  • Occasionally, if you undo after deleting a layer's IP Adapter, the UI will error. You'll just need to click Reset if this happens.
  • Occasionally, the Regional Control canvas may not show and brush strokes. Refresh the page to fix.

📈 Patch Nodes for v4.2.0

Enhancements

  • Regional Control UI

Fixes

  • Fixed inpainting models on canvas
  • IP Adapter starter models fixed
  • Fixed bug where temp files (tensors, conditioning) aren't cleaned up properly
  • FIxed trigger phrase form submit @joshistoast

💾 Installation and Updating

To install or update to v4.2.0a3, download the installer and follow the installation instructions.

To update, select the same installation location. Your user data will not be touched.

Missing models after updating from v3 to v4

See this FAQ.

What's Changed

Full Changelog: v4.2.0a2...v4.2.0a3

Contributors

lstein, dunkeroni, and 3 other contributors
Assets 5

v4.2.0a2

23 Apr 03:25
@psychedelicious psychedelicious
v4.2.0a2
8b9c4c6
Compare
Choose a tag to compare
v4.2.0a2 Pre-release
Pre-release

This is an alpha release. We suggest backing up your database in case there are any issues and you need to roll back.

🧪 Regional Control

Regional Control (name may change) gives you control over specific areas of the image. Draw a mask and set a positive prompt, negative prompt, or any number of IP Adapters to be applied to the masked region.

To support this powerful feature, we are introducing a new canvas editor. Here's a brief demo:

My.Movie.12.mp4

Full documentation to be included with the full release.

Your feedback is greatly appreciated as we continue to iterate on Regional Control.

🐛 Known Issues

Regional Control

These issues will be fixed for the full release.

  • Occasionally, if you undo after deleting a layer's IP Adapter, the UI will error. You'll just need to click Reset if this happens.
  • Occasionally, the Regional Control canvas may not show and brush strokes. Refresh the page to fix.

Other

  • Inpainting models on Canvas sometimes kinda give up and output mush. Apologies for the delay in rolling this fix out.

📈 Patch Nodes for v4.2.0

Enhancements

  • Regional Control UI

💾 Installation and Updating

To install or update to v4.2.0a2, download the installer and follow the installation instructions.

To update, select the same installation location. Your user data will not be touched.

Missing models after updating from v3 to v4

See this FAQ.

What's Changed

Full Changelog: v4.1.0...v4.2.0a2

Contributors

lstein, psychedelicious, and blessedcoolant
Assets 5

v4.1.0

17 Apr 22:05
@psychedelicious psychedelicious
v4.1.0
ac1071a
Compare
Choose a tag to compare
v4.1.0 Latest
Latest

Invoke v4.1.0 brings a many fixes and enhancements. The big ticket is Style and Composition IP Adapter.

🧪 Style and Composition IP Adapter (beta)

IP Adapter uses an image as a prompt. Images have two major components - their style and their composition - and you can choose either or both when using IP Adapter.

Use the new IP Adapter Method dropdown to select Full, Style, or Composition. The setting is applied per IP Adapter. You may need to delete and re-add active IP Adapters to see the dropdown.

No IP Adapter IP Adapter Image Full IP Adapter Style Only Composition Only

"a fierce wolf in an alpine forest", all using same seed - note how the Full method turns the wolf into a mouse-canine hybrid

Shout-out to @blessedcoolant for this feature!

📈 Patch Nodes for v4.1.0

Enhancements

  • Backend and nodes implementation for regional prompting and regional IP Adapter (UI in v4.2.0)
  • Secret option in Workflow Editor to convert a graph into a workflow. See #6181 for how to use it.
  • Assortment of UI papercuts
  • Favicon & page title indicate generation status @jungleBadger
  • Delete hotkey and button work with gallery selection @jungleBadger
  • Workflow editor perf improvements
  • Edge labels in workflow editor
  • Updated translations @Harvester62, @symant233, @Vasyanator
  • Updated docs @sarashinai
  • Improved torch device and precision handling

Fixes

  • multipleOf for invocations (for example, the Noise invocation's width and height have a step of 8)
  • Poor quality "fried" refiner outputs
  • Poor quality inpainting with gradient denoising and refiner
  • Canvas images appearing in the wrong places
  • The little eye defaulting to off in canvas staging toolbar
  • Premature OOM on windows (see shared GPU memory FAQ)
  • ~1s delay between queue items
  • Wonky model manager forms navigating away from UI @clsn

Invocation API

  • New method to get the filesystem path of an image: context.images.get_path(image_name: str, thumbnail: bool) @fieldOfView

Internal

💾 Installation and Updating

To install or update to v4.1.0, download the installer and follow the installation instructions.

To update, select the same installation location. Your user data will not be touched.

Missing models after updating from v3 to v4

See this FAQ.

🐛 Known Issues

  • Inpainting models on Canvas sometimes kinda give up and output mush. The fix didn't make it in to v4.1.0, we will aim to release a patch in by the weekend.

What's Changed

New Contributors

Full Changelog: v4.0.4...v4.1.0

Contributors

clsn, lstein, and 15 other contributors
Assets 5

v4.0.4

06 Apr 08:02
@psychedelicious psychedelicious
v4.0.4
2c45697
Compare
Choose a tag to compare
v4.0.4

🚨 v4 has some major changes. Please read the patch notes. 🚨

Patch Nodes for v4.0.4

This patch release includes the following changes:

  • Add fit bounding box to image when sending image to canvas
  • Small handful of canvas bugs fixed
  • Refiner models displayed in model manager
  • Fix OOM on Windows (see this FAQ entry for more detail)
  • Restore initial image recall for img2img

💾 Installation and Updating

To install or update to v4.0.4, download the installer and follow the installation instructions. To update, select the same installation location.

🎉 What's New in Invoke v4

💻 Simplified Installation, Updating and Configuration

We've simplified and streamlined installation, making it much faster and more reliable:

💖 New Model Manager

The model manager is rewritten in v4.0.0, both frontend and backend. This builds a foundation for future model architectures and brings some exciting new user-facing features:

  • All model installation happens via the UI (no configure script)
  • Queued model downloads
  • Per-model preview images
  • Per-model default settings - choose a model’s default VAE, Scheduler, CFG Scale, etc
  • User-defined trigger phrases for concepts/LoRAs and models - access by typing the < key in any prompt box
  • API key support for model marketplaces
  • 🚨 Autoimport removed - use Scan Folder instead

#️⃣ Model Hashing

When you first run v4, it may take a few minutes to start up as it does a one-time hash of all of your model files.

Do not panic.

Hashes provide a stable identifier for a model that is the same across every platform.

🚨 If you don’t care about this, you can press Ctrl+C to interrupt the process and disable hashing by setting hashing_algorithm: random setting in invokeai.yaml.

🎨 Canvas Improvements

The canvas uses a new method for compositing called gradient denoising. This eliminates the need for multiple “passes”, greatly reducing generation time on the canvas. This method also provides substantially improved visual coherence between the masked regions and the rest of the image.

The compositing settings on canvas allow for control over the gradient denoising process.

Major research & experimentation for this novel denoising implementation was led by @dunkeroni, and @blessedcoolant was responsible for managing integration into the canvas UI.

🐛 Known Issue

🚨 Inpainting models on Canvas sometimes kinda give up and output mush. We have a fix en-route, but it will need to wait for 4.1.0.

📈 Fixes and Enhancements

4.0.4

  • Add fit bounding box to image when sending image to canvas
  • Small handful of canvas bugs fixed
  • Refiner models displayed in model manager
  • Fix OOM on Windows (see this FAQ entry for more detail)
  • Restore initial image recall for img2img

4.0.2

  • Fix errors related to character encodings during install and startup
  • UI error on first launch of v4, requiring reset of UI
  • Cancel batch button not working
  • Improvements to Scan Folder
  • FAQ to fix some models not migrating to v4
  • Removed unused or wonky GPU options in installer
  • Root dir detection via venv path
  • Handful of cosmetic UI fixes
  • IP Adapter safetensor support

4.0.1

  • Minor updates that resolve performance issues on the canvas.
  • Some installation/updating fixes to improve experience.

4.0.0

Many small bug fixes, resolved papercuts, and warm fuzzies. Shouting out just a few notable goodies from the community:

🎁 Bonus: Invoke Training (Beta)

As of v4.0.0, all references to training in the core invoke script now point to the Invoke Training Repo. Invoke Training offers a simple user interface for:

  • Textual Inversion Training
  • LoRA Training
  • Dreambooth Training
  • Pivotal Tuning Training

Learn more on the Invoke Training repo, as well as our YT video on getting started

Models don't show up after upgrading

Follow these steps. If you are still missing some models, please create an issue on GitHub or ask for help on discord.

🤓 Developer Changes

v4.0.0 is versioned as a major release due to breaking changes:

  • The internal nodes API has been refactored to provide a stable public API. 🚨 Node authors should review the migration guide.
  • The internal graph execution engine is drastically simplified, resulting in more efficient and performant processing. This carries on from the changes in v3.6.0 in which graphs are no longer stored in the database.

🤝 Contributing

As a community-supported project, we rely on volunteers and enthusiasts for continued innovation and polish. Everything from minor documentation fixes to major feature additions are welcome. To get started as a contributor, please refer to How to Contribute or reach out in #dev-chat on Discord!

What's Changed

New Contributors

Full Changelog: v4.0.2...v4.0.4

Contributors

clsn, lstein, and 20 other contributors
Assets 5

v4.0.2

04 Apr 05:05
@psychedelicious psychedelicious
v4.0.2
8a32baf
Compare
Choose a tag to compare
v4.0.2

🚨 v4 has some major changes. Please read the patch notes. 🚨

🚨 🚨 🚨 Yes - Those patch notes 👇 🚨 🚨 🚨

🎉 What's New in 4.0.2

This is a patch release includes these changes:

  • Fix errors related to character encodings during install and startup
  • UI error on first launch of v4, requiring reset of UI
  • Cancel batch button not working
  • Improvements to Scan Folder
  • FAQ to fix some models not migrating to v4
  • Removed unused or wonky GPU options in installer
  • Root dir detection via venv path
  • Handful of cosmetic UI fixes

It also includes one notable feature:

  • IP Adapter safetensor support

💻 Simplified Installation, Updating and Configuration

We've simplified and streamlined installation, making it much faster and more reliable:

💖 New Model Manager

The model manager is rewritten in v4.0.0, both frontend and backend. This builds a foundation for future model architectures and brings some exciting new user-facing features:

  • All model installation happens via the UI (no configure script)
  • Queued model downloads
  • Per-model preview images
  • Per-model default settings - choose a model’s default VAE, Scheduler, CFG Scale, etc
  • User-defined trigger phrases for concepts/LoRAs and models - access by typing the < key in any prompt box
  • API key support for model marketplaces
  • 🚨 Autoimport removed - use Scan Folder instead

#️⃣ Model Hashing

When you first run v4, it may take a few minutes to start up as it does a one-time hash of all of your model files.

Do not panic.

Hashes provide a stable identifier for a model that is the same across every platform.

🚨 If you don’t care about this, you can press Ctrl+C to interrupt the process and disable hashing by setting hashing_algorithm: random setting in invokeai.yaml.

🎨 Canvas Improvements

The canvas uses a new method for compositing called gradient denoising. This eliminates the need for multiple “passes”, greatly reducing generation time on the canvas. This method also provides substantially improved visual coherence between the masked regions and the rest of the image.

The compositing settings on canvas allow for control over the gradient denoising process.

Major research & experimentation for this novel denoising implementation was led by @dunkeroni, and @blessedcoolant was responsible for managing integration into the canvas UI.

🐛 Known Issue

🚨 Inpainting models on Canvas sometimes kinda give up and output mush. We have a fix en-route, but it will need to wait for 4.1.0.

📈 Fixes and Enhancements

4.0.2

Fixes
  • Fix errors related to character encodings during install and startup
  • UI error on first launch of v4, requiring reset of UI
  • Cancel batch button not working
  • Improvements to Scan Folder
  • FAQ to fix some models not migrating to v4
  • Removed unused or wonky GPU options in installer
  • Root dir detection via venv path
  • Handful of cosmetic UI fixes
Features
  • IP Adapter safetensor support

4.0.1

Fixes
  • Minor updates that resolve performance issues on the canvas.
  • Some installation/updating fixes to improve experience.

4.0.0

Many small bug fixes, resolved papercuts, and warm fuzzies. Shouting out just a few notable goodies from the community:

🎁 Bonus: Invoke Training (Beta)

As of v4.0.0, all references to training in the core invoke script now point to the Invoke Training Repo. Invoke Training offers a simple user interface for:

  • Textual Inversion Training
  • LoRA Training
  • Dreambooth Training
  • Pivotal Tuning Training

Learn more on the Invoke Training repo, as well as our YT video on getting started

💾 Installation and Upgrading

🚨 To install or upgrade to version 4.0, download the zip file from the release notes ("Assets" section), unzip it, and follow the installation instructions. For upgrades, select the same installation location.

Download Installer

Models don't show up after upgrading

Follow these steps. If you are still missing some models, please create an issue on GitHub or ask for help on discord.

🤓 Developer Changes

v4.0.0 is versioned as a major release due to breaking changes:

  • The internal nodes API has been refactored to provide a stable public API. 🚨 Node authors should review the migration guide.
  • The internal graph execution engine is drastically simplified, resulting in more efficient and performant processing. This carries on from the changes in v3.6.0 in which graphs are no longer stored in the database.

🤝 Contributing

As a community-supported project, we rely on volunteers and enthusiasts for continued innovation and polish. Everything from minor documentation fixes to major feature additions are welcome. To get started as a contributor, please refer to How to Contribute or reach out in #dev-chat on Discord!

What's Changed

Full Changelog: v4.0.1...v4.0.2

Contributors

clsn, GGSSKK, and 17 other contributors
Assets 5

v4.0.1

02 Apr 14:47
@hipsterusername hipsterusername
v4.0.1
21b9e96
Compare
Choose a tag to compare
v4.0.1

🚨 4.0.0 has some major changes. Please read the patch notes. 🚨

🚨 🚨 🚨 Yes - Those patch notes 👇 🚨 🚨 🚨

🎉 What's New in 4.0

💻 Simplified Installation, Updating and Configuration

We've simplified and streamlined installation, making it much faster and more reliable:

💖 New Model Manager

The model manager is rewritten in v4.0.0, both frontend and backend. This builds a foundation for future model architectures and brings some exciting new user-facing features:

  • All model installation happens via the UI (no configure script)
  • Queued model downloads
  • Per-model preview images
  • Per-model default settings - choose a model’s default VAE, Scheduler, CFG Scale, etc
  • User-defined trigger phrases for concepts/LoRAs and models - access by typing the < key in any prompt box
  • API key support for model marketplaces
  • 🚨 Autoimport removed - use Scan Folder instead

#️⃣ Model Hashing

When you first run v4.0.0, it may take a few minutes to start up as it does a one-time hash of all of your model files.

Do not panic.

Hashes provide a stable identifier for a model that is the same across every platform.

🚨 If you don’t care about this, you can press Ctrl+C to interrupt the process and disable hashing by setting hashing_algorithm: random setting in invokeai.yaml.

🎨 Canvas Improvements

The canvas uses a new method for compositing called gradient denoising. This eliminates the need for multiple “passes”, greatly reducing generation time on the canvas. This method also provides substantially improved visual coherence between the masked regions and the rest of the image.

The compositing settings on canvas allow for control over the gradient denoising process.

Major research & experimentation for this novel denoising implementation was led by @dunkeroni, and @blessedcoolant was responsible for managing integration into the canvas UI.

🐛 Known Issue

🚨 Inpainting models on Canvas sometimes kinda give up and output mush. We have a fix en-route, but it will need to wait for 4.1.0.

📈 Fixes and Enhancements

Many small bug fixes, resolved papercuts, and warm fuzzies. Shouting out just a few notable goodies from the community:

4.01 Fixes

  • Minor updates that resolve performance issues on the canvas.
  • Some installation/updating fixes to improve experience.

🎁 Bonus: Invoke Training (Beta)

As of v4.0.0, all references to training in the core invoke script now point to the Invoke Training Repo. Invoke Training offers a simple user interface for:

  • Textual Inversion Training
  • LoRA Training
  • Dreambooth Training
  • Pivotal Tuning Training

Learn more on the Invoke Training repo, as well as our YT video on getting started

💾 Installation and Upgrading

🚨 To install or upgrade to version 4.0, download the zip file from the release notes ("Assets" section), unzip it, and follow the installation instructions. For upgrades, select the same installation location.

Download Installer

Models don't show up after upgrading

Follow these steps. If you are still missing some models, please create an issue on GitHub or ask for help on discord.

🤓 Developer Changes

v4.0.0 is versioned as a major release due to breaking changes:

  • The internal nodes API has been refactored to provide a stable public API. 🚨 Node authors should review the migration guide.
  • The internal graph execution engine is drastically simplified, resulting in more efficient and performant processing. This carries on from the changes in v3.6.0 in which graphs are no longer stored in the database.

🤝 Contributing

As a community-supported project, we rely on volunteers and enthusiasts for continued innovation and polish. Everything from minor documentation fixes to major feature additions are welcome. To get started as a contributor, please refer to How to Contribute or reach out in #dev-chat on Discord!

What's Changed

Read more

Contributors

clsn, lstein, and 22 other contributors
Assets 3

v4.0.0

01 Apr 23:20
@psychedelicious psychedelicious
v4.0.0
a105da6
Compare
Choose a tag to compare
v4.0.0

🚨 4.0.0 has some major changes. Please read the patch notes. 🚨

🚨 🚨 🚨 Yes - Those patch notes 👇 🚨 🚨 🚨

🎉 What's New in 4.0.0

💻 Simplified Installation, Updating and Configuration

We've simplified and streamlined installation, making it much faster and more reliable:

💖 New Model Manager

The model manager is rewritten in v4.0.0, both frontend and backend. This builds a foundation for future model architectures and brings some exciting new user-facing features:

  • All model installation happens via the UI (no configure script)
  • Queued model downloads
  • Per-model preview images
  • Per-model default settings - choose a model’s default VAE, Scheduler, CFG Scale, etc
  • User-defined trigger phrases for concepts/LoRAs and models - access by typing the < key in any prompt box
  • API key support for model marketplaces
  • 🚨 Autoimport removed - use Scan Folder instead

#️⃣ Model Hashing

When you first run v4.0.0, it may take a few minutes to start up as it does a one-time hash of all of your model files.

Do not panic.

Hashes provide a stable identifier for a model that is the same across every platform.

🚨 If you don’t care about this, you can press Ctrl+C to interrupt the process and disable hashing by setting hashing_algorithm: random setting in invokeai.yaml.

🎨 Canvas Improvements

The canvas uses a new method for compositing called gradient denoising. This eliminates the need for multiple “passes”, greatly reducing generation time on the canvas. This method also provides substantially improved visual coherence between the masked regions and the rest of the image.

The compositing settings on canvas allow for control over the gradient denoising process.

Major research & experimentation for this novel denoising implementation was led by @dunkeroni, and @blessedcoolant was responsible for managing integration into the canvas UI.

🐛 Known Issue

🚨 Inpainting models on Canvas sometimes kinda give up and output mush. We have a fix en-route, but it will need to wait for 4.1.0.

📈 Fixes and Enhancements

Many small bug fixes, resolved papercuts, and warm fuzzies. Shouting out just a few notable goodies from the community:

🎁 Bonus: Invoke Training (Beta)

As of v4.0.0, all references to training in the core invoke script now point to the Invoke Training Repo. Invoke Training offers a simple user interface for:

  • Textual Inversion Training
  • LoRA Training
  • Dreambooth Training
  • Pivotal Tuning Training

Learn more on the Invoke Training repo, as well as our YT video on getting started

💾 Installation and Upgrading

🚨 To install or upgrade to version 4.0, download the zip file from the release notes ("Assets" section), unzip it, and follow the installation instructions. For upgrades, select the same installation location.

Download Installer

🤓 Developer Changes

v4.0.0 is versioned as a major release due to breaking changes:

  • The internal nodes API has been refactored to provide a stable public API. 🚨 Node authors should review the migration guide.
  • The internal graph execution engine is drastically simplified, resulting in more efficient and performant processing. This carries on from the changes in v3.6.0 in which graphs are no longer stored in the database.

🤝 Contributing

As a community-supported project, we rely on volunteers and enthusiasts for continued innovation and polish. Everything from minor documentation fixes to major feature additions are welcome. To get started as a contributor, please refer to How to Contribute or reach out in #dev-chat on Discord!

📝 What's Changed

Read more

Contributors

clsn, lstein, and 22 other contributors
Assets 3

v4.0.0rc6

28 Mar 14:39
@psychedelicious psychedelicious
v4.0.0rc6
53c19ae
Compare
Choose a tag to compare
v4.0.0rc6 Pre-release
Pre-release

This is a Release Candidate. We strongly suggest backing up your database before testing to prevent data loss in case of any issues.

Please let us know if you run into anything unexpected.

What's Changed in RC6 from RC5

🎉 Barring any major issues, this will be the last RC before v4.0.0! 🎉

  • Updater removed - use the installer to update
  • Installation docs rewritten
  • FAQ updated
  • Iterate nodes now iterate in order - thanks @cgi-joe!
  • Fix soft lock when model install fails
  • Autoimport removed - use Scan Folder instead
  • Sane workflow sorting - thanks @clsn!
  • Numerous minor fixes and enhancements

What's New in 4.0

Simplified Installation, Updating and Configuration

We've simplified and streamlined installation, making it much faster and reliable:

New Model Manager

The model manager is rewritten in v4.0.0, both frontend and backend. This builds a foundation for future model architectures and brings some exciting new user-facing features:

  • All model installation happens via the UI (no configure script)
  • Queued model downloads
  • Per-model preview images
  • Per-model default settings - choose a model’s default VAE, Scheduler, CFG Scale, etc
  • User-defined trigger phrases for concepts/LoRAs and models - access by typing the < key in any prompt box
  • API key support for model marketplaces
  • Autoimport removed - use Scan Folder instead

Model Hashing

When you first run v4.0.0, it may take a few minutes to start up as it does a one-time hash of all of your model files.

Do not panic.

Hashes provide a stable identifier for a model that is the same across every platform.

If you don’t care about this, you can press Ctrl+C to interrupt the process and disable hashing by setting hashing_algorithm: random setting in invokeai.yaml.

Canvas Improvements

The canvas uses a new method for compositing called gradient denoising. This eliminates the need for multiple “passes”, greatly reducing generation time on the canvas. This method also provides substantially improved visual coherence between the masked regions and the rest of the image.

The compositing settings on canvas allow for control over the gradient denoising process.

Major research & experimentation for this novel denoising implementation was led by @dunkeroni, and @blessedcoolant was responsible for managing integration into the canvas UI.

Fixes and Enhancements

Many small bug fixes, resolved papercuts, and warm fuzzies. Shouting out some notable goodies from the community:

Bonus: Invoke Training (Beta)

As of v4.0.0, all references to training in the core invoke script now point to the Invoke Training Repo. Invoke Training offers a simple user interface for:

  • Textual Inversion Training
  • LoRA Training
  • Dreambooth Training
  • Pivotal Tuning Training

Learn more on the Invoke Training repo.

Installation and Upgrading

To install or upgrade to version 4.0, download the zip file from the release notes ("Assets" section), unzip it, and follow the installation instructions. For upgrades, select the same installation location.

💾 Download Installer

💻 Developer Changes

v4.0.0 is versioned as a major release due to breaking changes:

  • The internal nodes API has been refactored to provide a stable public API. Node authors should review the migration guide.
  • The internal graph execution engine is drastically simplified, resulting in more efficient and performant processing. This carries on from the changes in v3.6.0 in which graphs are no longer stored in the database.

Contributing

As a community-supported project, we rely on volunteers and enthusiasts for continued innovation and polish. Everything from minor documentation fixes to major feature additions are welcome. To get started as a contributor, please refer to How to Contribute or reach out in #dev-chat on Discord!

What's Changed

New Contributors

Full Changelog: v4.0.0rc5...v4.0.0rc6

Contributors

clsn, lstein, and 17 other contributors
Assets 3