* resolved conflicts
* changed logo and some design changes
* feedback changes
* resolved conflicts
* changed logo and some design changes
* feedback changes
* lint fixed
* added translations
* some requested changes done
* all feedback changes done and replace links in settingsmenu comp
* fixed the gap between deps verisons & chnaged heights
* feat(ui): minor about modal styling
* feat(ui): tag app endpoints with FetchOnReconnect
* fix(ui): remove unused translation string
---------
Co-authored-by: psychedelicious <4822129+psychedelicious@users.noreply.github.com>
Co-authored-by: Kent Keirsey <31807370+hipsterusername@users.noreply.github.com>
Use each language's own language for their option in the language select. This falls back to the english translation if the language name isn't translated.
* add basic functionality for model metadata fetching from hf and civitai
* add storage
* start unit tests
* add unit tests and documentation
* add missing dependency for pytests
* remove redundant fetch; add modified/published dates; updated docs
* add code to select diffusers files based on the variant type
* implement Civitai installs
* make huggingface parallel downloading work
* add unit tests for model installation manager
- Fixed race condition on selection of download destination path
- Add fixtures common to several model_manager_2 unit tests
- Added dummy model files for testing diffusers and safetensors downloading/probing
- Refactored code for selecting proper variant from list of huggingface repo files
- Regrouped ordering of methods in model_install_default.py
* improve Civitai model downloading
- Provide a better error message when Civitai requires an access token (doesn't give a 403 forbidden, but redirects
to the HTML of an authorization page -- arrgh)
- Handle case of Civitai providing a primary download link plus additional links for VAEs, config files, etc
* add routes for retrieving metadata and tags
* code tidying and documentation
* fix ruff errors
* add file needed to maintain test root diretory in repo for unit tests
* fix self->cls in classmethod
* add pydantic plugin for mypy
* use TestSession instead of requests.Session to prevent any internet activity
improve logging
fix error message formatting
fix logging again
fix forward vs reverse slash issue in Windows install tests
* Several fixes of problems detected during PR review:
- Implement cancel_model_install_job and get_model_install_job routes
to allow for better control of model download and install.
- Fix thread deadlock that occurred after cancelling an install.
- Remove unneeded pytest_plugins section from tests/conftest.py
- Remove unused _in_terminal_state() from model_install_default.
- Remove outdated documentation from several spots.
- Add workaround for Civitai API results which don't return correct
URL for the default model.
* fix docs and tests to match get_job_by_source() rather than get_job()
* Update invokeai/backend/model_manager/metadata/fetch/huggingface.py
Co-authored-by: Ryan Dick <ryanjdick3@gmail.com>
* Call CivitaiMetadata.model_validate_json() directly
Co-authored-by: Ryan Dick <ryanjdick3@gmail.com>
* Second round of revisions suggested by @ryanjdick:
- Fix type mismatch in `list_all_metadata()` route.
- Do not have a default value for the model install job id
- Remove static class variable declarations from non Pydantic classes
- Change `id` field to `model_id` for the sqlite3 `model_tags` table.
- Changed AFTER DELETE triggers to ON DELETE CASCADE for the metadata and tags tables.
- Made the `id` field of the `model_metadata` table into a primary key to achieve uniqueness.
* Code cleanup suggested in PR review:
- Narrowed the declaration of the `parts` attribute of the download progress event
- Removed auto-conversion of str to Url in Url-containing sources
- Fixed handling of `InvalidModelConfigException`
- Made unknown sources raise `NotImplementedError` rather than `Exception`
- Improved status reporting on cached HuggingFace access tokens
* Multiple fixes:
- `job.total_size` returns a valid size for locally installed models
- new route `list_models` returns a paged summary of model, name,
description, tags and other essential info
- fix a few type errors
* consolidated all invokeai root pytest fixtures into a single location
* Update invokeai/backend/model_manager/metadata/metadata_store.py
Co-authored-by: psychedelicious <4822129+psychedelicious@users.noreply.github.com>
* Small tweaks in response to review comments:
- Remove flake8 configuration from pyproject.toml
- Use `id` rather than `modelId` for huggingface `ModelInfo` object
- Use `last_modified` rather than `LastModified` for huggingface `ModelInfo` object
- Add `sha256` field to file metadata downloaded from huggingface
- Add `Invoker` argument to the model installer `start()` and `stop()` routines
(but made it optional in order to facilitate use of the service outside the API)
- Removed redundant `PRAGMA foreign_keys` from metadata store initialization code.
* Additional tweaks and minor bug fixes
- Fix calculation of aggregate diffusers model size to only count the
size of files, not files + directories (which gives different unit test
results on different filesystems).
- Refactor _get_metadata() and _get_download_urls() to have distinct code paths
for Civitai, HuggingFace and URL sources.
- Forward the `inplace` flag from the source to the job and added unit test for this.
- Attach cached model metadata to the job rather than to the model install service.
* fix unit test that was breaking on windows due to CR/LF changing size of test json files
* fix ruff formatting
* a few last minor fixes before merging:
- Turn job `error` and `error_type` into properties derived from the exception.
- Add TODO comment about the reason for handling temporary directory destruction
manually rather than using tempfile.tmpdir().
* add unit tests for reporting HTTP download errors
---------
Co-authored-by: Lincoln Stein <lstein@gmail.com>
Co-authored-by: Ryan Dick <ryanjdick3@gmail.com>
Co-authored-by: psychedelicious <4822129+psychedelicious@users.noreply.github.com>
Per user feedback, this is preferrable to letting them expand when the window grows.
Also bumps `react-resizable-panels` now that one of my PRs is merged to fix an issue.
* feat: allow bfloat16 to be configurable in invoke.yaml
* fix: `torch_dtype()` util
- Use `choose_precision` to get the precision string
- Do not reference deprecated `config.full_precision` flat (why does this still exist?), if a user had this enabled it would override their actual precision setting and potentially cause a lot of confusion.
---------
Co-authored-by: psychedelicious <4822129+psychedelicious@users.noreply.github.com>
- Add various brand images, organise images
- Create favicon for docs pages (light blue version of key logo)
- Rename app title to `Invoke - Community Edition`
Add `FetchOnReconnect` tag, tagging relevant queries with it. This tag is invalidated in the socketConnected listener, when it is determined that the queue changed.
- Add checks to the "recovery" logic for socket connect events to reduce the number of network requests.
- Remove the `isInitialized` state from `systemSlice` and make it a nanostore local to the socketConnected listener. It didn't need to be global state. It's also now more clearly named `isFirstConnection`.
- Export the queue status selector (minor improvement, memoizes it correctly).
- Fixed a bug where after you load more, changing boards doesn't work. The offset and limit for the list image query had some wonky logic, now resolved.
- Addressed major lag in gallery when selecting an image.
Both issues were related to the useMultiselect and useGalleryImages hooks, which caused every image in the gallery to re-render on whenever the selection changed. There's no way to memoize away this - we need to know when the selection changes. This is a longstanding issue.
The selection is only used in a callback, though - the onClick handler for an image to select it (or add it to the existing selection). We don't really need the reactivity for a callback, so we don't need to listen for changes to the selection.
The logic to handle multiple selection is moved to a new `galleryImageClicked` listener, which does all the selection right when it is needed.
The result is that gallery images no long need to do heavy re-renders on any selection change.
Besides the multiselect click handler, there was also inefficient use of DND payloads. Previously, the `IMAGE_DTOS` type had a payload of image DTO objects. This was only used to drag gallery selection into a board. There is no need to hold onto image DTOs when we have the selection state already in redux. We were recalculating this payload for every image, on every tick.
This payload is now just the board id (the only piece of information we need for this particular DND event).
- I also removed some unused DND types while making this change.
There was a lot of convoluted, janky logic related to trying to not mount the context menu's portal until its needed. This was in the library where the component was originally copied from.
I've removed that and resolved the jank, at the cost of there being an extra portal for each instance of the context menu. Don't think this is going to be an issue. If it is, the whole context menu could be refactored to be a singleton.
* ci: add docker build timout; log free space on runner before and after build
* docker: bump frontend builder to node=20.x; skip linting on build
* chore: gitignore .pnpm-store
* update code owners for docker and CI
---------
Co-authored-by: Millun Atluri <Millu@users.noreply.github.com>
I was troubleshooting a hotkeys issue on canvas and thought I had broken the tool logic in a past change so I redid it moving it to nanostores. In the end, the issue was an upstream but with the hotkeys library, but I like having tool in nanostores so I'm leaving it.
It's ephemeral interaction state anyways, doesn't need to be in redux.
There's a challenge to accomplish this due to our slice structure - the model is stored in `generationSlice`, but `canvasSlice` also needs to have awareness of it. For example, when the model changes, the canvas slice doesn't know what the previous model was, so it doesn't know whether or not to optimize the size.
This means we need to lift the "should we optimize size" information up. To do this, the `modelChanged` action creator accepts the previous model as an optional second arg.
Now the canvas has access to both the previous model and new model selection, and can decide whether or not it should optimize its size setting in the same way that the generation slice does.
Closes #5452
For some reason `ReturnType<typeof useListImagesQuery>` isn't working correctly, and destructuring `queryResult` it results in `any`, when the hook is used.
I've removed the explicit return typing so that consumers of the hook get correct types.
Workflow building would fail when a current image node was in the workflow due to the strict validation.
So we need to use the other workflow builder util first, which strips out extraneous data.
This bug was introduced during an attempt to optimize the workflow building logic, which was causing slowdowns on the workflow editor.
* do not show toast if 403 is triggered by lack of image access
* remove log
* lint
---------
Co-authored-by: Mary Hipp <maryhipp@Marys-MacBook-Air.local>
* feat(ui): get rid of convoluted socket vs appSocket redux actions
There's no need to have `socket...` and `appSocket...` actions.
I did this initially due to a misunderstanding about the sequence of handling from middleware to reducers.
* feat(ui): bump deps
Mainly bumping to get latest `redux-remember`.
A change to socket.io required a change to the types in `useSocketIO`.
* chore(ui): format
* feat(ui): add error handling to redux persistence layer
- Add an error handler to `redux-remember` config using our logger
- Add custom errors representing storage set and get failures
- Update storage driver to raise these accordingly
- wrap method to clear idbkeyval storage and tidy its logic up
* feat(ui): add debuggingLoggerMiddleware
This simply logs every action and a diff of the state change.
Due to the noise this creates, it's not added by default at all. Add it to the middlewares if you want to use it.
* feat(ui): add $socket to window if in dev mode
* fix(ui): do not enable cancel hotkeys on inputs
* fix(ui): use JSON.stringify for ROARR logger serializer
A recent change to ROARR introduced limits to the size of data that will logged. This ends up making our logs far less useful. Change the serializer back to what it was previously.
* feat(ui): change diff util, update debuggerLoggerMiddleware
The previous diff library would present deleted things as `undefined`. Unfortunately, a JSON.stringify cycle will strip those values out. The ROARR logger does this and so the diffs end up being a lot less useful, not showing removed keys.
The new diff library uses a different format for the delta that serializes nicely.
* feat(ui): add migrations to redux persistence layer
- All persisted slices must now have a slice config, consisting of their initial state and a migrate callback. The migrate callback is very simple for now, with no type safety. It adds missing properties to the state. A future enhancement might be to model the each slice's state with e.g. zod and have proper validation and types.
- Persisted slices now have a `_version` property
- The migrate callback is called inside `redux-remember`'s `unserialize` handler. I couldn't figure out a good way to put this into the reducer and do logging (reducers should have no side effects). Also I ran into a weird race condition that I couldn't figure out. And finally, the typings are tricky. This works for now.
- `generationSlice` and `canvasSlice` both need migrations for the new aspect ratio setup, this has been added
- Stuff related to persistence has been moved in to `store.ts` for simplicity
* feat(ui): clean up StorageError class
* fix(ui): scale method default is now 'auto'
* feat(ui): when changing controlnet model, enable autoconfig
* fix(ui): make embedding popover immediately accessible
Prevents hotkeys from being captured when embeddings are still loading.
The new select component appears to close itself before calling the onchange handler. This short-circuits the autoconnect logic. Tweaked so the ordering is correct.
Centralize the initial/min/max/etc values for all numerical params. We used this for some but at some point stopped updating it.
All numerical params now use their respective configs. Far fewer hardcoded values throughout the app now.
Also updated the config types a bit to better accommodate slider vs number input constraints.
- Use the virtuoso grid item container and list containers to calculate imagesPerRow, skipping manual compensation for padding of images
- Round the imagesPerRow instead of flooring - we often will end up with values like 4.99999 due to floating point precision
- Update `getDownImage` comments & logic to be clearer
- Use variables for the ids in query selectors, preventing future typos
- Only scroll if the new selected image is different from the prev one
- Fix preexisting bug where gallery network requests were duplicated when triggering infinite scroll
- Refactor `useNextPrevImage` to not use `state => state` as an input selector - logic split up into different hooks
- Remove use instant scroll for arrow key navigation - smooth scroll is janky when you hold the arrow down and it fires rapidly
- Move gallery nav hotkeys to GalleryImageGrid component, so they work whenever the gallery is open (previously didn't work on canvas or workflow editor tabs)
- Use nanostores for gallery grid refs instead of passing context with virtuoso's context feature, making it much simpler to do the imperative gallery nav
- General gallery hook/component cleanup
Pending resolution of https://github.com/reduxjs/reselect/issues/635, we can patch `reselect` to use `lruMemoize` exclusively.
Pin RTK and react-redux versions too just to be safe.
This reduces the major GC events that were causing lag/stutters in the app, particularly in canvas and workflow editor.
A bug that caused panels to be collapsed on a fresh indexedDb in was fixed in dd32c632cd, but this re-introduced a different bug that caused the panels to expand on window resize, if they were already collapsed.
Revert the previous change and instead add one imperative resize outside the observer, so that on startup, we set both panels to their minimum sizes.
* replace custom header with custom nav component to go below settings
* add option for custom gallery header
* add option for custom app info text on logo hover
* add data-testid for tabs
* remove descriptions
* lint
* lint
---------
Co-authored-by: Mary Hipp <maryhipp@Marys-MacBook-Air.local>
We are now using the lefthand vertical strip for the settings menu button. This is a good place for the status indicator.
Really, we only need to display something *if there is a problem*. If the app is processing, the progress bar indicates that.
For the case where the panels are collapsed, I'll add the floating buttons back in some form, and we'll indicate via those if the app is processing something.
just make it like a normal button - normal and hover state, no difference when its expanded. the icon clearly indicates this, and you see the extra components
On one hand I like the color but on the other it makes this divider a focus point, which doesn't really makes sense to me. I tried several shades but think it adds a bit too much distraction for your eyes.
There was an extra div, needed for the fullscreen file upload dropzone, that made styling the main app containers a bit awkward.
Refactor the uploader a bit to simplify this - no longer need so many app-level wrappers. Much cleaner.
Removed logic related to aspect ratio from the components.
When the main bbox changes, if the scale method is auto, the reducers will handle the scaled bbox size appropriately.
Somehow linking up the manual mode to the aspect ratio is tricky, and instead of adding complexity for a rarely-used mode, I'm leaving manual mode as fully manual.
Cannot figure out how to allow the bbox to be transformed when aspect ratio is locked from all handles. Only the bottom right handle works as expected.
As a workaround, when the aspect ratio is locked, you can only resize the bbox from the bottom right handle.
- Do not _merge_ prompt and style prompt when concat is enabled - either use the prompt as style, or use the style directly.
- Set style prompt metadata correctly.
- Add metadata recall for style prompt.
`react-select` has some weird behaviour where if the value is `undefined`, it shows the last-selected value instead of nothing. Must fall back to `null`
Ensure workflow editor model selector component gets a value
This introduced some funky type issues related to ONNX models. ONNX doesn't work anyways (unmaintained). Instead of fixing the types to work with a non-working feature, ONNX is now removed entirely from the UI.
- Remove all refs to ONNX (and Olives)
- Fix some type issues
- Add ONNX nodes to the nodes denylist (so they are not visible in UI)
- Update VAE graph helper, which still had some ONNX logic. It's a very simple change and doesn't change any logic. Just removes some conditions that were for ONNX. I tested it and nothing broke.
- Regenerate types
- Fix prettier and eslint ignores for generated types
- Lint
* Udpater suggest db backup when installing RC
* Update invokeai_update.py to be more specific
* Update invokeai_update.py
* Update invokeai_update.py
* Update invokeai_update.py
* Update invokeai_update.py
- Prompt must have an open curly brace followed by a close curly brace to enable dynamic prompts processing
- If a the given prompt already had a dynamic prompt cached, do not re-process
- If processing is not needed, user may invoke immediately
- Invoke button shows loading state when dynamic prompts are processing, tooltip says generating
- Dynamic prompts preview icon in prompt box shows loading state when processing, tooltip says generating
- Support grid size of 8 on canvas
- Internal canvas math works on 8
- Update gridlines rendering to show 64 spaced lines and 32/16/8 when zoomed in
- Bbox manipulation defaults to grid of 64 - hold shift to get grid of 8
Besides being something we support internally, supporting 8 on canvas avoids a lot of hacky logic needed to work well with aspect ratios.
Canvas and non-canvas have separate width and height and need their own separate aspect ratios. In order to not duplicate a lot of aspect ratio logic, the components relating to image size have been modularized.
- Fix `weight` and `begin_step_percent`, the constraints were mixed up
- Add model validatort to ensure `begin_step_percent < end_step_percent`
- Bump version
- Store workflow in nanostore as singleton instead of building for each consumer
- Debounce the build (already was indirectly debounced)
- When the workflow is needed, imperatively grab it from the nanostores, instead of letting react handle it via reactivity
This drastically reduces the computation needed when moving the cursor. It also correctly separates ephemeral interaction state from redux, where it is not needed.
Also removed some unused canvas state.
This uses the previous implementation of the memoization function in reselect. It's possible for the new weakmap-based memoization to cause memory leaks in certain scenarios, so we will avoid it for now.
If the user specifies `torch-sdp` as the attention type in `config.yaml`, we can go ahead and use it (if available) rather than always throwing an exception.
* add base definition of download manager
* basic functionality working
* add unit tests for download queue
* add documentation and FastAPI route
* fix docs
* add missing test dependency; fix import ordering
* fix file path length checking on windows
* fix ruff check error
* move release() into the __del__ method
* disable testing of stderr messages due to issues with pytest capsys fixture
* fix unsorted imports
* harmonized implementation of start() and stop() calls in download and & install modules
* Update invokeai/app/services/download/download_base.py
Co-authored-by: Ryan Dick <ryanjdick3@gmail.com>
* replace test datadir fixture with tmp_path
* replace DownloadJobBase->DownloadJob in download manager documentation
* make source and dest arguments to download_queue.download() an AnyHttpURL and Path respectively
* fix pydantic typecheck errors in the download unit test
* ruff formatting
* add "job cancelled" as an event rather than an exception
* fix ruff errors
* Update invokeai/app/services/download/download_default.py
Co-authored-by: psychedelicious <4822129+psychedelicious@users.noreply.github.com>
* use threading.Event to stop service worker threads; handle unfinished job edge cases
* remove dangling STOP job definition
* fix ruff complaint
* fix ruff check again
* avoid race condition when start() and stop() are called simultaneously from different threads
* avoid race condition in stop() when a job becomes active while shutting down
---------
Co-authored-by: Lincoln Stein <lstein@gmail.com>
Co-authored-by: Ryan Dick <ryanjdick3@gmail.com>
Co-authored-by: psychedelicious <4822129+psychedelicious@users.noreply.github.com>
Co-authored-by: Kent Keirsey <31807370+hipsterusername@users.noreply.github.com>
The graph library occasionally causes issues when the default graph changes substantially between versions and pydantic validation fails. See #5289 for an example.
We are not currently using the graph library, so we can disable it until we are ready to use it. It's possible that the workflow library will supersede it anyways.
* remove MacOS Sonoma check in devices.py
As of pytorch 2.1.0, float16 works with our MPS fixes on Sonoma, so the check is no longer needed.
* remove unused platform import
* add code to repopulate model config records after schema update
* reformat for ruff
* migrate model records using db cursor rather than the ModelRecordConfigService
* ruff fixes
* tweak exception reporting
* fix: build frontend in pypi-release workflow
This was missing, resulting in the 3.5.0rc1 having no frontend.
* fix: use node 18, set working directory
- Node 20 has a problem with `pnpm`; set it to Node 18
- Set the working directory for the frontend commands
* Don't copy extraneous paths into installer .zip
* feat(installer): delete frontend build after creating installer
This prevents an empty `dist/` from breaking the app on startup.
* feat: add python dist as release artifact, as input to enable publish to pypi
- The release workflow never runs automatically. It must be manually kicked off.
- The release workflow has an input. When running it from the GH actions UI, you will see a "Publish build on PyPi" prompt. If this value is "true", the workflow will upload the build to PyPi, releasing it. If this is anything else (e.g. "false", the default), the workflow will build but not upload to PyPi.
- The `dist/` folder (where the python package is built) is uploaded as a workflow artifact as a zip file. This can be downloaded and inspected. This allows "dry" runs of the workflow.
- The workflow job and some steps have been renamed to clarify what they do
* translationBot(ui): update translation files
Updated by "Cleanup translation files" hook in Weblate.
Co-authored-by: Hosted Weblate <hosted@weblate.org>
Translate-URL: https://hosted.weblate.org/projects/invokeai/web-ui/
Translation: InvokeAI/Web UI
* freeze yaml migration logic at upgrade to 3.5
* moved migration code to migration_3
---------
Co-authored-by: Lincoln Stein <lstein@gmail.com>
Co-authored-by: psychedelicious <4822129+psychedelicious@users.noreply.github.com>
Co-authored-by: Hosted Weblate <hosted@weblate.org>
The VAE decode on linear graphs was getting cached. This caused some unexpected behaviour around image outputs.
For example, say you ran the exact same graph twice. The first time, you get an image written to disk and added to gallery. The second time, the VAE decode is cached and no image file is created. But, the UI still gets the graph complete event and selects the first image in the gallery. The second run does not add an image to the gallery.
There are probbably edge cases related to this - the UI does not expect this to happen. I'm not sure how to handle it any better in the UI.
The solution is to not cache VAE decode on the linear graphs, ever. If you run a graph twice in linear, you expect two images.
This simple change disables the node cache for terminal VAE decode nodes in all linear graphs, ensuring you always get images. If they graph was fully cached, all images after the first will be created very quickly of course.
- "Reset Workflow Editor" -> "New Workflow"
- "New Workflow" gets nodes icon & is no longer danger coloured
- When creating a new workflow, if the current workflow has unsaved changes, you get a dialog asking for confirmation. If the current workflow is saved, it immediately creates a new workflow.
- "Download Workflow" -> "Save to File"
- "Upload Workflow" -> "Load from File"
- Moved "Load from File" up 1 in the menu
This model was a bit too strict, and raised validation errors when workflows we expect to *not* have an ID (eg, an embedded workflow) have one.
Now it strips unknown attributes, allowing those workflows to load.
- Handle an image file not existing despite being in the database.
- Add a simple pydantic model that tests only for the existence of a workflow's version.
- Check against this new model when migrating workflows, skipping if the workflow fails validation. If it succeeds, the frontend should be able to handle the workflow.
Currently translated at 98.1% (1340 of 1365 strings)
translationBot(ui): update translation (Russian)
Currently translated at 84.2% (1150 of 1365 strings)
translationBot(ui): update translation (Russian)
Currently translated at 83.1% (1135 of 1365 strings)
Co-authored-by: Васянатор <ilabulanov339@gmail.com>
Translate-URL: https://hosted.weblate.org/projects/invokeai/web-ui/ru/
Translation: InvokeAI/Web UI
- use simpler pattern for migration dependencies
- move SqliteDatabase & migration to utility method `init_db`, use this in both the app and tests, ensuring the same db schema is used in both
This fixes a problem with `Annotated` which prevented us from using pydantic's `Field` to specify a discriminator for a union. We had to use FastAPI's `Body` as a workaround.
