Previously the color of a link would simply use the type of the target
slot and fallback to the type of the origin slot. When a connection is
made to a node that accepts the any type ('*'), the link has the green
color of an unknown type.
Instead, when a connection is made, the type of a link is now calculated
as the greatest common type of the source and destination. This means
that connections to reroutes are correctly colored.
| Before | After |
| ------ | ----- |
| <img width="360" alt="before"
src="https://github.com/user-attachments/assets/a5544730-e69a-4c85-af33-b303bb30ae71"
/>| <img width="360" alt="after"
src="https://github.com/user-attachments/assets/7d7b59fd-1b79-440b-a97d-a1657313c484"
/>|
The code for calculating common types already exists, it has simply been
moved into litegraph and given a more descriptive name.
Resolves#7196
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-7211-Color-links-as-common-type-2c16d73d365081188460f6b5973db962)
by [Unito](https://www.unito.io)
---------
Co-authored-by: Alexander Brown <drjkl@comfy.org>
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
## Summary
Adds an interactive video tutorial dialog to help users find CivitAI
model URLs during the Upload Model wizard.
## Changes
- **New Component**: Created reusable `VideoHelpDialog.vue` component
- Full-width video player with floating close button
- Configurable props: `videoUrl`, `loop`, `showControls`
- Custom ESC key handling to prevent parent dialog from closing
- Click backdrop to dismiss
- 70% dark backdrop for better video focus
- **Upload Model Flow**: Integrated video help button in step 1 footer
- "How do I find this?" button opens tutorial video
- Video demonstrates finding model URLs on CivitAI
- PostHog tracking attribute maintained (`upload-model-step1-help-link`)
## Review Focus
- ESC key event handling uses capture phase to prevent propagation to
parent dialogs
- Component follows existing patterns from `MediaVideoTop.vue` and
`BaseModalLayout.vue`
- Video player accessibility (ARIA labels, keyboard navigation)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-7177-feat-Add-video-help-dialog-to-Upload-Model-flow-2c06d73d36508148963ad9ee60038ea3)
by [Unito](https://www.unito.io)
---------
Co-authored-by: Claude <noreply@anthropic.com>
## Summary
Fixes an issue where Playwright HTML reports were not being generated
for `chromium-2x`, `chromium-0.5x`, and `mobile-chrome` test runs,
causing 404 errors when accessing report links in PR comments.
## Problem
Only the first report link (chromium) had contents - the other three
browsers returned 404 errors. Investigation revealed that artifacts for
non-chromium browsers were only uploading 1 file (report.json) instead
of the full HTML report with ~32 files.
The root cause was that these browsers run very few tests (1-6 tests
each), and when using `--reporter=html --reporter=json` together,
Playwright would only generate the JSON report without the HTML assets.
## Solution
Changed the workflow for non-chromium browsers to use the same approach
as the sharded chromium tests:
1. Run tests with `--reporter=blob`
2. Generate HTML and JSON reports separately using `merge-reports`
This ensures both HTML and JSON reports are always generated with
complete assets, regardless of test count.
## Testing
The fix can be verified by:
1. Checking that this PR's workflow run uploads similar file counts for
all browsers
2. Confirming all 4 report links are accessible and show proper HTML
content
Related to #7186🤖 Generated with [Claude Code](https://claude.com/claude-code)
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-7193-bugfix-Fix-E2E-test-report-generation-for-non-chromium-browsers-2c06d73d365081ba8ea3ed0d3f5d8d38)
by [Unito](https://www.unito.io)
Co-authored-by: Claude <noreply@anthropic.com>
## Summary
Adds step-specific `data-attr` attributes throughout the Upload Model
wizard to enable PostHog analytics tracking and action creation,
following PostHog's recommended best practices for element labeling.
## Changes
- **What**: Added `data-attr` attributes to all key interactive elements
in the Upload Model flow (buttons, inputs, selectors)
- **Step-based naming**: Attributes include step numbers for funnel
analysis (e.g., `upload-model-step1-continue-button`)
- **Coverage**: Entry button, URL input, help link, navigation buttons
(cancel/back/continue/confirm/finish), and model type selector
## Benefits
- Enables creation of stable PostHog actions using CSS selectors like
`[data-attr="upload-model-step2-confirm-button"]`
- Allows funnel analysis to track user progression through the 3-step
upload wizard
- Identifies drop-off points and help-seeking behavior
- Follows PostHog best practice of using data attributes over CSS
classes (especially important with Tailwind)
## Review Focus
- Naming consistency and clarity of data-attr values
- Completeness of coverage across the upload flow
🤖 Generated with [Claude Code](https://claude.com/claude-code)
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-7173-feat-Add-PostHog-data-attr-attributes-to-Upload-Model-flow-2c06d73d365081699861d3d910250e32)
by [Unito](https://www.unito.io)
Co-authored-by: Claude <noreply@anthropic.com>
Previously, right-clicking on a Vue node would deselect all other
selected nodes because the pointerup event handler was calling
toggleNodeSelectionAfterPointerUp regardless of which mouse button was
released.
This fix skips selection handling when the right mouse button (button 2)
is released, allowing the context menu to operate on the existing
selection.
## Summary
- Fixes right-click deselecting all selected nodes when using Vue node
rendering
- Now right-clicking preserves the existing selection, allowing context
menu
actions on multiple nodes
## Problem
When multiple nodes were selected and user right-clicked on one of them,
the
`pointerup` event handler would call
`toggleNodeSelectionAfterPointerUp`, which
deselected everything except the clicked node. This broke multi-node
context menu
operations.
## Solution
Skip selection handling in `onPointerup` when `event.button === 2`
(right-click).
The context menu handler manages selection independently
fix https://github.com/Comfy-Org/ComfyUI_frontend/issues/7136
Before
https://github.com/user-attachments/assets/23ac5e03-c464-44b7-8950-67c14da9e02b
After
https://github.com/user-attachments/assets/9d1bd6a8-6386-442b-9dc4-6bc8fbe4a0a8
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-7162-fix-preserve-node-selection-on-right-click-2bf6d73d365081acaf75f2fc845bbffb)
by [Unito](https://www.unito.io)
---------
Co-authored-by: GitHub Action <action@github.com>
## Summary
- Allow default browser copy (Ctrl+C / Cmd+C) when text is selected
anywhere in the document
- Previously, the graph node copy handler intercepted copy events even
in dialogs
## Problem
Users could not copy error messages from the PromptExecutionError dialog
or other modal dialogs. When pressing Ctrl+C with text selected in a
dialog, the graph copy handler would intercept the event and prevent the
default browser copy behavior.
## Solution
Add a `hasTextSelection()` check to `shouldIgnoreCopyPaste()`. When the
user has any text selected in the document, the function returns `true`,
allowing the default browser copy to proceed.
## Test plan
- [ ] Open an error dialog (trigger a workflow error)
- [ ] Click "Show Report" to expand error details
- [ ] Select some text in the dialog
- [ ] Press Ctrl+C (or Cmd+C on Mac)
- [ ] Paste elsewhere to verify the text was copied
- [ ] Verify graph node copy still works when no text is selected
https://github.com/user-attachments/assets/30a0c501-95ee-4148-b321-3d60339a41c5
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-7166-Fix-copy-not-working-when-text-is-selected-in-dialogs-2bf6d73d36508182a240fd3153cb6969)
by [Unito](https://www.unito.io)
---------
Co-authored-by: GitHub Action <action@github.com>
The output type of a matchType output is initialized to
COMFY_MATCHTYPE_V3, but is updated soon after to the value calculated
from input types. Under some difficult to reproduce circumstances, this
output type may be incorrectly evaluated in connections to other switch
nodes. Since the initial type is never a valid connection, this can
produce errors.
Instead, the output type of a matchtype node is initialized to allow
connections to anything to ensure that the subsequent restriction of
output types is guaranteed to be directional
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-7161-On-adding-output-matchType-initialize-type-2bf6d73d3650819ab169ffe9a4ecfeb4)
by [Unito](https://www.unito.io)
---------
Co-authored-by: Alexander Brown <drjkl@comfy.org>
…s metadata
- Reorder handleFile() to check workflow before parameters
- Add validation to prevent JSON parse errors from crashing imports
- Fix loadGraphData() to use explicit type validation instead of falsy
check
- Ensures ComfyUI-generated PNGs with both metadata types load the
workflow, not parameters
Fixes issue where large workflows (e.g., 634 nodes) were replaced with
basic A1111 format when importing PNG files.
## Summary
Fixed workflow loading from PNG images to prioritize workflow metadata
over parameters, preventing large workflows from being replaced with
basic A1111 format.
## Changes
- **What**: Reordered `handleFile()` to check workflow before
parameters, added JSON parse error handling and validation, fixed
`loadGraphData()` to use explicit type checking instead of falsy check
- **Dependencies**: None
## Review Focus
The key issue was in `handleFile()` where parameters were checked before
workflow, causing ComfyUI-generated PNGs (which contain both workflow
and parameters metadata) to incorrectly import as A1111 format. The fix
ensures:
1. Workflow is always checked first and validated properly
2. Parameters are only used as a fallback when no workflow exists
3. Invalid/malformed workflow data doesn't crash the import process
Additionally, `loadGraphData()` was using a falsy check (`if
(!graphData)`) which could incorrectly replace valid but falsy values.
Now uses explicit type validation.
Tested with real-world PNG containing 634-node workflow (780KB) +
parameters (1KB) - now correctly loads the workflow instead of
discarding it.
<!-- Fixes #ISSUE_NUMBER -->
## Screenshots (if applicable)
N/A - Backend logic fix, no UI changes
Fixes#6633
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-7154-Fix-workflow-loading-from-PNG-images-with-both-workflow-and-parameter-2bf6d73d365081ecb7a6c4bf6b6ccd51)
by [Unito](https://www.unito.io)
---------
Co-authored-by: Alexander Brown <DrJKL0424@gmail.com>
Co-authored-by: Alexander Brown <drjkl@comfy.org>
## Summary
Adds a dedicated upgrade modal that appears when users without private
models access try to upload models, providing a clear path to upgrade
their subscription.
## Changes
- **New upgrade modal**: Created `UploadModelUpgradeModal` with
dedicated body, header, and footer components
- **Conditional rendering**: Modified `AssetBrowserModal` to show
upgrade modal when `privateModelsEnabled` flag is false
- **Subscription integration**: Connected upgrade flow to existing
subscription system via `showSubscriptionDialog()`
- **Localization**: Added localization keys for upgrade messaging
## Review Focus
- Conditional logic in `AssetBrowserModal.handleUploadClick()` based on
feature flags
- Component naming consistency (all upgrade-related components prefixed
with `UploadModelUpgrade`)
- Footer component refactoring maintains existing upload wizard behavior
🤖 Generated with [Claude Code](https://claude.com/claude-code)
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-7124-feat-add-upgrade-modal-for-model-upload-when-private-models-disabled-2be6d73d36508147b72eea8a1d6ab772)
by [Unito](https://www.unito.io)
---------
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Alexander Brown <drjkl@comfy.org>
## Summary
Feature flags for model upload button and asset update options now check
remote config from `/api/features` first, falling back to websocket
feature flags.
## Changes
- **What**: Added `model_upload_button_enabled` and
`asset_update_options_enabled` to `RemoteConfig` type
- **What**: Updated feature flag getters to prioritize remote config
over websocket flags
- **Why**: Enables dynamic feature control without requiring websocket
connection, consistent with other feature flags pattern
## Review Focus
- Pattern consistency with other remote config feature flags
- Proper fallback behavior when remote config is unavailable
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-7143-feat-Add-remote-config-support-for-model-upload-and-asset-update-feature-flags-2bf6d73d3650819cb364f0ab69d77dd0)
by [Unito](https://www.unito.io)
Co-authored-by: Claude <noreply@anthropic.com>
This pull request refines the loading and error handling logic for both
the `VideoPreview.vue` and `ImagePreview.vue` components. The main
improvements include making the loading skeleton more accurate and
visually consistent, updating how loading and error states are managed
when URLs change, and ensuring that the main media elements are hidden
while loading. These changes enhance the user experience by providing
clearer feedback during media load operations.
**Loading and error state improvements:**
* The loading skeleton in both `VideoPreview.vue` and `ImagePreview.vue`
now only appears when loading and no error is present, with updated
styling and fixed dimensions for better consistency. (`VideoPreview.vue`
[[1]](diffhunk://#diff-17b5c19b4628f22e45570b66a85ed1fc16e931dd368fe420584d487e522ab8aaL29-R41)
`ImagePreview.vue`
[[2]](diffhunk://#diff-0c0b17c5c68464e0284398ad42b823509d414c9cf297f3bc2aa2b00e0f9c2015L29-R48)
* The main video and image elements are now hidden (using the
`invisible` class) while loading, preventing display glitches before the
media is ready. (`VideoPreview.vue`
[[1]](diffhunk://#diff-17b5c19b4628f22e45570b66a85ed1fc16e931dd368fe420584d487e522ab8aaL29-R41)
`ImagePreview.vue`
[[2]](diffhunk://#diff-0c0b17c5c68464e0284398ad42b823509d414c9cf297f3bc2aa2b00e0f9c2015L29-R48)
* The loading state (`isLoading`) is now set to `true` whenever new URLs
are provided, and reset appropriately when navigating between media
items, ensuring accurate feedback to the user. (`VideoPreview.vue`
[[1]](diffhunk://#diff-17b5c19b4628f22e45570b66a85ed1fc16e931dd368fe420584d487e522ab8aaL145-R152)
`ImagePreview.vue`
[[2]](diffhunk://#diff-0c0b17c5c68464e0284398ad42b823509d414c9cf297f3bc2aa2b00e0f9c2015L164-R176)
[[3]](diffhunk://#diff-0c0b17c5c68464e0284398ad42b823509d414c9cf297f3bc2aa2b00e0f9c2015L224-R236)
**Code consistency and maintainability:**
* Both components now import and use the shared `cn` utility for
conditional class names, improving code consistency and maintainability.
(`VideoPreview.vue`
[[1]](diffhunk://#diff-17b5c19b4628f22e45570b66a85ed1fc16e931dd368fe420584d487e522ab8aaR115)
`ImagePreview.vue`
[[2]](diffhunk://#diff-0c0b17c5c68464e0284398ad42b823509d414c9cf297f3bc2aa2b00e0f9c2015R132)
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-7094-Fix-skeleton-loaders-for-Image-Video-Previews-2bd6d73d3650817989e1f4d597094016)
by [Unito](https://www.unito.io)
Increases the `/features` endpoint polling interval (runtime config
polling) from the current 30 seconds to 10 minutes.
## Background
The polling was originally added for two main purposes:
1. Server alert badges
2. Extra assurance that states are synchronized between frontend and
backend
However, both of these use cases are not critical:
- Server alert badges are unlikely to be needed in practice. WebSocket
(WS) can be used eventually as an alternative
- For synchronization, the system should be redesigned to be explicit
about marking client state as stale, or use WebSocket/Server-Sent Events
(WS/SSE) on a per-data basis rather than polling for the entire feature
flag set
## Motivation
The reason to reduce polling frequency is that per-user feature flags
are being added, which will make `/features` endpoint handling
significantly heavier. The endpoint will no longer just return static
JSON with high cache age, making frequent polling more costly.
## Future Considerations
- Eventually migrate server alert badges to use WebSocket
- Design a system that explicitly marks client state as stale when
needed
- Consider using WebSocket or Server-Sent Events for targeted data
updates instead of polling entire feature flag set
┆Issue is synchronized with this [Notion
page](https://www.notion.so/PR-7100-cloud-increase-feature-flag-polling-interval-to-10min-from-30s-2bd6d73d365081dfa8abeec901d0d975)
by [Unito](https://www.unito.io)
---------
Co-authored-by: Alexander Brown <drjkl@comfy.org>
