- Update `README.md` to recommend "Nano Banana (non-Pro)" for Asset Extraction, with "Nano Banana Pro" as a fallback.
- Update `README_CN.md` with the same recommendation translated into Chinese.
- Fix typo "Nanao" to "Nano" in `README_CN.md` and ensure consistent use of "Nano Banana Pro".
Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>
- Created `EnlargableImage` component in `src/ui/components/EnlargableImage.tsx`.
- Updated `ExtractionStage` to use `EnlargableImage` for displaying extraction results.
- Supports full-screen image viewing with zoom-in/out cursors and click-to-close behavior.
Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>
* Refactor Extractor workflow to use atomic extraction
* Refactor Extractor workflow to use atomic extraction
* Refactor Extractor workflow to use atomic, sequential extraction
* fix(extractor): resolve overflow issues in SelectionStage
- Constrain `ScrollArea` in `SelectionStage` with `min-h-0` to ensure proper scrolling within flex container.
- Add `overflow-hidden` to `SelectionStage` container.
- Add `overflow-hidden` to `ExtractorDialog` content area to prevent content spillover.
- Ensure header in `SelectionStage` does not shrink (`shrink-0`).
* fix: ensure equipment is updated on re-extraction
- Verified that `src/api/extract.ts` uses `equipmentService.updateEquipment` when `previousEquipmentId` is provided.
- Confirmed that `src/service/equipment.ts` performs an SQL UPDATE and cleans up the old asset file.
- No new code changes were needed as the previous submission already covered this requirement.
* fix: use default extract model in re-extract dialog
- Updated `ExtractorDialog` to fetch `default_extract_model` from settings.
- Passed `defaultModel` to `ExtractionStage`.
- Updated `ExtractionStage` to initialize the re-extract model selection with `defaultModel`.
* fix: correct setting key for default extract model
- Updated `ExtractorDialog` to use `step_refine_model` instead of `default_extract_model` to fetch the default extraction model from settings, as confirmed by the `ExtractionFlowDesigner` configuration.
* fix: update extraction stage layout and buttons
- Enforced a square aspect ratio for the image container in extraction cards using `aspect-square`.
- Removed the "Cancel" button from the footer in `ExtractionStage`.
- Updated the footer to be always visible.
- Changed the "Finish" button to display "Extracting..." and be disabled while extraction is in progress, and "Finish" when complete.
* fix: resolve lint errors and refine extraction UI
- Removed unused `onCancel` prop from `ExtractionStage` to fix linting error.
- Updated `ExtractorDialog` to remove the unused `onCancel` callback passed to `ExtractionStage`.
- Ensured consistent image sizing (square aspect ratio) in extraction cards.
- Refined footer behavior to always show actions, with dynamic "Extracting..." state.
* refine ui
---------
Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>
Co-authored-by: yiling ji <njjyl723@gmail.com>
Configures `Bun.build` with `publicPath: '/'` to ensure that generated script and style tags in `index.html` use absolute paths (e.g., `/chunk-....js`). This fixes "Failed to load module script" errors when accessing the application via deep links (e.g., `/characters/...`) in production environments where the browser would otherwise attempt to load assets relative to the nested path.
Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>
* Add Chinese README and language switcher
* Add Chinese README and language switcher
* Update READMEs with model recommendation for Refine Asset step
---------
Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>
Updated model recommendations for equipment extraction, character generation, and image generation. Added notes on budget-friendly alternatives and potential limitations of non-Pro models.
- Fix `RefineStage` overflow by applying `overflow-hidden` to the parent container and `min-h-0` to the `ScrollArea`.
- Update `generateTextureSheet` to use Chroma Green background.
- Update `refineAsset` prompt to extract the main equipment into a white background, using name and description for context.
- Update tests and API to support these changes.
Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>
* Update Look Details Dialog with image zoom and scrollable list
* add missing migration
* minor change
* fix test
---------
Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>
Co-authored-by: yiling ji <njjyl723@gmail.com>
- Updated `getGridDimensions` in `src/service/extraction.ts` to use a 2x2 grid for 4 or fewer assets, and a 3x3 grid otherwise.
- Updated `src/service/extraction.test.ts` to reflect the new grid dimensions in the `cropAssets` test.
Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>
Renamed the keyframes in `AnalyzeStage.tsx` to `scanline-analyze` to prevent conflict with the global `scanline` animation.
Also removed unused `scanline` keyframes from `RefineStage.tsx` to avoid potential future conflicts.
Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>
* feat(ai): use OpenAI Image Edit API instead of Responses API
Replaces the deprecated/unofficial Responses API usage for image generation with the standard `client.images.edit` API.
This change supports multi-image inputs and follows the user's provided example for `gpt-image-1.5` model compatibility.
- Updates `generateImageOpenAI` in `src/service/ai.ts` to use `client.images.edit`.
- Handles conversion of text prompts and inline/reference images to the required format using `toFile`.
- Updates `src/service/ai.test.ts` to mock the new API call and verify correct behavior.
* feat(ai): use OpenAI Image Edit API instead of Responses API
Replaces the deprecated/unofficial Responses API usage for image generation with the standard `client.images.edit` API.
This change supports multi-image inputs and follows the user's provided example for `gpt-image-1.5` model compatibility.
- Updates `generateImageOpenAI` in `src/service/ai.ts` to use `client.images.edit`.
- Removes `response_format` parameter as it's not supported/needed for GPT image models (which default to base64).
- Handles conversion of text prompts and inline/reference images to the required format using `toFile`.
- Updates `src/service/ai.test.ts` to mock the new API call and verify correct behavior.
---------
Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>
* Fix toast visibility z-index issue and Create Look dialog error handling
- Moved `<Toaster />` outside of `SidebarProvider` in `Layout.tsx` and increased z-index to ensure toasts appear above dialog overlays.
- Updated `CreateLookDialog.tsx` to use `toast.error()` for API errors instead of displaying them inline at the bottom of the dialog.
* Fix Extractor error handling, toast visibility, and Create Look dialog errors
- Updated `ExtractorDialog.tsx` to reset the stage to `'selection'` on refinement error, preventing infinite loading.
- Updated `CreateLookDialog.tsx` to use `toast.error()` for API errors instead of inline text.
- Moved `<Toaster />` in `Layout.tsx` outside of `SidebarProvider` to ensure toasts appear above dialog overlays.
- Ensured `package.json` and `sonner.tsx` remain clean (reverted temporary changes).
* fix sonner
* fix lint
---------
Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>
Co-authored-by: yiling ji <njjyl723@gmail.com>
- Moved `GithubIcon` to `src/ui/components/Icons.tsx` as `Icons.Github`.
- Updated `src/ui/Layout.tsx` to use the new `Icons.Github` component.
- Increased icon size to `h-7 w-7`.
- Deleted `src/ui/components/GithubIcon.tsx`.
- Removed `subCategory` field from Equipment model (DB), API, and UI.
- Generated Prisma migration `20260219022619_remove_subcategory_and_update_settings`.
- Implemented API Key masking (******) in settings service and UI.
- Updated Settings page to show 'Saved' state for API Key/URL buttons.
- Renamed 'Extraction Flow Defaults' to 'Default Models' in Settings.
- Fixed Fitting Room scrolling issue by adding `min-h-0` constraints.
- Updated tests and verifying scripts.
- Added download option to VibeCard dropdown menu when an image is present.
- Added download option to Fitting Room generated look card dropdown menu.
- Implemented client-side download logic using temporary anchor tag.
- Verified with typecheck and tests.
- Updated `docker-compose.yml` to pull `yilingj/forgery:latest` instead of building locally.
- Updated `README.md` with instructions for using Docker Compose with the pre-built image.
- Added instructions for standalone `docker run` execution in `README.md`.
- Added `.github/workflows/release.yml` for multi-platform Docker builds.
- Configured triggers for `main` branch pushes and `v*` tags.
- Uses `docker/metadata-action` for tagging:
- `latest` is only updated on semantic version tags (e.g., `v1.0.0`).
- `main` branch pushes are tagged with `main` and commit SHA.
- Supports `linux/amd64` and `linux/arm64`.
- Uses hardcoded `yilingj` username and `DOCKERHUB_TOKEN` secret.
- Added `.github/workflows/release.yml` for multi-platform Docker builds.
- Configured triggers for `main` branch pushes and `v*` tags.
- Uses `docker/metadata-action` for tagging:
- `latest` is only updated on semantic version tags (e.g., `v1.0.0`).
- `main` branch pushes are tagged with `main` and commit SHA.
- Supports `linux/amd64` and `linux/arm64`.
- Requires `DOCKERHUB_USERNAME` and `DOCKERHUB_TOKEN` secrets.
- Added `.github/workflows/release.yml` to build and push Docker images to Docker Hub.
- Configured trigger on `push` to `main` branch and `v*` tags.
- Enabled multi-platform builds for `linux/amd64` and `linux/arm64`.
- Uses `docker/metadata-action` for automatic tagging (`latest`, semver).
- Requires `DOCKERHUB_USERNAME` and `DOCKERHUB_TOKEN` secrets.
- Update `VibeCard` to support custom `children` content and ensure text readability with a gradient overlay.
- Refactor Outfit list in `Equipments.tsx` to use a grid of `VibeCard` components.
- Implement a 3x3 equipment preview grid inside Outfit cards, with overflow (+N) handling.
- Center the preview grid to 80% height as per design requirements.
- Update `CreateOutfitDialog` to support editing existing outfits (PUT request).
- Change category selection in `CreateOutfitDialog` to single-select.
- Update equipment selection grid in `CreateOutfitDialog` to fixed 3 columns.
- Add "Edit" action to Outfit cards.
- Fitting Room:
- Constrain page height to viewport.
- Use `ScrollArea` for "Generated Looks".
- Ensure character image uses `object-contain` and empty state is centered.
- Extractor:
- Add "Save as Outfit" checkbox and name input to `RefineStage`.
- Implement API call to create outfit upon completion.
- Dockerfile:
- Copy `example-data` directory to final image.
- Fixed Fitting Room page overflow by constraining height and using ScrollArea for the Generated Looks section.
- Added "Save as Outfit" checkbox and input to the Refine stage of the Extractor dialog.
- Installed @radix-ui/react-checkbox and added Shadcn Checkbox component.
- Updated Dockerfile to include example-data directory in the final image.
- Removed hardcoded `BUILTIN_POSES` and `BUILTIN_EXPRESSIONS` constants.
- Updated `PoseService` and `ExpressionService` to only list and delete items from the database.
- Removed the `type` field ('builtin' | 'custom') from `PoseItem` and `ExpressionItem`.
- Updated `GenerationService` to resolve pose and expression assets strictly from the database (`data/files`) instead of falling back to `public/`.
- Updated `Poses.tsx` and `Expressions.tsx` to remove "SYSTEM" vs "CUSTOM" UI distinction and enable deletion for all items.
- Updated `pose.test.ts` and created `expression.test.ts` to verify the new behavior.
- Fixed `generation.test.ts` to mock database calls correctly for pose/expression lookups and avoid polluting other tests.
- Removed hardcoded `BUILTIN_POSES` and `BUILTIN_EXPRESSIONS` constants.
- Updated `PoseService` and `ExpressionService` to only list and delete items from the database.
- Removed the `type` field ('builtin' | 'custom') from `PoseItem` and `ExpressionItem`.
- Updated `GenerationService` to resolve pose and expression assets strictly from the database (`data/files`) instead of falling back to `public/`.
- Updated `Poses.tsx` and `Expressions.tsx` to remove "SYSTEM" vs "CUSTOM" UI distinction and enable deletion for all items.
- Updated `pose.test.ts` and created `expression.test.ts` to verify the new behavior.
google-labs-jules[bot]