feat: adapt image tuning by engine

This commit is contained in:
inman
2026-05-29 14:32:02 +08:00
parent 4b21d2999c
commit e36f28a668
12 changed files with 138 additions and 10 deletions

View File

@@ -295,6 +295,27 @@
- Local HTTP smoke test uploaded a PNG through `/api/v1/assets`, fetched `/api/v1/assets/:id`, and downloaded matching binary bytes from `/api/v1/assets/:id/download`.
- Mock-provider task flow created a queued job through `/api/v1/jobs`, Worker processed it to `succeeded`, generated output asset carried the API client tag, and download returned an attachment response.
## Session: 2026-05-29 - Engine-Aware Image Tuning
### Implementation
- **Status:** complete
- Actions taken:
- Replaced the image generation text-influence range slider with select options.
- Added create-page engine detection from `/api/health`.
- For Jimeng image generation, the UI now shows `文本影响` options: `创意 35`, `均衡 50`, `贴合 70`, `严格 85`, and submits `scale`.
- For EvoLink image generation, the UI now shows `生成质量` options: `快速`, `标准`, `精细`, and submits `quality`.
- Added per-request EvoLink `quality` support in the payload builder.
- Updated public API request typing, OpenAPI schema, API docs, and focused tests.
### Verification
- **Status:** complete
- Results:
- `npm test`: 7 files / 22 tests passed.
- `npm run build`: production build succeeded.
- Agent-browser snapshot on `/create` showed current EvoLink mode rendering `生成质量` with `快速 / 标准 / 精细`.
- Desktop and mobile screenshots showed the create-page controls fitting cleanly.
- Browser page errors list was empty.
## Error Log - Server Deployment Support
| Timestamp | Error | Attempt | Resolution |
|-----------|-------|---------|------------|