Update Match3D/image-generation docs & code

Adds/updates documentation, assets and implementation for Match3D and puzzle image generation workflows. Key changes: decision logs and pitfalls updated to prefer VectorEngine Gemini for Match3D material sheets and to require edits (multipart) for 1:1 container reference images; guidance added for when to use APIMart vs VectorEngine. .env.example clarified APIMart/Responses config. Many new public assets and PPT visuals added. Code changes across frontend and backend: updated shared contracts, server-rs match3d/puzzle/image-generation handlers, VectorEngine/OpenAI image generation clients, and multiple React components/tests to handle UI/background/container image signing, edits workflow, and puzzle UI background resolution. Added src/services/puzzle-runtime/puzzleUiBackgroundSource.ts and related test updates. Includes notes about multipart HTTP/1.1 requirement and test/verification commands in docs.
This commit is contained in:
2026-05-14 20:34:45 +08:00
parent d33c937ebc
commit 548db78ca7
103 changed files with 6687 additions and 3270 deletions

View File

@@ -47,7 +47,7 @@ Default body:
}
```
For a reference image, add:
For weak visual references in text-to-image generation, add:
```json
{
@@ -55,6 +55,26 @@ For a reference image, add:
}
```
For image-to-image work that must follow a reference image closely, use the VectorEngine edits endpoint instead of the generations `image` array:
```text
POST {VECTOR_ENGINE_BASE_URL}/v1/images/edits
Authorization: Bearer {VECTOR_ENGINE_API_KEY}
Content-Type: multipart/form-data
```
Multipart fields:
```text
model=gpt-image-2
prompt=<prompt>
n=1
size=1024x1024
image=@reference.png
```
Prefer edits for workflows where the reference image controls composition, pose, container shape, or layout. In this repository, Match3D container UI generation uses edits with `public/match3d-background-references/pot-fused-reference.png` as the `image` part.
Accept image output from `data[].url`, `data[].b64_json`, or direct nested `url` fields. VectorEngine GPT-image-2-all currently returns synchronously; do not poll APIMart task endpoints.
## Environment