le-gpt-image v1.3：完整引入 GPT-Image2-Skill Reference Gallery（163 条精调 prompt / 31 类 / craft + 官方提示词指南，MIT，附 NOTICE 与原版 LICENSE，仅删预览图外链）；SKILL.md 工作流升级为 Gallery 优先（查库选案例做底改写，勿从零现写）；对照实测同主题海报 Gallery 框架质感显著优于即兴 prompt

2026-10-06 22:49:19 +08:00
parent d995f725d0
commit 3b1ee9f997
46 changed files with 4150 additions and 8 deletions
@@ -0,0 +1,45 @@
+# GPT Image 2.5: layout and exact text
+
+Read for posters, UI previews, educational diagrams and multi-panel images.
+Source check: **2026-09-09**;
+[provenance and validation boundary](openai-image-2.5.md#source-boundaries).
+
+## Official examples to consult selectively
+
+- [Exact text](https://developers.openai.com/api/docs/guides/image-prompting?model=gpt-image-2.5#render-exact-text)
+- [Interface previews](https://developers.openai.com/api/docs/guides/image-prompting?model=gpt-image-2.5#create-an-interface-preview)
+- [Scientific and educational visuals](https://developers.openai.com/api/docs/guides/image-prompting?model=gpt-image-2.5#create-scientific-and-educational-visuals)
+- [Slides, diagrams and charts](https://developers.openai.com/api/docs/guides/image-prompting?model=gpt-image-2.5#build-slides-diagrams-and-charts)
+- [Comic strips](https://developers.openai.com/api/docs/guides/image-prompting?model=gpt-image-2.5#turn-a-story-into-a-comic-strip)
+
+Treat these as task examples, not guarantees of spelling or factual accuracy.
+Do not infer that all text-heavy tasks require the highest quality setting.
+
+## Local brief-to-QA checklist
+
+Use the checklist below first. If a specific gap remains, consult only the relevant
+[craft.md](craft.md) section (1–7, 11 or 16); these are shared techniques, not 2.5 syntax.
+
+| Task | Freeze in the brief | Acceptance check |
+|---|---|---|
+| Poster | Exact approved strings, text block positions, hierarchy, margins | Every character, repeated/missing copy, clipping and reading order |
+| UI preview | Screen count, layout regions, component names, state and data | Labels, alignment, plausible controls; image output is not working UI |
+| Diagram/chart | Supplied facts/data, nodes, directed edges, axes, units and legend | Relations and values against the brief, not just visual attractiveness |
+| Comic/panel sheet | Grid dimensions, panel order, identity anchors and per-panel action | Panel count, sequence, consistent character/outfit and no unwanted duplicates |
+
+Keep required copy verbatim and distinguish it from explanatory instructions.
+Do not invent missing data, captions or citations. An attractive scientific
+visual still needs subject-matter checking; rendering does not validate it.
+
+For an event-poster starting point, consult **Typography-led event poster** in
+[community templates](templates-gpt-image-2.5.md). Its older source image is not
+evidence of 2.5 reproduction. A reference-based pose sheet also needs the
+[editing](openai-image-2.5-editing.md) attachment and preservation checks.
+
+## Refinement decision
+
+Compare the output against the frozen brief at readable zoom. Name the failed
+string, region or relation before suggesting one focused correction. If the
+user authorizes another call, keep the chosen model and approved copy unless
+they explicitly agree to change them. For translation of an existing design,
+start with the editing slice rather than regenerating a new layout.