SkillPedia Skills Encyclopedia
中文 EN

Skills 2026-06-08 05:41 UTC

product-shots

Committed by whiletrue0x. Update Category: New Features & Capability Expansion. Commit details: add: text-overlay use_case in image-gen — gpt-image-2 routes for on-image text. This update introduces new capabilities and expands supported use cases.

whiletrue0x 063c508 motiful/product-shots

CommitRaw commit message

add: text-overlay use_case in image-gen — gpt-image-2 routes for on-image text

Promoted the lesson from the gallery regen pass into the image-gen skill so
every caller benefits, not just ad-creative. Gemini Nano Banana garbles
small text (<30pt); gpt-image-2 renders crisp headlines, callout labels,
CTAs, and price chips. The text-overlay branch wins over photorealistic /
image-to-image / creative defaults.

Changes:
- image-gen/references/model-selection.md: added "text-overlay" → gpt-image-2
  to select_default_model, with trigger-keyword detection. New decision-tree
  branch placed FIRST with "CHECK FIRST" marker. Family Differences table
  gains an "On-image text rendering" row. Catalogue strengths updated.
- image-gen/references/parameter-spec.md: new top row in Family-Specific
  Defaults — "Any output that must carry on-image text" overrides per-caller
  defaults below. Per-caller rows split (main-image secondary as callout vs
  visual variant; social-post text-bearing vs purely visual carousel).
- image-gen/SKILL.md: Step 1 comment lists "text-overlay" as a use_case and
  warns it wins over all others.
- ad-creative/SKILL.md: image-gen dispatch now passes use_case=text-overlay
  unconditionally (ad copy always present).
- main-image/SKILL.md: image-gen dispatch picks use_case based on whether
  spec.has_callout_labels — tech-spec Slot-2 routes to gpt-image-2; visual
  variants stay on Gemini i2i.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>