Paths relative to the payload root (LifeOS/install/); verified in the 7.40.4 release and byte-identical at current main.
The defect. skills/Art/Tools/Generate.ts:498-501 rejects gpt-image-1/gpt-image-2/compare: "OpenAI image models and compare mode were REMOVED 2026-07-30 … Use --model nano-banana-pro (the default)." skills/Art/SKILL.md:230 documents the removal and says not to reintroduce the path. But four doc references still route work to the removed model, so an agent following any of them lands on a hard error:
skills/Art/SKILL.md:43 — lists GPT-Image-2 in the model lineup.skills/Art/SKILL.md:47 — "text-heavy cards want GPT-Image-2".skills/Remotion/Workflows/GeneratedContentVideo.md:22 — "Art skill configured (Nano Banana Pro or GPT-Image-2)".skills/Remotion/Tools/Ref-ai-pipeline.md:17 — "Art skill (Nano Banana Pro / GPT-Image-2)".
A payload-wide search shows these four are the only remaining recommendations; every other gpt-image mention is the removal notice or the rejection code itself. The skill's own model table (Art/SKILL.md:238) already routes text-heavy stat cards to nano-banana-pro. Not a duplicate of #380/#348/#326 — those were size-validation bugs when the OpenAI path was still live.
Fix (PR attached, 3 files, +4/−4):
Art/SKILL.md:43 → lineup becomes "using Flux and Nano Banana Pro (Gemini 3 Pro)".Art/SKILL.md:47 → "(text-heavy cards and editorial headers want Nano Banana Pro)", matching the table at :238.- Both Remotion references → "(Nano Banana Pro)", preserving the
#1760 credit comment and the pipeline diagram's arrow alignment.
Paths relative to the payload root (
LifeOS/install/); verified in the 7.40.4 release and byte-identical at currentmain.The defect.
skills/Art/Tools/Generate.ts:498-501rejectsgpt-image-1/gpt-image-2/compare: "OpenAI image models and compare mode were REMOVED 2026-07-30 … Use --model nano-banana-pro (the default)."skills/Art/SKILL.md:230documents the removal and says not to reintroduce the path. But four doc references still route work to the removed model, so an agent following any of them lands on a hard error:skills/Art/SKILL.md:43— lists GPT-Image-2 in the model lineup.skills/Art/SKILL.md:47— "text-heavy cards want GPT-Image-2".skills/Remotion/Workflows/GeneratedContentVideo.md:22— "Art skill configured (Nano Banana Pro or GPT-Image-2)".skills/Remotion/Tools/Ref-ai-pipeline.md:17— "Art skill (Nano Banana Pro / GPT-Image-2)".A payload-wide search shows these four are the only remaining recommendations; every other
gpt-imagemention is the removal notice or the rejection code itself. The skill's own model table (Art/SKILL.md:238) already routes text-heavy stat cards tonano-banana-pro. Not a duplicate of #380/#348/#326 — those were size-validation bugs when the OpenAI path was still live.Fix (PR attached, 3 files, +4/−4):
Art/SKILL.md:43→ lineup becomes "using Flux and Nano Banana Pro (Gemini 3 Pro)".Art/SKILL.md:47→ "(text-heavy cards and editorial headers want Nano Banana Pro)", matching the table at:238.#1760credit comment and the pipeline diagram's arrow alignment.