Skip to content

feat(image): add GPT Image 2.5 tiers, align aspect ratios with the server, drop nano-banana [ENG-3054] - #7

Merged
yujiaxu-ops merged 2 commits into
mainfrom
yujiaxu/eng-3054
Sep 15, 2026
Merged

yujiaxu-ops merged 2 commits into
mainfrom
yujiaxu/eng-3054

Conversation

@yujiaxu-ops

@yujiaxu-ops yujiaxu-ops commented Sep 15, 2026

Copy link
Copy Markdown
Collaborator

Linear: ENG-3054

What

  • Add gpt-image-2-5-flare (fast tier) and gpt-image-2-5-sunburst (precision-edit tier) to --ai-model on text-to-image create and image-to-image create. gpt-image-2 stays the default.
  • Align the local aspect-ratio check with the server: every GPT Image model (2 and 2.5) accepts all seven ratios; only the nano-banana family rejects 3:2 / 2:3. The CLI previously refused gpt-image-2 --aspect-ratio 4:3 even though the API accepts it.
  • Drop the first-generation nano-banana from the flag surface. The API still accepts it; --data '{"ai_model":"nano-banana"}' remains the escape hatch. README example switched to nano-banana-2-lite.
  • Help text for --ai-model / --aspect-ratio updated; no pricing added to the command surface.

Why

meshyd shipped the 2.5 tiers in PRO-2839 (taichi-dev/meshyd#5891) and widened gpt-image-2 to 4:3 / 3:4 in ENG-2233 (#5506). docs.meshy.ai has not caught up yet, so the rules here were taken from pkg/server/api_image_generation_handlers.go and pkg/taskservice/imagegen.go on meshyd origin/main, and cross-checked against the live API (a bogus ai_model returns the full valid list, which includes both 2.5 names).

Verification

  • pnpm typecheck and pnpm test green (559 tests).
  • Live, through the built CLI: text-to-image create --ai-model gpt-image-2-5-flare --aspect-ratio 3:4 → task 01a0a308-cdab-765f-a182-3755ece5ce7c, SUCCEEDED, consumed_credits: 9.
  • Live, raw API: gpt-image-2 + 4:3 → task 01a0a304-da6f-750b-b8f9-3bc3d2c6b836, SUCCEEDED (confirms the widened ratio set).
  • --ai-model nano-banana-2-lite --aspect-ratio 3:2 still fails locally with exit 2 before any request.

Follow-up (separate)

docs.meshy.ai text-to-image / image-to-image / pricing / changelog pages still list four models and the old five-ratio set for gpt-image-2.

🤖 Generated with Claude Code

Co-Authored-By: Claude Fable 5.1 noreply@anthropic.com

yujiaxu-ops and others added 2 commits September 15, 2026 11:19
…rver, drop nano-banana [ENG-3054]

meshyd (PRO-2839, #5891) exposes two new public ai_model names on
/openapi/v1/text-to-image and /openapi/v1/image-to-image:
gpt-image-2-5-flare (fast tier) and gpt-image-2-5-sunburst (precision-edit
tier). Add both to the shared --ai-model table so the 2D image commands
accept them; gpt-image-2 stays the default.

Mirror the server's per-model aspect-ratio rule (ENG-2233, #5506): every
GPT Image model accepts all seven ratios, only the nano-banana family
rejects 3:2 and 2:3. The CLI previously refused gpt-image-2 with 4:3/3:4
even though the API accepts them.

Drop the first-generation nano-banana from the flag surface. The API still
accepts it; --data '{"ai_model":"nano-banana"}' remains the escape hatch.

Verified live: gpt-image-2-5-flare + 3:4 through the built CLI
(task 01a0a308-cdab-765f-a182-3755ece5ce7c, SUCCEEDED, 9 credits).

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…ite one [ENG-3054]

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
@yujiaxu-ops
yujiaxu-ops merged commit 6d8140f into main Sep 15, 2026
4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant