[Arlieeee]

Adds Meshy T2.5 (ai_model: "meshy-t2.5"), a Smart Topology model, to image-to-3d and text-to-3d preview. Without --ai-model, smart topology behaves as before: no ai_model is sent and the API picks meshy-t2.

Added

• --ai-model meshy-t2|meshy-t2.5 on image-to-3d create and text-to-3d create (preview), valid with --model-type smart-topology only. Standard mode keeps no model flag. multi-image-to-3d gets none because the API has no Smart Topology there.
• --target-polycount is checked per model before the request: meshy-t2 100–15,000, meshy-t2.5 100–30,000. image-to-3d had no local range check before.
• Polycount default (D-078): meshy-t2 keeps the CLI's 10,000 (D-076). With meshy-t2.5 the CLI sends no polycount, so the API's 15,000 applies. The 10,000 pin was there to lift meshy-t2's 4,000 default; with meshy-t2.5 it would only lower the model's own default.
• These checks run on the merged payload, so --data gets the same rules. A Smart Topology ai_model without model_type: "smart-topology" fails with exit 2, and so does topology: "quad" with smart topology (the output is triangle-only).
• Docs: docs/usage.md, skills/meshy-cli/SKILL.md (one line), decisions D-078, migration notes "Changes in 0.4.3", endpoint contracts (smart_topology) and the capability matrix. Credits are the same as meshy-t2: 5 for a mesh only, and for image-to-3d 15 with 2K/4K textures or 20 with 8K. A text-to-3d refine of a meshy-t2.5 preview works the same as one of a meshy-t2 preview.
• Version 0.4.3.

Verification

• pnpm typecheck && pnpm test: 639/639 pass. New loopback tests check that t2.5 is accepted with smart topology on image-to-3d and text-to-3d preview and that the request body carries ai_model: "meshy-t2.5". They also check that 30,000 is accepted for t2.5 and refused for t2, that t2.5 leaves the default polycount to the API while t2 still sends 10,000, and that each refusal (without smart topology, with quad, on refine, on multi-image, an unknown model) exits 2 before any request.
• node dist/index.js --version prints 0.4.3. Compared with main, the command/flag tree only adds image-to-3d create --ai-model and text-to-3d create --ai-model. Nothing was removed.
• These changes were not run against the live API.

Release notes (draft)

Highlights

• Meshy T2.5 Smart Topology. image-to-3d and text-to-3d preview accept --model-type smart-topology --ai-model meshy-t2.5, with --target-polycount up to 30,000 (default 15,000). Without --ai-model, smart topology runs meshy-t2 as before (100–15,000, default 10,000). Credits are the same as meshy-t2.
• Behaviour changes to note.
  • On image-to-3d, a smart-topology --target-polycount outside the model's range now fails before the request (exit 2) instead of being refused by the API.
  • A Smart Topology ai_model without --model-type smart-topology, or topology: "quad" with smart topology (sent through --data), now fails before the request (exit 2).

npm install -g [email protected]   # or @meshy-ai/[email protected]

🤖 Generated with Claude Code