Skip to content

v0.6.10: Creative Lab option checks, Smart Topology on Meshy T2.5, image aspect ratio rules

Choose a tag to compare

@Arlieeee Arlieeee released this 09 Nov 05:13
86a59b3

Added

  • meshy_text_to_image says so when a pose does not get the aspect ratio asked for. The API renders
    an a-pose / t-pose image on a nano-banana model at 3:4 whatever aspect_ratio is (the GPT
    Image models keep it), and it does so for the 1:1 default too. The request is sent as before; the
    result message now carries a short note when pose_mode is set on a nano-banana model and
    aspect_ratio is not 3:4. The pose_mode and aspect_ratio descriptions say the same.

Changed

  • Smart Topology without ai_model now runs meshy-t2.5 (was meshy-t2), on meshy_text_to_3d
    and meshy_image_to_3d, so a game asset gets the newer model's sharper detail unless the caller
    names ai_model: "meshy-t2". Without target_polycount the API's own default for the model
    applies: 15,000 faces on meshy-t2.5 (4,000 on meshy-t2). The tool descriptions and server
    instructions name meshy-t2.5 as the default.
  • meshy_creative_lab: keychain name_text is described as what it is, a flat coloured detail
    printed on the face (not raised or cut), not engraving. The build keeps it; name only labels
    the task.
  • Published to npm through trusted publishing (GitHub OIDC), with provenance; no npm token is
    stored.

Fixed

  • meshy_creative_lab checks build_options against the product's ranges before the prototype is
    posted, so a bad value no longer comes back after the prototype has been charged.
    export_resolution is 0 (the server default) or a whole number from 64 to 512; the lamp,
    keychain / fridge-magnet, keycap and fidget-pixel options are held to their documented ranges,
    hex fidget pieces must be 6 or 8 mm, and lamp include_result_json: true needs
    output_format: "zip". Options the server adds later still pass through.
  • meshy_image_to_3d with Smart Topology and topology: "quad" is refused, as meshy_text_to_3d
    already did, instead of quietly dropping quad and returning triangles. Use meshy_remesh for
    quads.
  • meshy_text_to_image / meshy_image_to_image refuse a nano-banana ai_model with
    aspect_ratio 3:2 or 2:3 before any request (only the GPT Image models take those two).