InputLightweight
Text-to-3D
A 3D model generated from a sentence.
A 3D model generated from a sentence, then loaded into the scene. The fastest way to get an object that does not exist yet. The result arrives messy and needs cleanup before it looks good.
- Breaks when
- topology is bad; needs cleanupslow generation
- Also called
- Meshy, Tripo, Shap-E, TRELLIS
Prompts
Prompt to placed asset
A pipeline script: send a prompt to a text-to-3D API, poll the task until done, download the glTF, run gltf-transform (dedup, weld, simplify to 20k triangles, resize textures to 1024, Meshopt), and write the result to /public/models with a JSON manifest (prompt, date, triangle count). The page loads the manifest and places the newest model on a turntable with an environment map.
Judge the mesh
Write a mesh report for a generated glTF: triangle count, non-manifold edges, texture sizes, bounding box, whether lighting is baked into the albedo (variance of luminance vs a flat-lit render), and a recommended cleanup list. Print it as a small HTML overlay next to the model.
Splat instead of mesh
Compare the same prompt’s output as a mesh and as gaussian splats for a product-page hero: load both, measure file size, load time and fps, and write the decision rule (splats for organic/scanned looks, meshes for lighting and shadows).