3D Models
Generate 3D models in Studio from text or images, with control over topology, texturing, and polygon budget. Rig and animate characters, exported as optimized GLB.
3D Model Generation#
Studio turns text prompts or reference images into 3D models, exported as web-optimized GLB files. Provide multiple views for higher fidelity, and control texturing, topology, and polygon budget.
3D generation runs as a background job — submit it and come back when it's done.
Choosing a model#
Studio offers several 3D models with different strengths — some balance detail and speed, some maximize fidelity, and others specialize in hard-surface shapes or accept multiple reference views. The picker explains what each is best at, and the settings panel only shows the options that model supports.
| Model | Best for | Reference views | Extra options |
|---|---|---|---|
| Hunyuan3D | Clean meshes with balanced detail and speed | Up to 4 | Custom polygon count, PBR, topology |
| Hunyuan 3.1 Pro | Highest fidelity, sharpest geometry | Up to 8 | Custom polygon count, PBR |
| Trellis 2 | Subjects with intersecting volumes | 1 | Custom polygon count, texture size, seed |
| Meshy | Strong quality straight from its raw mesh, with remesh support | Up to 4 | Custom polygon count, PBR, topology, remesh, symmetry, seed (text-to-3D only) |
| Rodin | Hard-surface and mechanical shapes | Up to 5 | Texture size, PBR, topology, seed |
| Rodin 2.5 | Sculpt-level detail with production-grade control | Up to 5 | PBR, seed |
| Hi3D | Ultra-dense meshes up to 2M faces from a single image | 1 | Custom polygon count, PBR |
The lineup evolves over time, so the picker in the app is always the source of truth — new models appear there before they appear here.
Text-to-3D and image-to-3D#
- Text-to-3D — Describe the object. If you don't provide a reference image, Studio generates one automatically to anchor the model.
- Image-to-3D — Upload one or more reference images. For models that accept multiple views, you can supply different angles (front, back, sides, top, bottom) for sharper results.
You can also send an image generated in the Studio image view straight into 3D as the reference.
Settings#
- Quality — Low, medium, or high, mapping to a polygon budget.
- Custom polygon count — Override the quality preset with a specific target.
- Texture size — 1K, 2K, or 4K.
- PBR — Toggle physically-based rendering textures on or off.
- Topology — Triangle or quad meshes.
- Remesh — Clean up the topology of the generated mesh (available with some models).
- Symmetry — Control mirrored geometry (available with some models).
- Seed — Reproduce a previous run (available with some models).
- Pose — Force characters into a T-pose or A-pose, so they come out ready for rigging.
The exact options shown depend on the model you pick.
Output and optimization#
Models are exported as GLB and automatically optimized for the web (including Draco mesh compression) so they load fast in scenes and previews. You can preview the model in Studio's 3D viewer, download the GLB, and reuse it.
Rigging and animation#
Turn a finished character model into an animated one. Hover a completed 3D generation and click Rig to open the animation picker and bake clips onto an auto-rigged skeleton:
- Browse the full animation library — hundreds of motions across Daily actions, Walk & run, Fighting, Dancing, and Body movements, with search and a curated Featured set.
- Pick up to 10 clips, then press Start rigging.
- The result is saved as a new 3D asset: a single GLB containing the rig and every clip you chose, with the original model left untouched.
- The 3D viewer gains an animation switcher so you can play each clip and preview the motion.
Need a different set later? Click Animations on a rigged model to reopen the picker with its current clips pre-selected, adjust the selection, and bake a new version — the original stays put.
Rigging runs as a background job, just like generation, and the rigged model lands in your Asset Library ready to download or drop into a scene.
Auto-rigging works best on humanoid characters standing in a T-pose or A-pose with arms and legs clearly separated from the body — the rig picker shows a visual guide of good and bad poses. To generate rig-ready characters, set the Pose option to T-pose or A-pose before you generate.
Using your models#
Finished models are saved to your Asset Library and can be:
- Downloaded as GLB
- Loaded into a Chat 3D scene as a referenced model
- Dropped into a Canvas design as a 3D object
When using a model that accepts multiple reference views, supplying clean front, back, and side images of the same subject produces noticeably sharper, more accurate geometry than a single photo.