Guides
Open-source text to 3D: how to generate 3D models from text with open models
Open source text to 3D model options: TRELLIS and Hunyuan3D text modes, and the reliable text-to-image-to-3D route in ComfyUI.
2 min read

Quick answer
Is there an open-source text to 3D model?
Yes. TRELLIS (MIT) accepts text prompts directly, and Hunyuan3D supports text-to-3D too. But the most reliable open route is two steps: generate an image of the object with an open image model, then turn that image into 3D with TRELLIS, Hunyuan3D or TripoSR. Image-to-3D models are stronger than direct text-to-3D.
Direct text to 3D vs text to image to 3D
Direct: you type "a brass lantern" and the 3D model generates the object. TRELLIS's README lists text prompts as an input alongside images, and Hunyuan3D supports text-to-3D. It's one step, but you can't see the design before the 3D step runs.
Two-step: an image model draws the lantern; you pick the best of several images; an image-to-3D model builds it. More steps, but much more control over the look, and image-to-3D is where open models are strongest.
The two-step pipeline
- Prompt an image of one object, centred, plain background, three-quarter view, even lighting.
- Generate several and pick the clearest.
- Remove the background if needed.
- Image to 3D with TRELLIS, TRELLIS 2, Hunyuan3D or TripoSR. See open-source image to 3D.
- Clean up in Blender: scale, decimate or retopologise.
ComfyUI is the usual place to chain these: ComfyUI 3D.
Prompts that work for 3D
- Name one object: "a wooden treasure chest with iron bands".
- Add view and background: "three-quarter view, plain grey background, studio lighting, no shadows".
- Avoid scenes, crowds, text and logos.
- For games, add a style: "stylised, low-poly, hand-painted".
More: AI 3D prompts.

Licences across the chain
Both the image model and the 3D model have licences. A permissive 3D model doesn't help if the image model forbids commercial use. Check both: open-source 3D AI licences.
Hardware
Running an image model and a 3D model locally needs a strong GPU; you can run them one after the other to save memory. See GPU for AI 3D generation.
Text, image or sketch to 3D in one place
Vi3W handles both routes in the browser: text to 3D directly, or text to image to 3D with a choice of 25 image-to-3D models, including TRELLIS and Hunyuan 3D. Results are retopologised and exported as GLB, FBX or OBJ. Try Vi3W.
Frequently asked questions
open source text to 3d model
Use TRELLIS's text mode for quick tries, or chain an open image model into an open image-to-3D model in ComfyUI for better control over the look.
best open source text to 3d model
The best results usually come from text to image to 3D, because you can pick the best image before spending GPU time on 3D.
Last reviewed

