Text to Textured 3D Asset in Blender: StableGen + TRELLIS.2 + Qwen Pipeline
You type a description. A few seconds later, Blender shows you a rigged, textured 3D character—all without touching a terminal, leaving the viewport, or stitching together six different tools. That's the workflow people are building now with StableGen, a free Blender addon that chains TRELLIS.2 mesh generation, SkinTokens auto-rigging, and diffusion-based texturing in one run.
What's actually happening here
StableGen is a Blender add-on that textures meshes and scenes with diffusion models and generates new meshes from a prompt or image, rendering guidance images from cameras in your scene, sending the work to a ComfyUI server, and projecting the results back onto the meshes. In Blender you set up prompt, mesh, cameras, and projection; on the ComfyUI server side, mesh generation happens with TRELLIS.2, rigging with SkinTokens, and diffusion with SDXL or Qwen Image.
The real magic: you don't build separate workflows for each step. You describe a character and pick a preview image; TRELLIS.2 builds the mesh, SkinTokens rigs it, texture views are generated and projected onto it, or you can start from an image, skip the rig, or keep the texture TRELLIS.2 makes itself.
What people are making with it
A cyberpunk woman character has been generated and PBR-textured from a text prompt inside Blender with StableGen, with text prompt in and PBR-textured 3D character out, generated entirely inside Blender. Creators are shipping untextured subway scenes, retexturing them three different ways (default, fantasy palace, cyberpunk) in scene-wide texturing runs. Others are using Qwen Refine and Local Edit modes to touch up detail on existing textures.
StableGen v0.3.0 transformed it from a texturing tool into an end-to-end 3D asset creation pipeline, with 30+ built-in presets organized across 4 architecture groups, including TRELLIS.2 pipeline presets. The RunPod template someone built adds SDXL texturing, Qwen Image Edit texturing, PBR decomposition, TRELLIS.2 image-to-3D, and even a text-to-animation add-on.
What's actually good, what hits the wall
The honest part: you want Blender 4.2–4.5 or 5.1+, a CUDA-capable NVIDIA GPU with 8 GB VRAM minimum (16 GB+ if you want FLUX.1-dev or Qwen), and a separate ComfyUI install. That's not nothing. And StableGen's alternative texture paths sidestep some licensing restrictions, but you should check your pipeline's licensing before a paid project.
On the upside: StableGen is free on GitHub, it lives inside Blender (no round-trip to a web app), and the pipeline actually works. Creators hitting quality walls can layer: use TRELLIS.2 for the base mesh, then hand-tune with Qwen edits for highlights and detail.
How to try it
- Clone or download StableGen from GitHub.
- Have ComfyUI installed on the same machine or network—run the installer to pull custom nodes and models; package 13 sets up Qwen Image 2.1, package 8 mesh generation, package 14 auto-rigging.
- Install the StableGen addon in Blender, point it at your ComfyUI server.
- Set Architecture to TRELLIS.2, choose a preset tagged Mesh + texture, then click Generate asset and pick a preview image.
For a hosted backend without owning GPU hardware, someone published a RunPod template with the whole stack preloaded—ComfyUI, models, and all custom nodes.