Texture Diffusion v2 - AI texturing addon (two modes, cloud providers, pay per image)

I just put out Texture Diffusion v2, a full rewrite of an addon I released a couple of years ago.

It textures 3D models using cloud AI image models, directly inside Blender. You pick a provider (Replicate, Google Gemini, or fal.ai), pick a model, generate, and project the result on your mesh.

There are two workflows that share the same Generate panel:

mail_001

  • Global Texture: you place duplicates of your object at several angles inside a single camera frame. The AI sees the whole object in one shot, and the addon assembles a coherent texture across every viewpoint. Single frame instead of separate per-view generations, so the seams don’t show.

mail_002

  • Live Paint: you frame an area, generate, and paint the result on the mesh with Blender’s stencil brush. Good for adding detail to a specific zone.

The two feed each other: lay down a base with Global Texture, then refine zones with Live Paint.

A few things worth noting:

  • No subscription. You buy the addon once, then pay your provider per image (a few cents each, usually $0.02 to $0.16).
  • No heavy GPU needed, the generation runs in the cloud.
  • Works on Blender 5.0 and 5.1.

Launch week pricing is $19 until May 27, then $24.

Gumroad: https://adriflex.gumroad.com/l/texturesdiffusion
Superhive: https://superhivemarket.com/products/textures-diffusion

Happy to answer questions.

2 Likes

Does it only create albedo?

1 Like

Yes, it only creates the albedo!
To convert the image to PBR, you’ll need to look into other tools.

1 Like

Amazing, but Trellis gives all PBR textures, and it runs locally…so why not implement Trellis2 into your add-on?

1 Like

I agree, TRELLIS.2 is really interesting. The main reason I didn’t build TD v2 around it is the local setup: it needs a serious GPU and a fairly heavy CUDA/ComfyUI install, which is a lot to ask from many Blender users.

TD v2 goes the other way: strong cloud image-edit models, simple setup, works on almost any machine. And for pure texture quality / artistic control, I’m still getting better results with those models.

A local TRELLIS or ComfyUI backend could be a nice optional feature later though.