Blender 4.0 – 5.1 add-on
From text to
stunning textures
Type a prompt. Get a finished PBR material on your mesh — albedo, roughness and normal, wired into a Principled BSDF and wrapped around the whole model. Not just the side facing the camera.
One-time purchase · $10+ · Bring your own API key, no subscription
- AI backends
- 5Gemini · OpenAI · fal.ai · Replicate · ComfyUI
- Cross-view consistency
- 25×measured, 0.224 → 0.009
- Dependencies
- 0nothing to pip install
- Mesh types tested
- 116-face cube to 20k icosphere
The problem everyone else has
Eight cameras. One texture.
Point an AI at your model from one angle and you get one good-looking view. Point it from eight angles and you get eight different materials fighting each other — a patchwork with seams down every edge.
Framanen textures your model one view at a time. Before each camera it re-renders the model with the texture painted so far and asks the AI to continue it, masking off what is already done. Every view extends a single coherent texture instead of inventing its own.
What it does
A texturing pipeline, not a filter
One dropdown, your own key
Google Gemini, OpenAI (gpt-image-1, DALL·E 3), fal.ai, Replicate, or your own local ComfyUI server. Switch any time. No subscription and no markup on your inference.
A material, not a colour map
Albedo is de-lit so baked shading is not shaded twice. Roughness comes from luminance and local contrast. Normals from Sobel gradients. All wired up for you.
No manual seam painting
Each view is weighted by how squarely it faces the surface, and a projective depth test rejects anything the camera could not see. UV padding and masked diffusion clean up the rest.
Your UVs, handled
No UV map? It unwraps. Mirrored or overlapping islands — common on characters and fatal for projection painting? It measures the overlap, re-unwraps, and tells you it did.
Fix one spot, keep the rest
Select faces in Edit Mode, describe the change, and only that region is re-generated and composited back over the existing texture.
Blender stays responsive
Network calls run on worker threads with a live progress readout. Press ESC in the viewport to cancel a run at any point.
Structure first
It sends the AI your geometry, not just words
Every camera renders four passes. The depth and normal maps go to the model with your prompt, so what comes back matches your mesh's real silhouette and volume.
What comes out
Three maps, already plugged in
The AI only produces colour. Framanen derives the rest and builds the Principled BSDF for you. Prefer to art-direct it? Procedural mode rebuilds the same derivation live in shader nodes.
How it works
Five steps, one button
-
Cameras are placed for you
1, 4, 6 or 8 views, fitted automatically to your mesh.
-
Passes are rendered
Base colour, depth, camera-space normal and an edge map, per camera.
-
The AI generates, view by view
Each view sees what the previous one painted, so they agree.
-
Results are projected back
Blended per pixel into one UV texture, seams handled.
-
The material is built
Roughness and normal derived, Principled BSDF assigned.
What you get
In the download
- The add-on — a single zip, installed the normal way through Preferences
- Illustrated user guide (English) — install, panel map, setting recipes and a troubleshooting table
- Kullanım kılavuzu (Türkçe) — the same guide, fully translated
- Both guides are single self-contained HTML files. Open them offline; no internet needed.
- Under 1 MB total.
What you need
Requirements
- Blender 4.0 – 5.1. Developed and tested against 4.5 LTS and 5.1.
- Cycles enabled. It is by default — you do not need to change your scene's engine.
- An API key from Gemini, OpenAI, fal.ai or Replicate — or a local ComfyUI server, which costs nothing to run.
- API usage is billed by your provider, not by this add-on.
Honest limits — better to know before you buy
- A fast base layer, not a finished hero asset. Close-up hero props still want a refining pass. For set dressing, background props, prototypes and look development it is often enough as-is.
- Sequential mode is only as faithful as your provider's inpainting. Gemini and OpenAI respect the mask well; fal.ai, Replicate and ComfyUI have no inpainting endpoint, so there the partial render is attached as a reference instead.
- Areas no camera can see stay empty. Deep pockets need more views or a pass with the inpaint tool.
- Speed depends on your provider. Local work — render, projection, bake, PBR — is about 30 seconds for 8 views. Expect a few minutes for a full texture set end to end.
- Developed and tested on Windows. The code is plain Python with no platform-specific calls, so macOS and Linux should work, but they are not yet verified.
Questions
Before you buy
Is this a subscription?
No. One-time purchase. You supply your own API key and pay your provider directly for what you use.
Can I use it commercially?
Yes — on any number of your own machines, for commercial work.
Which provider should I start with?
Google Gemini. It is fast, inexpensive, handles multi-image input well, and supports the masked inpainting that sequential mode relies on.
Will it work on my mesh?
It has been tested headless across 11 mesh types — from a 6-face cube to a 20,000-face icosphere, including meshes with no UV map, with mirrored UVs, with several material slots, and objects rotated, scaled non-uniformly and sitting away from the world origin.
Does it break when Blender updates?
It has no external dependencies, which is the usual cause of breakage. It is verified on both Blender 4.5 LTS and 5.1.
What licence is the code?
GPL-3.0-or-later, like every Blender add-on that uses the bpy API. The user
guides, their images and the Framanen name are not covered by that licence.
Do I need to understand ControlNet?
No. Pick a preset, write a prompt, press the button. The advanced controls are there when you want them.
Name your price
Texture your next model from a sentence
Currently $10+ — pay what you think it is worth. Leave the field blank and Gumroad will ask you at checkout.
or continue without entering an amount — you can set it on the next step.