BLENDER

AI

get 35% off on all products with coupon tonda35

Generate seamless PBR textures, 360° HDR environments, and 3D meshes directly inside Blender — using local offline AI models on your GPU or cloud APIs with no GPU required.

🎨

Object Tab

Generate full PBR texture sets — diffuse, normal, roughness, AO, displacement — and apply them directly to your active object.

🌐

World Tab

Generate 360° equirectangular HDR environment maps and apply them as your scene world — complete with a Mapping node for rotation.

🧊

3D Tab

Generate 3D meshes from text or image input. Models import directly into the viewport as new mesh objects, ready to use.

Local Model — runs on your GPU

Download AI models (FLUX, SDXL, Juggernaut XL, StableMaterials, LDM3D, TripoSR, Shap-E and 20+ more) from HuggingFace. They run entirely on your machine — no internet after download, no API costs, no data leaves your PC.

Requires: Nvidia GPU · Python environment setup · Model download

Cloud API — no GPU required

Use external APIs with your own API key. Supports fal.ai (1000+ models), Replicate (any model by ID), OpenAI (GPT Image, DALL-E), HuggingFace Inference, Google Nano Banana for textures — and Meshy & Tripo AI for 3D.

Requires: API key from provider · Internet connection · No GPU needed

Non-blocking generation with live progress bar

Blender never freezes. Generation runs in a background thread. A progress bar fills in real time with step-level updates from the model.

Full PBR map generation

Generates diffuse, normal, roughness, AO, and displacement maps. StableMaterials outputs real multi-channel PBR. All maps are wired into a Principled BSDF node tree automatically.

Gallery with history and quick re-apply

Every generation is saved to disk with a thumbnail. Browse past results in a visual gallery, re-apply any texture or world with one click, or delete what you don't need.

Texture tiling control in the gallery

Set a scale value and click Apply to update the Mapping node on the active material. Scale up to tile more, scale down to stretch — without re-generating.

Auto-apply checkboxes per tab

Enable "Apply to Object", "Set as World", or "Import 3D Model" to have results applied immediately on generation finish — no extra clicks needed.

VRAM check before generation

Before launching any local model, the addon checks available GPU memory and compares it against the model's requirement. You get a clear warning instead of an out-of-memory crash mid-generation.

Set Up the Python Environment

Open Edit → Preferences → Add-ons → Kretos Blender AI. Click Set Up Environment. This installs PyTorch (CUDA GPU build) and all AI dependencies into an isolated folder — not into Blender's own Python. Takes 5-15 minutes on first run.

Download a Model

In the same Preferences panel, open Offline Texture & World Models and click Download next to any model. Start with SDXL 1.0 or DreamShaper 8 for a good balance of quality and VRAM use.

Open the Panel

In the 3D Viewport, press N to open the sidebar. Click the Kretos AI  tab. You'll see two panels: Gallery (top) and Generate (below).

Write a Prompt and Generate

In the Generate panel, select Object tab. Set source to Local Model. Pick your downloaded model. Type a prompt like "weathered red brick wall, seamless, 4K". Click Generate Texture.

Apply to Your Object

When generation finishes, the result appears in the Gallery. Select your object, then click Apply to Object in the Gallery panel. Or enable Apply to Object after generating to do this automatically.

Windows Only

This addon is currently compatible with Windows only. macOS and Linux support is planned for a future release. The Python environment setup, CUDA detection, and some file-locking workarounds are Windows-specific in this version.

Nvidia GPU strongly recommended for local models

The addon installs a CUDA 12.8 PyTorch build. Your GPU driver must be version 520 or newer. Generation on CPU is possible but extremely slow (30+ minutes per image). Check your VRAM against each model's requirement before downloading.

First-time environment setup downloads ~3-4 GB

PyTorch CUDA (~2.5 GB) + diffusers/transformers/shap-e and other packages (~500 MB). This happens once. After that, reinstalling the addon does not re-download if the environment folder is intact.

HuggingFace models may require a token or license agreement

FLUX.1 Dev requires accepting a license on the HuggingFace website before downloading. Set your HF token in Preferences under HuggingFace. Models like FLUX.1 Schnell and StableMaterials are fully open (Apache 2.0 / MIT) and download without a token.

Close other GPU-heavy applications before generating

Games, other AI tools, or a viewport with heavy geometry can leave VRAM occupied. The addon checks free VRAM before launching and warns you, but if generation still fails with an out-of-memory error, close other applications and try again.

HuggingFace hf-inference provider dropped image generation (July 2025)

If you use the HuggingFace cloud option, set the inference provider to fal-ai — not hf-inference. As of July 2025, hf-inference only handles CPU tasks like embeddings and classification. Image generation goes through fal-ai or replicate using the same HF token.

When reporting a bug, include:

Report bugs through the product page where you purchased this addon. We read every report and push fixes regularly.

⚠ Beta — v2.1.0

This addon is in active development. Core features work but you may encounter bugs or edge cases. Every report helps — see the bug report section in the documentation page for what to include.

KretosAI

Blender Addon · v2.1.0 Beta · Windows

Compatible with Blender 5.0+ · Windows only
This addon is open to bug reports — help us improve it.
© 2025 All rights reserved.

© 2025