Stable Diffusion in Maya

Mandala (Tibetan: དཀྱིལ་འཁོར།) A traditional painting made from colored sand. Once complete, Mandalas are dismantled, reflecting the traditional Buddhist belief in the transitory nature of material life.

Ever wondered what Stable Diffusion would be like as a Maya renderer?

Spoiler alert: It’s awesome. Mandala implements most of Stable Diffusion functionalities into Maya, without relying on external tools.

Update: V1.17

New camera mapping system! works much better than previously.

Update: V1.16

Lots of bug fixes
Revamped UI
Image editing functions
Magic AI eraser brush function
All modes now working
… and LOTS of new stuff.

Update: V1.13 Alpha

The create 3D models function, using depth maps and Alpha, is now fully working! Perfect for animation, or 2.5D. NERF, full-3D models in one of the next updates!
A few minor bug-fixes and UI improvements.
New menu functions: Create Alpha/Depth, Normals and Enhance Faces.
Attributes have been renamed and reorganized. The face enhancer has been moved out of the Upscaler pipeline.

Update: V1.12 Alpha

Added a “create 3D model” function. This is a WIP. Right now, it just creates an alpha-cut model.

Update: V1.11 Alpha

Fixed a bug when saving images with an alpha channel. The alpha is also taken into account when creating textures or image planes.
Fixed a bug with LORA weights not patching correctly on torch 2.0

Update: V1.1 Alpha

Can now render with an Alpha channel! Using Dichotomous Image Segmentation (DIS), or the RGBA pass’ alpha when using ControlNet or Depth2Img.

V1.0 Alpha

Initial release


Features

1. Integration in Maya

Arnold Aovs are directly fed into Stable Diffusion. Canny edges, Hough lines, Sobel edges, and Openpose-style joints are also automatically rendered and used as input to condition Controlnet and Depth2img, in just a few clicks.

2. ControlNet

Use ControlNet to turn basic 3D models into awesome, fully lit, textured images in 3 clicks. No complex setup, no connection to make manually, and no external software to run!

3. Depth2Img

Like ControlNet, Depth2Img uses a depth pass to generate anything you want. It’s amazing for concept art, making quick textures, experimental stuff, etc.

4. Texture Generator

Generate tileable, high-res texture maps with normal maps out of a simple prompt!

5. Camera Mapping

Mandala includes an advanced, layered camera mapping function that takes into account the camera occlusion and its frustum to create masks! This makes it super-easy to texture assets using AI.

6. Model Manager

Fully featured ML model manager. Add any model from the awesome HuggingFace.co or Civita.ai and try exciting new things on-the-fly!

7.Prompt Editor

Craft your prompts in an un-cluttered, no-nonsense UI, with spell checking, history & presets.

8. Render View Editor

A Maya-style render view allows you to save images and settings with ease – in my tests, it’s actually faster than Maya’s original Renderview!

9. Inpainting

The good old region renders now in-paints the image to make corrections.

10. Render Animations

Make cool experimental animations! Batch rendering doesn’t work yet, but render animation does. It even saves all your settings along with the PNG files so you can re-use them later!