Texture Studio native material model

Model Name: Displacement Base

This model generates a 32-bit depth map from high-quality, properly formatted diffuse maps. It is trained on texture depth rather than object depth. (Diffuse -> Displacement)

The model is a component of Texture Studio, currently in development; the repository is at (https://github.com/gnaservicesinc/Texture_Studio)

It is designed for integration into a broader pipeline. Although it may be used independently, it was developed for a specific application. One potential use outside the program involves generating diffuse maps with a LoRA—such as https://civitai.com/models/939491/diffuse-map-surfaces—and subsequently producing a depth map for use as a texture in Blender. This aligns with the intended use case; however, performance may vary because training utilized photographic diffuse maps rather than AI-generated inputs.

Texture Studio will include a LoRA trainer for refining this model’s behavior, as well as a trainer for base models such as this one.

This version requires high-quality diffuse maps free of noise, blurring, hot spots, uneven illumination, and shadows, that are properly framed. The presence of such artifacts in input material may result in suboptimal output.

Target: Displacement (a height map / depth map). Each training map employs its native grid without padding or resizing. The application evaluates this numeric checkpoint using Apple MPSGraph in Float32. This compact model trains all weights from scratch and includes complete standalone weights; no base model or LoRA adapter is required. Source images and an interpreter are not provided. Visual inspection is recommended prior to model selection.

Downloads last month
5
Safetensors
Model size
1,000k params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support