Original experimental image generation model developed under the XenLogic project.
Model information
- Developer: XenLogic
- Version: 0.5
- Architecture: Custom PyTorch diffusion model
- Parameters: 30,323,593
- Image resolution: 32x32
- Training task: Colored geometric compositions
- Stable Diffusion: Not used
Architecture
The model uses a custom convolutional diffusion U-Net, spatial conditioning and a residual refinement network.
It was developed through successive experimental versions of XENON IMAGEN.
Capabilities
- Colored geometric shape generation
- Two-object composition generation
- Left/right spatial conditioning
- Text-conditioned denoising
Limitations
This is an experimental research checkpoint.
It is not a general-purpose photorealistic text-to-image generator. Some shape, color and spatial prompts may fail.
The checkpoint is a PyTorch state dictionary, not a standalone Diffusers or Transformers pipeline.
Reconstructing the model requires the original Python architecture classes used for v0.5.
Weights
model.safetensors contains the full v0.5 model state
dictionary, including its base and refinement modules.
Model author
XenLogic — XENON IMAGEN research project.
- Downloads last month
- 40