ZYI 1 TINY
ZYI 1 TINY is a compact text-to-image model based on a Diffusion Transformer (DiT) trained using Rectified Flow.
Architecture
- ~59M trainable parameters
- Latent Diffusion Transformer
- Rectified Flow
- 256x256 image training
- VAE latent space
- Text-conditioned generation
- FLAN-T5 text embeddings
- PyTorch
Model family
ZYI — compact generative image models.
Checkpoints
The repository contains training checkpoints, including the latest and best available checkpoints produced during training.