tiny-sd-models

This is a DIFFUSION model trained with tiny-stable-diffusion.

Model Description

This is a Diffusion Transformer (DiT/MMDiT) trained for text-to-image generation in latent space.

Architecture

• Type: DiT or MMDiT (Multi-Modal Diffusion Transformer)
• Conditioning: CLIP text embeddings

Usage

import torch
from src.models.vae import create_vae  # or appropriate model import

# Load checkpoint
checkpoint = torch.load("model.pt", map_location="cpu")

# Create model and load weights
model = create_model(...)  # Use config from checkpoint
model.load_state_dict(checkpoint["model_state_dict"])

License

MIT License