An Image-to-Video (I2V) Generation Model from scratch in PyTorch to demystify video diffusion/flow-matching models

NanoI2V is a step-by-step educational repository for building a full Image-to-Video model from the ground up.

Core building blocks included:

• 3D VAEs & Latent video manipulation
• Diffusion Transformer (DiT) architecture
• Flow Matching & Diffusion trajectories
• Image Conditioning & CFG (Classifier-Free Guidance)
• Rotary Position Embeddings (RoPE)

If you're looking for a readable, modular project to learn how modern video generation works under the hood (or to use as a starting point for your own experiments), check it out:

🔗 Repo:https://github.com/Shubham2376G/NanoI2V

Drop a star if you find it helpful, and let me know what you think!