Show HN: Lemon Slice Live – Have a video call with a transformer model
Analysis
Lemon Slice introduces a real-time talking avatar demo using a custom diffusion transformer (DiT) model. The key innovation is the ability to generate avatars from a single image without pre-training or rigging, unlike existing platforms. The article highlights the technical challenges, particularly in training a fast DiT model for video streaming at 25fps. The demo's focus is on ease of use and versatility in character styles.
Key Takeaways
Reference
“Unlike existing avatar video chat platforms like HeyGen, Tolan, or Apple Memoji filters, we do not require training custom models, rigging a character ahead of time, or having a human drive the avatar.”