krea/krea-realtime-video
Full fine-tune vs Wan-AI/Wan2.1-T2V-14B. Fine-tune: 100% of the weights changed, across 40 of 40 layers.
Tensors changed: Attention 640/640, Norms 240/240, MLP 160/160, other 45/45, Embeddings 10/10
1 derivatives checked
- 6chan/krea-realtime-video-fp8: Quantization
2 other repos hold byte-for-byte identical weight files.