mlx-community/LongCat-Video-Avatar-1.5-bf16-dmd-merged

Text-to-Video • Updated May 28 • 6

Note Recommended. DMD step-distillation LoRA pre-merged into the DiT. 8-step inference, smallest load step, simplest pipeline code. Use this unless you specifically need to toggle between 50-step base and 8-step distilled paths.

mlx-community/LongCat-Video-Avatar-1.5-bf16

Text-to-Video • Updated May 28

Note Alternate. Base bf16 weights + separate DMD LoRA file. Lets you switch between the 50-step base inference path and the 8-step DMD distilled path at runtime by toggling the LoRA.

meituan-longcat/LongCat-Video-Avatar-1.5

Updated Jun 4 • 2.06k • 818

Note Upstream PyTorch reference checkpoint. The MLX weights in this collection are converted from here by recipes/convert_longcat_avatar.py in the GitHub repo.

meituan-longcat/LongCat-Video

Text-to-Video • Updated Oct 29, 2025 • 1.8k • • 568

Note Base LongCat-Video model. Source for the umT5-XXL text encoder, Wan 2.1 VAE, and SentencePiece tokenizer that the Avatar 1.5 overlay reuses.