mlx-community/LongCat-Video-bf16

Text-to-Video • Updated about 6 hours ago

Note Base bf16 weights + cfg_step_lora + refinement_lora as separate files. ~42 GB. T2V / I2V / Continuation / 720p refinement / Long-Video / Interactive — all six task variants share this DiT.

meituan-longcat/LongCat-Video

Text-to-Video • Updated Oct 29, 2025 • 2.6k • • 502

Note Upstream PyTorch reference checkpoint. The MLX weights in this collection are converted from here by recipes/convert_longcat_video.py in the github.com/xocialize/longcat-video-mlx repo.

mlx-community/LongCat-Video-q4

Text-to-Video • Updated about 5 hours ago

Note 4-bit quantized variant. ~25 GB on disk, ~32 GB unified-memory minimum. DiT Linears quantized via mlx.nn.quantize(bits=4, group_size=64); skips adaLN_modulation, t_embedder, y_embedder, final_layer.linear. Best for 32 GB Macs.

mlx-community/LongCat-Video-q8

Text-to-Video • Updated about 5 hours ago

Note 8-bit quantized variant. ~31 GB on disk, ~48 GB unified-memory minimum. DiT Linears quantized via mlx.nn.quantize(bits=8, group_size=64); same skip patterns as q4. Best balance of disk savings vs quality.