LongCat Video Avatar 1.5 BF16 Dmd Merged

--- license: mit library_name: mlx pipeline_tag: text-to-video tags: - mlx - apple-silicon - video-generation - audio-driven-video - longcat - distilled - flow-matching base_model: - meituan-longcat/LongCat-Video-Avatar-1.5 language: - en - zh --- Part of the [LongCat-Video-Avatar 1.5 — MLX](https://huggingface.co/collections/mlx-community/longcat-video-avatar-15-mlx-6a185d1af4a43074d882e375) collection. # LongCat-Video-Avatar-1.5-bf16-dmd-merged (MLX) Apple MLX bf16 weights for [LongCat-Video-Avatar-1.5](https://github.com/meituan-longcat/LongCat-Video) — Meituan's audio-driven video diffusion model — with the DMD step-distillation LoRA **pre-merged** into the DiT weights. Recommended variant for inference: produces the same outputs as the base + LoRA combination but in **8 sampling steps** with no LoRA loading at runtime. ## TL;DR | | | |---|---| | **Architecture** | Wan 2.1 VAE + umT5-XXL + Whisper-Large-v3 + 48-block Avatar DiT (with DMD LoRA pre-merged) | | **Params** | ~13.6 B DiT + ~11 B umT5 + ~0.6 B Whisper encoder + 0.5 B VAE | | **Format** | bf16, sharded safetensors (HF-style per-component subdirs) | | **Disk** | ~43 GB | | **Hardware** | Apple Silicon M-series, 64 GB+ unified memory recommended for 480p | | **Inference** | 8-step DMD distilled, FlowMatchEulerDiscreteScheduler with shift=7.0 | | **License** | MIT (matches upstream Meituan) | ## Quick start ```bash # 1. Pull weights (~43 GB) hf download mlx-community/LongCat-Video-Avatar-1.5-bf16-dmd-merged \ --local-dir ./weights # 2. Set up inference (Python 3.12) git clone https://github.com/xocialize/longcat-avatar-mlx cd longcat-avatar-mlx python3.12 -m venv .venv .venv/bin/pip install -e ".[parity]" .venv/bin/pip install librosa Pillow imageio imageio-ffmpeg # 3. Run end-to-end .venv/bin/python scripts/run_inference.py \ --weights ./weights/.. \ --variant merged \ --num-frames 93 \ --out output.mp4 ``` Programmatic usage: ```python import json import pathlib import mlx.core as mx from longcat_video_avatar.pipeline_mlx import LongCatAvatarPipeline, PipelineConfig from longcat_video_avatar.models.autoencoder_kl_wan import AutoencoderKLWan from longcat_video_avatar.models.avatar.longcat_video_dit_avatar import ( LongCatVideoAvatarTransformer3DModel, ) from longcat_video_avatar.models.umt5 import UMT5EncoderModel from longcat_video_avatar.models.whisper import Whis...

Source context: 0 downloads · 6 likes · Pipeline text-to-video · Library mlx · Repo mlx-community/LongCat-Video-Avatar-1.5-bf16-dmd-merged