Skip to content

longcat video avatar 1 5 mlx.stub

Nicolas Cravino edited this page Jul 10, 2026 · 1 revision

id: longcat-video-avatar-1-5-mlx name: LongCat Video Avatar 1.5 MLX repo_path: ~/Documents/sw30labs/repos/longcat-video-avatar-1.5-mlx remote_url: https://github.com/sw30labs/longcat-video-avatar-1.5-mlx.git primary_language: Python framework: MLX (longcat-avatar-mlx runtime wrapper) category: local-inference-mlx stacks: [mlx, cli] ingested: 2026-07-09 last_commit_date: 2026-05-28

LongCat Video Avatar 1.5 MLX

Purpose

Local MLX app for mlx-community/LongCat-Video-Avatar-1.5-q4-dmd-merged — a talking-avatar video model driven by a portrait image + audio + prompt. Wraps the upstream xocialize/longcat-avatar-mlx runtime and gives it the same local shape as the other MLX projects.

Architecture

Follows the local-mlx-wrapper-pattern. install.sh (optional 24 GB q4 model pull), longcat CLI (sample-inputs, generate), stdlib GUI on port 8796. Outputs .mp4 + .npy + .json metadata. Conservative defaults (256×432, 29 frames, DMD 8-step q4); upstream full example is 480×832, 93 frames.

Dependencies

huggingface-hub (hf-xet/hf-transfer) for weight fetch; runtime deps live in the upstream checkout's venv. ~24 GB minimum unified memory per the model card.

CI/CD

No CI workflows. install.sh / gui.sh bootstrap.

Entities and topics

Clone this wiki locally