🎤 LongCat-Video-Avatar 1.5

Audio-driven avatar video generation by Meituan LongCat — 8-step distilled, INT8, single GPU. Model · Code

Mode

Image + Audio → Video Text + Audio → Video

Reference image

Speech / singing (trimmed to 20s)

Prompt

Longer, detailed prompts (appearance, action, scene) work best.

Resolution

Segments

1 ≈ 3.7s; each extra ≈ +3.2s

1 4

Result

Examples

Mode	Reference image	Speech / singing (trimmed to 20s)	Prompt	Resolution	Segments	ref_img_index	mask_frame_range	Seed