🎤 LongCat-Video-Avatar 1.5: Audio-Image-to-Video

Upload a reference image + audio clip + a short text prompt. Generates a lip-synced video (5-15s) using Meituan's LongCat-Video-Avatar 1.5 (INT8 DiT + DMD2 8-step distilled), running on ZeroGPU.

💡 Tips — Longer, more descriptive prompts (appearance, action, scene) give better consistency. Use Isolate vocals when the clip has music or background noise. Try Exact 8-step if you want maximum fidelity over speed.

Reference image

Driving audio

Prompt

Video length (seconds)

Past 5s the video is built by chaining continuation segments (one per extra 4.5s), so GPU time grows with it. Supply audio at least as long as the video — the rest is padded with silence. At 720p a PRO quota tops out near 9s; 480p covers the full range.

5 15

Resolution

720p runs on a 96 GB slice and costs well beyond a free quota — PRO only.

480p 720p

Seed

Audio preprocessing

Clean speech (fast) Isolate vocals (quality)

Acceleration

Exact 8-step is highest fidelity but needs a PRO quota; the DBCache modes fit a free one.

Exact 8-step DBCache fast DBCache faster

Output

Examples