MultiTalk
Audio-driven model that generates conversational video of several people from multi-stream audio and a reference image.
Generate a conversation between several people, each lip-synced to their own audio.
MultiTalk is a framework from MeiGen-AI for audio-driven multi-person conversational video. Given separate audio streams, a reference image and a prompt, it renders people interacting with matching lip motion. It suits developers with a strong GPU.
What it is
The README describes multi-stream audio input plus a prompt, with multi-GPU inference, TeaCache, APG and a low-VRAM mode that runs 480P on one RTX 4090. The current code supports 480P only on a single GPU, and 720P needs multiple GPUs.
Who it's for
Suits researchers and developers building multi-speaker clips such as interviews. It does not suit users wanting a studio, templates or built-in voices.
Key features
Multi-stream audio
Separate audio tracks drive each person.
Prompt-guided interaction
A text prompt describes how the people interact.
Low-VRAM mode
480P on a single RTX 4090.
Multi-GPU inference
torchrun scripts spread work across GPUs.
TeaCache and APG
Acceleration and artifact-reduction options.
ComfyUI and Wan2GP
Community integrations reach 8 GB VRAM.
Up and running in 3 steps
Our honest verdict
What's great
- Apache-2.0 licence on the code.
- Handles two or more speakers in one scene.
- Documented low-VRAM and multi-GPU paths.
- Community integrations for ComfyUI and 8 GB cards.
Where it falls short
- Needs an NVIDIA GPU, PyTorch 2.4.1 with CUDA 12.1 and flash-attn.
- The current code only does 480P; 720P needs multiple GPUs.
- No voices, editor or text-to-speech.
- Base model weights must be downloaded separately and are large.
Pricing & license
Free under Apache-2.0. The README lists no hosted or paid plan, so the only cost is your own or rented GPU.
Open remote AI-training roles
768 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Can MultiTalk make more than one person speak?
Yes. It takes a multi-stream audio input and drives each person with their own audio.
Does it run on one GPU?
Yes at 480P: the README says low-VRAM mode enables 480P on a single RTX 4090. 720P needs several GPUs.
Is MultiTalk free for commercial use?
The code is Apache-2.0; check the licence of the base model weights it downloads.
