Skip to content
ComfyDeploy
Get started

Talking Videos with Wan 2.1 MultiTalk

Feed in a voice track and a reference image or video, and this workflow generates a video where the subject speaks in sync. It suits creators making talking-head clips and character videos.

Run on ComfyDeployDownload JSON

How it works

The workflow is built on the ComfyUI-WanVideoWrapper nodes. WanVideoModelLoader loads Wan2_1-T2V-14B_fp8_e4m3fn together with the MultiTalkModelLoader weights, and two LoRA selectors add Wan2.1_T2V_14B_FusionX_LoRA and the lightx2v cfg-step distill LoRA, which allow fewer sampling steps.

Audio is loaded with LoadAudio, optionally cropped with AudioCrop, and split into voice and background with AudioSeparation. DownloadAndLoadWav2VecModel and MultiTalkWav2VecEmbeds convert the speech into conditioning for the lips and face.

Your reference image (or video, via VHS_LoadVideo) can have its background removed with BiRefNetRMBG. A VideoResolutionSelector chooses the size from the aspect ratio and quality setting (HQ, MQ or LQ). An optional LLM step, using an OpenAI provider node, turns your action prompt into a fuller description.

After decoding, the video can get color adjustment and LTXVFilmGrain, and VHS_VideoCombine muxes the audio back in.

Tips

  • Set the frame count to match your audio length, since the audio is what drives the animation.
  • The RESOLUTION option offers I2V480p, I2V720p, T2V1.3B and T2V14B, so pick a smaller one for quick tests.
  • Use a clean, front-facing portrait with the face clearly visible for the best sync.
  • The LLM prompt step needs an OpenAI key on the machine. Without it, type the full action prompt yourself.

What you can use it for

  • Talking-head clips from a still portrait
  • Character dialogue scenes with a written action prompt
  • Vertical short-form videos
  • Prototype speaking avatars for a product

Run it in three steps

  1. Open the template

    Click Run on ComfyDeploy. The workflow, its custom nodes and models are set up for you on a cloud machine.

  2. Add your inputs

    Provide reference image or video and audio clip in the playground or the ComfyUI editor.

  3. Run, share or deploy

    Run it on an L40S, share it with your team as a playground, or deploy it as an API endpoint.

Frequently asked questions

What GPU does this MultiTalk workflow need?

It is configured for an L40S GPU on ComfyDeploy. It loads the 14B Wan 2.1 model in fp8 with block swap to manage VRAM.

Which models does it use?

It uses Wan2_1-T2V-14B_fp8_e4m3fn with the WanVideo_2_1_Multitalk_14B_fp8 model, plus the FusionX and lightx2v distill LoRAs. Wav2Vec handles the audio embeddings.

How do I run it as an API?

Deploy it on ComfyDeploy and send the image or video, audio and prompt to the endpoint. The external input nodes define the API fields.

Related workflows

All Wan workflows