Sound-to-Video with Wan 2.2 S2V
Upload a voice or music track and one reference image, and Wan S2V animates the subject and syncs it to the audio. Good for avatars and talking characters.
- L40S
- Avatars & lip sync
- Image to video
- Free template
How it works
The workflow uses core ComfyUI nodes and subgraphs, with no custom packages. UNETLoader loads Wan2_1-S2V-14B_fp8_e5m2_scaled_KJ, CLIPLoader loads umt5_xxl_fp8_e4m3fn_scaled, and VAELoader loads wan_2.1_vae. A lightx2v_T2V_14B_cfg_step_distill_v2 LoRA is applied through LoraLoaderModelOnly so sampling needs only a few steps.
Audio comes in through LoadAudio and is encoded with AudioEncoderLoader using wav2vec2_large_english_fp16 and AudioEncoderEncode. WanSoundImageToVideo combines the audio embeddings, your reference image and the prompt into the video latent. KSampler with ModelSamplingSD3 generates it.
For long audio, extension subgraphs continue the video in chunks, using LatentCut, LatentConcat and ImageFromBatch. A workaround in the notes doubles the first frame and removes it after decoding, because the VAE can overbake the first frame. CreateVideo and SaveVideo write the MP4.
Tips
- The notes include a table of steps and cfg for runs with and without the lightning LoRA, so check it before changing sampler settings.
- Use a clear, front-facing image for better lip movement.
- For videos longer than one chunk, add extension subgraphs rather than raising the frame count.
- Describe the setting and mood in the prompt, and let the audio drive the mouth movement.
What you can use it for
- Talking characters from a still image
- Lip-sync for voiceovers
- Music-driven animation
- Longer clips by chaining extension chunks
Run it in three steps
Open the template
Click Run on ComfyDeploy. The workflow, its custom nodes and models are set up for you on a cloud machine.
Add your inputs
Provide reference image and audio track in the playground or the ComfyUI editor.
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 workflow need?
It is set to run on an L40S GPU on ComfyDeploy. It loads Wan2_1-S2V-14B in fp8.
Can I make longer videos?
Yes. The workflow notes say to copy and paste the Video S2V Extend subgraph, and each copy extends the video by the chunk length, 77 frames by default.
How do I run it as an API?
Deploy it on ComfyDeploy and expose the image, audio and prompt inputs. The workflow uses a ComfyDeploy node for the output.
