Multitalk: Generate talking head video from image and audio

Multitalk Try it →Watch full video →AI Search · Jul 2025

Demo summary

The user demonstrates uploading a reference image and an audio clip to Multi-Talk, configuring background removal and text prompts to generate a video of a woman speaking in a park.

Step-by-step

Select a character injection mode, such as 'inject only people or objects'
Upload a reference image of your character
Toggle 'remove background' if you want to place the character in a new setting
Upload the audio clip for the character to speak
Enter a text prompt describing the final scene and background
Select the aspect ratio and resolution
Calculate and set the number of frames based on audio duration (seconds multiplied by 25)
Adjust inference steps and guidance scales if needed, then click Generate

Options

Generate video using only a text description without a reference image
Keep the original background from the reference image
Activate T-cache or Meg-cache to speed up generation by up to 2.5x
Use temporal or spatial upsamplers for higher quality output

Watch out for

The first run requires downloading large files, including a 15GB text-video fusion model
Lowering inference steps to speed up generation will sacrifice video quality
T-cache and Meg-cache speed up generation by skipping steps, which may impact quality

Tips

Upload a reference image rather than using text-only for more control over the character
If the lip-sync is inaccurate, set the audio guidance to a higher value
Leave advanced settings at default as they usually work fine without adjustment
Use a GPU with sufficient VRAM (e.g., 16GB) to avoid needing speed-optimization caches

Highlights

“usually everything just works fine without you needing to change any of these settings”

All demos from “Make AI videos with talking + pose + reference control. MultiTalk & VACE tutorial”

5:271:27Overview of the Wan2GP interface for Multi-TalkThe creator walks through the Wan2GP Gradio interface, explaining how to select the Multi-Talk model and the specific 'Vase Multi-Talk Fusion X' version for better performance on low VRAM.Multitalk· AI Animation Generator
8:224:37Generate talking head video from image and audioCurrentThe user demonstrates uploading a reference image and an audio clip to Multi-Talk, configuring background removal and text prompts to generate a video of a woman speaking in a park.Multitalk· AI Avatar Video Generator
13:571:32Simulate angry expressions with Multi-TalkThe demo shows how to use an angry reference image and matching audio to generate a highly expressive video that captures the pitch and intensity of the speaker's anger.Multitalk· AI Lip Sync Generator
15:291:10Animate sad emotions and cryingThe creator demonstrates Multi-Talk's ability to handle complex emotions by animating a sad character who pauses and breathes in sync with a crying audio track.Multitalk· AI Lip Sync Generator
17:441:28Lip-syncing anime charactersA demonstration of applying Japanese audio to an anime still image, showing how the tool handles non-human characters and different languages.Multitalk· AI Lip Sync Generator
19:393:03Animate multiple speakers in a podcast sceneThe video shows how to configure Multi-Talk for two speakers by uploading an image of two people and two sequential audio clips, assigning voices based on their position in the frame.Multitalk· AI Avatar Video Generator
22:173:21Parallel multi-speaker animationThe user demonstrates a more advanced multi-speaker setup where two audio tracks are played in parallel to animate a conversation between two people in a single reference image.Multitalk· AI Avatar Video Generator
26:322:34Transfer human motion with VACE and Multi-TalkThe demo shows how to use a control video of a person dancing to drive the body movements of a reference image while simultaneously applying a Spanish lip-sync track.Multitalk· Video to Video
Watch “Make AI videos with talking + pose + reference control. MultiTalk & VACE tutorial” →

AI Avatar Video Generator

8:224:37Generate talking head video from image and audioCurrentThe user demonstrates uploading a reference image and an audio clip to Multi-Talk, configuring background removal and text prompts to generate a video of a woman speaking in a park.AI Search
19:393:03Animate multiple speakers in a podcast sceneThe video shows how to configure Multi-Talk for two speakers by uploading an image of two people and two sequential audio clips, assigning voices based on their position in the frame.AI Search
22:173:21Parallel multi-speaker animationThe user demonstrates a more advanced multi-speaker setup where two audio tracks are played in parallel to animate a conversation between two people in a single reference image.AI Search
0:500:47Multi-person conversational video generationMultiTalk is shown animating a group image where two separate people interact and respond to each other using different audio tracks.NadimExplainsAI