LeapTalk

LeapTalk

Turns a portrait photo and a speech clip into a lip-synced talking-head video, as a demo of the LeapTalk model for researchers, developers and creators.

Made by hugging-apps on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: This Space is a demo of the LeapTalk model by z-rx. Feed it one face picture and some spoken audio, and the output is a clip of that face talking in sync with the sound. The model makes each video chunk in a single network pass, and the Space runs on Hugging Face's zero-a10g hardware.

Who it suits: Researchers and developers who want to try audio-driven talking-head generation will get the most from it. Creators who want a short talking portrait from a still image can use it too.

Industry | Who uses it | How they use it | Example
Research | ML researchers | Run the LeapTalk pipeline on their own photo and audio | Checking how steady a face stays across a clip
Video production | Creators | Animate a still portrait to a speech clip | A short talking portrait
Education | Teachers | Make a presenter-style clip from a photo and a narration | A narrated explainer with a face
Software | Developers | Read the pipeline code in the repository | Seeing how the model parts connect

If you have no portrait photo or speech clip to feed it, you can skip it.

Where to use it:

• A web browser: open the Space's page on Hugging Face
• A Windows PC, a Mac or a Linux machine that has Python 3.12
• Tools that speak MCP, since Hugging Face lists the Space with an MCP tag

Use case | Who | What they get back
Animate a portrait | Creators | A lip-synced video of the face speaking your clip
Test the model | Researchers | A short clip to judge how well the face holds steady
Study the pipeline | Developers | Inference code and a list of the weights it uses

How to set it up: Online use needs nothing. For your own copy you need Python 3.12 and pip. The source names no GPU requirement for a local copy, but the hosted Space runs on zero-a10g.

• Python 3.12 with pip, on Windows, macOS or Linux
• Gradio 6.23.1, the version this project pins
• The Python packages listed in requirements.txt

1. Install the Hugging Face hub package.
2. Download the Space's files into a folder.
3. Move into that folder.
4. Install Gradio 6.23.1 and the packages in requirements.txt.
5. Run app.py.

pip install -U huggingface_hub
hf download hugging-apps/leaptalk-talking-head --repo-type space --local-dir leaptalk-talking-head
cd leaptalk-talking-head
pip install gradio==6.23.1
pip install -r requirements.txt
python app.py

How to use it: Give the app a portrait photo and a speech clip, then start the generation. The video comes out at 512x512 and 25 frames per second, built chunk by chunk. The repository's examples folder holds sample assets to try first.

python app.py

Example: You upload a portrait photo and a clip of a few seconds of narration. The Space returns a video of that face speaking the clip, with the lips following the sound.

For newcomers and experts: A talking-head video shows a still face speaking along with an audio clip. To try it, open the Space, use one of the example assets, and run it. Use only faces and voices you have the right to use.

Its method, named Bridge Forcing, frames the task as a Brownian-bridge transport from data to data. The distilled model makes each chunk with one network evaluation and keeps identity and style stable over long rollouts. Chunks hold 33 frames, with 5 overlapping motion frames. The base video model is SoulX-FlashHead 1.3B, the audio encoder is wav2vec2-base-960h, and the multi-GPU path is disabled here.

You give a portrait photo and a speech clip
wav2vec2 encodes the audio
LeapTalk makes each video chunk in one step
Chunks join into a 512x512, 25 fps video

Good to know:

• License: apache-2, as listed on Hugging Face
• Cost to run: none named. The hosted Space uses Hugging Face's zero-a10g hardware
• Data it reaches: the photo and audio you give it go to the Space's backend, and the source names no account or clipboard use
• Use only photos and voices you have the right to use
• Not for: multi-GPU runs, since that path is disabled in this Space

On Hugging Face, 13 November 2026

• Likes: 9
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/hugging-apps/leaptalk-talking-head
• Files: https://huggingface.co/spaces/hugging-apps/leaptalk-talking-head/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.