skallagrimr / latentsync

LatentSync: generate high-quality lip sync animations from images and videos

Cold

Public
179 runs
A100 (80GB)
GitHub
Weights
License

Run with an API

Playground API Examples README Versions

Input

Run this model in Node.js with one line of code:

npx create-replicate --model=skallagrimr/latentsync

or set up a project from scratch

Install Replicate’s Node.js client library:

npm install replicate

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Import and set up the client:

import Replicate from "replicate";

const replicate = new Replicate({
  auth: process.env.REPLICATE_API_TOKEN,
});

Run skallagrimr/latentsync using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

const output = await replicate.run(
  "skallagrimr/latentsync:e8eb9aa4503dd2db8e84d0308349b5f43bda29bdb5643b7fb876583b15176036",
  {
    input: {
      seed: 0,
      guidance_scale: 1
    }
  }
);

// To access the file URL:
console.log(output.url()); //=> "http://example.com"

// To write the file to disk:
fs.writeFile("my-image.png", output);

To learn more, take a look at the guide on getting started with Node.js.

Install Replicate’s Python client library:

pip install replicate

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Import the client:

import replicate

Run skallagrimr/latentsync using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

output = replicate.run(
    "skallagrimr/latentsync:e8eb9aa4503dd2db8e84d0308349b5f43bda29bdb5643b7fb876583b15176036",
    input={
        "seed": 0,
        "guidance_scale": 1
    }
)
print(output)

To learn more, take a look at the guide on getting started with Python.

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Run skallagrimr/latentsync using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

curl -s -X POST \
  -H "Authorization: Bearer $REPLICATE_API_TOKEN" \
  -H "Content-Type: application/json" \
  -H "Prefer: wait" \
  -d $'{
    "version": "e8eb9aa4503dd2db8e84d0308349b5f43bda29bdb5643b7fb876583b15176036",
    "input": {
      "seed": 0,
      "guidance_scale": 1
    }
  }' \
  https://api.replicate.com/v1/predictions

To learn more, take a look at Replicate’s HTTP API reference docs.

You can run this model locally using Cog. First, install Cog:

brew install cog

If you don’t have Homebrew, there are other installation options available.

Run this to download the model and run it in your local environment:

cog predict r8.im/skallagrimr/latentsync@sha256:e8eb9aa4503dd2db8e84d0308349b5f43bda29bdb5643b7fb876583b15176036 \
  -i 'seed=0' \
  -i 'guidance_scale=1'

To learn more, take a look at the Cog documentation.

Run this to download the model and run it in your local environment:

docker run -d -p 5000:5000 --gpus=all r8.im/skallagrimr/latentsync@sha256:e8eb9aa4503dd2db8e84d0308349b5f43bda29bdb5643b7fb876583b15176036
curl -s -X POST \
  -H "Content-Type: application/json" \
  -d $'{
    "input": {
      "seed": 0,
      "guidance_scale": 1
    }
  }' \
  http://localhost:5000/predictions

To learn more, take a look at the Cog documentation.

Output

No output yet! Press "Submit" to start a prediction.

Run time and cost

This model runs on Nvidia A100 (80GB) GPU hardware. We don't yet have enough runs of this model to provide performance information.

Readme

Unofficial Implementation of bytedance/LatentSync

This is an unofficial implementation of bytedance/LatentSync. This version has been improved to support both images and videos as input for lip-syncing.

LatentSync: Audio Conditioned Latent Diffusion Models for Lip Sync

We present LatentSync, an end-to-end lip sync framework based on audio-conditioned latent diffusion models without any intermediate motion representation. This approach diverges from previous diffusion-based lip sync methods that rely on pixel space diffusion or two-stage generation.

Our framework leverages the powerful capabilities of Stable Diffusion to directly model complex audio-visual correlations. However, we found that diffusion-based lip sync methods tend to exhibit inferior temporal consistency due to inconsistencies in the diffusion process across different frames.

To address this, we propose Temporal REPresentation Alignment (TREPA) to enhance temporal consistency while preserving lip-sync accuracy. TREPA utilizes temporal representations extracted by large-scale self-supervised video models to align the generated frames with the ground truth frames.

LatentSync

Framework

LatentSync utilizes Whisper to convert melspectrograms into audio embeddings, which are then integrated into the U-Net via cross-attention layers.

The reference and masked frames are channel-wise concatenated with noised latents as input to U-Net.
During training, we use a one-step method to estimate clean latents from predicted noises.
These latents are then decoded to obtain the estimated clean frames.
TREPA, LPIPS, and SyncNet losses are applied in the pixel space to improve performance.