React models

react-1 introduces the first performance control primitive for video editing. As an expressive AI talking head model, react-1 can synchronize the lip movements, facial expressions, and head movements to match a target audio, while following an emotional prompt. We describe a couple of workflows below that are only possible using react-1.

Key Features

• Model Modes: You can choose which facial region to edit: just the mouth, or the facial expressions, or even the head movements
• Expressive lipsync: react-1 operates on a much larger facial region, giving you the the most expressive mouth movements that match the speech
• Facial Expressions: facial expressions rewritten using an emotion prompt. and every tiny micro-expressions are perfectly in sync with the speech.
• Head Movements: react-1 can also synchronize your head movements to match the pacing, prosody, and intonation of the new dialogue.

Model modes

The model can be controlled with three modes of operation: lips, face, and head.

This allows you to specify the spatial region you want to edit. You can only opt for lipsync, or additionally choose for facial expressions or head movements as well. The default is face.

Emotion prompts

You can guide the facial expressions by specifying the emotion prompts. You can also choose not to specify one, in which case, the model will follow the emotional context of the input video. Please see usage below for more details.

Usage

react-1 is available through the same /v2/generate API endpoint used for standard lipsync, with additional parameters to control the emotional and movement effects.

1
from sync import Sync
2
from sync.common import Audio, Video, GenerationOptions
3
4
sync = Sync()
5
6
response = sync.generations.create(
7
    input=[
8
        Video(url="https://assets.sync.so/docs/example-video.mp4"),
9
        Audio(url="https://assets.sync.so/docs/example-audio.wav")
10
    ],
11
    model="react-1",
12
    options=GenerationOptions(
13
        prompt="happy",
14
        model_mode="face"
15
    )
16
)

Best Practices

Current Limitations

The following features are not yet supported for react-1:

• Input Duration: react-1 supports inputs up to 15 seconds in duration. For longer content, consider breaking your video into segments.
• Segments: Multi-segment generation with different audio inputs is not available. Process each segment separately if needed.
• Speaker Selection: The active_speaker_detection option is not supported, including both automatic detection (auto_detect) and manual selection via bounding box or frame number. Ensure your input video contains a single, clearly visible speaker.
• Occlusion detection: The occlusion_detection_enabled option for handling partially hidden faces is not available for react-1.

When to use react-1

react-1 is ideal for:

• Short-form content (≤ 15 seconds) requiring emotional expressions
• Videos where natural head movements enhance the result
• Content that benefits from emotion-aware facial expressions
• Projects where you want more dynamic and expressive lipsync results

For longer content (> 15 seconds) or when you only need standard lipsync, consider using lipsync-2 or lipsync-2-pro instead.

Pricing

react-1 is available on paid subscription plans (Creator and above). Pricing is usage-based at 25 fps and differs from lipsync models — each Sync Labs model has its own per-second rate.

react-1 is the most expressive model and is priced higher than lipsync models. See the pricing page for a full comparison.

FAQs