Act_two is Runway’s next-generation AI performance capture and character animation tool: it ingests a short driving performance (a webcam or phone video of someone acting a scene) plus a character reference (image or video) and generates an animated character performance that transfers body, facial expression and hand motion to the character. Act-Two is offered inside Runway’s web product and as a model available through API ecosystem.
Introduction to the Act-Two model
act_two is designed to democratize high-fidelity animation by turning short “driving performance” clips into fully animated character sequences. The model focuses on expressive fidelity — transferring facial micro-expressions, lip and mouth motion, finger/hand gestures and full-body posture — while also adding plausible environmental motion when the character input is an image. act_two is positioned as an evolution of Runway’s earlier Act models and is tightly integrated into Runway’s Gen-4 video toolset and API.
what act-two does
- Full-body performance transfer: maps head, face, torso and hands from a single driving video to a character reference.
- Character input flexibility: accepts either a character image or a reference video as target.
- Gesture control: when using a character image, you can drive hand/body gestures through the driving clip and adjust gesture influence.
- Automatic environmental motion: adds subtle background/environment motion for image-based characters to avoid “floating” results.
Technical details & task constraints
Inputs
- Driving performance: a video that contains the acting performance (movement, gestures, audio).
- Character reference: either a still image or a video of the character you want animated. ([Runway][1])
Outputs & formats
Supported aspect ratios and resolutions include 1280×720 (16:9), 720×1280 (9:16), 960×960 (1:1) and a small set of other presets; 24 FPS is the standard frame rate for outputs. There is auto-cropping to match aspect ratio targets.
Processing notes: best results when driving performance and character face the same general direction and occupy similar screen space; inputs with extreme perspective mismatches or very distant/low-resolution subjects can produce degraded results. Runway’s moderation and content filters apply to uploaded assets; tasks may be rejected if content violates policy.
Limitations and known failure modes
- Short-duration focus: act_two is optimized for short clips (3s minimum; typical workflows use under 30s). For feature-length mocap you’ll still need traditional capture or chunked workflows.
- Moderation / content safety: Runway’s moderation can block or fail tasks for flagged inputs; accounts with excessive moderation failures may be rate-limited or suspended. Plan content policy compliance into automation.
- Edge cases: extremely complex multi-person performances, highly occluded hands, or ultra-stylized references can produce artifacts (jitter, incorrect hand poses, or expression mismatch). Manual cleanup or hybrid pipelines (lighthand rotoscoping / keyframe repair) may still be required.
- Not a full motion-capture replacement in all cases: while Act-Two can replace many traditional setups for short scenes and prototyping, high-end film/CGI pipelines that require sub-millimeter accuracy, multiple actors interacting physically, or on-set timing sync will still rely on marker systems / performance capture stages.
Typical use cases
- Virtual production & previs — rapid blocking and acting tests without a mocap stage.
- Indie game & animation prototyping — fast character motion generation for short scenes.
- Commercials & social content — produce character spots and animated talent cheaply and quickly.
- VFX insertions & motion replacement — augment existing footage by driving a stylized character from an actor’s take.
Comparison with other current solutions
act_two vs Pika Labs / Kaiber / Sora (high level)
- Act-Two (Runway): excels at performance fidelity for characters (head/face/body/hands) with a single driving clip paradigm and straightforward API integration for short videos; predictable credit pricing for seconds of output.
- Pika Labs: often highlighted for flexible prompt-to-video and style transfer; may focus more on general video generation and stylization rather than targeted performance transfer.
- Kaiber: strong at style transforms, music-driven visuals, and general scene generation, but not necessarily as specialized in per-character mocap fidelity.
- Sora (and similar premium VFX pipelines): oriented toward cinematic quality and extended scene generation; stronger for long sequences and film VFX but more resource-intensive and possibly less accessible for rapid prototypes.
How to call Act-Two API from CometAPI
Required Steps
- Log in to cometapi.com. If you are not our user yet, please register first
- Get the access credential API key of the interface. Click “Add Token” at the API token in the personal center, get the token key: sk-xxxxx and submit.
- Get the url of this site: https://api.cometapi.com/
Use Method
- Select the “
act_two - Replace <YOUR_API_KEY> with your actual CometAPI key from your account.
- Insert your question or request into the content field—this is what the model will respond to.
- . Process the API response to get the generated answer.
CometAPI provides a fully compatible REST API—for seamless migration. Key details to API doc:
- Endpoint:
https://api.cometapi.com/runwayml/v1/character_performance - Model Parameter:
act_two - Authentication:
Bearer YOUR_CometAPI_API_KEY - Content-Type:
application/json.
curl --location --request POST 'https://api.cometapi.com/runwayml/v1/character_performance' \
--header 'X-Runway-Version: 2024-11-06' \
--header 'Authorization: {{api-key}}' \
--header 'Content-Type: application/json' \
--data-raw '{
"character": {
"type": "video",
"uri": "https://filesystem.site/cdn/20250818/wAKbHUoj5EHyqZvEdJbFXn10wXBMUn.mp4"
},
"reference": {
"type": "video",
"uri": "https://filesystem.site/cdn/20250818/wAKbHUoj5EHyqZvEdJbFXn10wXBMUn.mp4"
},
"bodyControl": true,
"expressionIntensity": 3,
"seed": 4294967295,
"model": "act_two",
"ratio": "1280:720",
"contentModeration": {
"publicFigureThreshold": "auto"
}
}'