Character-driven Runway video: drive a character with a reference image and motion direction for consistent, animated characters.
This model gets stronger as the shot becomes more explicit. Give it a subject, a move, a frame, and a mood so the output feels directed instead of guessed.
Best results start with a directed prompt or a strong first frame.
Gen-4 Act-Two on Pixio is Runway’s character-driven video model. You provide a reference image of a character (or person) and a text prompt that describes how they should move or act; the model generates video that keeps the character consistent across the clip. Use it when you need a specific character or spokesperson to perform an action or deliver a scene—talking, gesturing, or moving—without character drift.
Gen-4 Act-Two on Pixio is Runway’s character-driven video model. You provide a reference image of a character (or person) and a text prompt that describes how they should move or act; the model generates video that keeps the character consistent across the clip. Use it when you need a specific character or spokesperson to perform an action or deliver a scene—talking, gesturing, or moving—without character drift.
| Mode | Input | Best for |
|---|---|---|
| Character to Video | One character reference image + prompt | Character performs the described action; consistency from reference |
| Option | Values | Notes |
|---|---|---|
| Reference | One image (character/person) | Clear face and body; front or three-quarter view works best |
| Duration | Depends on backend | Check Pixio for limits |
| Prompt | Action, expression, camera | Describe what the character does, not their appearance |
Credits depend on duration and plan; check the model card in Pixio for current rates.
Act-Two is built for one character in, one character out: the reference image defines who we see, and the prompt defines what they do. The model keeps the character’s look consistent while animating motion and expression. Use it for spokesperson clips, character moments, or when you need a specific person/character to perform an action. For talking head with lip-sync and voice, combine with Runway Act-One or other voice/lip-sync tools.
Describe the character’s action and expression, not their look. The reference image defines appearance.
Keep one clear action per prompt.
| Scenario | Best choice |
|---|---|
| Character-driven clip from one reference | Gen-4 Act-Two |
| Talking head + lip-sync + voice | Fabric, Character 3, OmniHuman, or Act-One + voice |
| General image-to-video (no character focus) | Gen-4 (Image to Video) or Seedance 2 Pro |
| Restyle existing video | Gen-4 Aleph |
Start with a strong first frame when consistency matters more than surprise.
Keep each prompt focused on one primary motion direction.
Use shorter runs for iteration, then scale up for finals.
For narratives, structure the idea as Shot 1 / Shot 2 / Shot 3 instead of one flat blob.
A strong video prompt gives the scene a subject, a move, camera behavior, and a mood to hold onto.
Start from language and push for camera intent, pacing, atmosphere, and shot design in one move.
Start from a frame or reference when consistency matters more than improvisation.
Continue or refine the clip without throwing away the visual language you already established.
Gen-4 Act-Two works well when the prompt needs motion, framing, and visual direction, not just subject matter.
Use it for sequences that need a strong first frame, continuity, or a clearly controlled camera idea.
Treat each generation like a shot brief instead of a loose caption to get more cinematic outputs.
Start with either a directed text brief or a strong frame, depending on how locked the look already is.
Write the motion like a director: subject, action, camera behavior, environment, lighting, and tone.
Iterate fast on shorter runs, then move to stronger finals once the rhythm feels right.
Use it to build a stronger first frame, then hand that frame to the video model for motion and continuity.
Pair it with frame extraction, merge tools, or image prep so the motion workflow stays clean end to end.
