Updated: 2026-05

1. About This Page

This section covers techniques for rendering the same character across multiple shots, with a particular focus on the use of Act-Two (motion capture-based character animation). This is one of Runway’s greatest strengths.

2. Why Is It Difficult?

Since video-generating AI creates each shot independently, the characters’ faces and clothing will change from one shot to the next if left unchecked.

• Shot 1: Woman in a red coat
• Shot 2: Woman in a blue coat (for some reason; a different person even though the prompt is the same)

Runway solves this problem in three ways:

1. Character Reference (reference-based, lightweight)
2. Act-Two (motion capture-based, resource-intensive but powerful)
3. Aleph (a different approach that alters a person’s appearance by editing existing video footage)

3. Review of Character References

As mentioned in the References section, this is the most basic method.

1. Prepare an image of the character in full-front view, with no background, and in bright light
2. Add it to the Character Reference when generating all shots
3. Use a prompt to specify the character’s actions

The success rate is moderate. For short pieces of about 3 to 4 shots, this should be sufficient. As the piece gets longer, the alignment gradually drifts.

4. What is Act-Two?

A next-generation motion capture model released in 2025. It allows you to transfer performances captured by a webcam onto any character.

Role	Description
Driving Performance	A video of you acting in front of a webcam
Character Reference	An image (or video) of the character you want to animate
Output	A video of the character moving according to your performance

Movements captured:

• Facial expressions and mouth movements
• Head, neck, and upper body
• Hand movements (simple gestures)

5. How to Use Act-Two

5.1 Terms of Use

• Standard plan or higher (Not available on the Free plan)
• A PC with a webcam (or pre-recorded video is also acceptable)

5.2 Basic Workflow

1. Open the left navigation menu → Apps → Act-Two
2. Upload a character image to Character Reference
3. Record a video of your Driving Performance on the spot or upload a video file
4. Click Generate
5. Results will appear in a few seconds to a minute

5.3 Tips for Acting

• Stand 50–80 cm away from the camera so that your upper body is visible
• Keep your face facing slightly forward (turning your head too far to the side reduces accuracy)
• Bright lighting and a plain background are best
• Avoid exaggerated expressions; acting naturally makes it easier for the AI to replicate

6. Preparing Character Images

Checklist for character images used in Act Two:

• ✓ Full-face or three-quarter view
• ✓ Plain background (or solid color)
• ✓ Bright lighting with minimal shadows
• ✓ Face clearly visible
• ✓ Upper body visible
• ✗ Dark, side profile, or complex background

You can either generate character images in advance using Generative Images or select them from free stock images. In class, students are expected to bring their own character images.

7. Sample Workflow

Examples of creating scenes where the protagonist is surprised:

1. Prepare a character image (a woman in a red coat)
2. Open Act-Two and set it as the Character Reference
3. Stand in front of your webcam and act out a “surprised expression” for 10 seconds (open your mouth and widen your eyes)
4. Click Generate → A video will appear showing the character making the same expression

Note: You are not the woman in the red coat yourself; only her performance is transferred. Her facial features and clothing remain as shown in the character image.

8. Credit Usage

Act-Two is more resource-intensive than standard video generation. Generating a 10-second video costs approximately 100–150 cr (check the official documentation each time).

• Free: Not available (Standard or higher)
• Standard: 625 cr/month: Enough to try Act-Two 4–6 times
• Pro: 2,250 cr/month: 15–20 times

For the class, use Act-Two only for 1–2 key scenes; for everything else, the standard I2V is sufficient.

9. Shots Suitable for Act-Two and Shots Not Suitable for Act-Two

9.1 Facing

• Close-ups of characters’ faces, shots where facial expressions take center stage
• Short monologues, reactions
• Speaking directly to the camera

9.2 Not Suitable

• Intense full-body movements (such as running or dancing)
• Scenes involving multiple characters
• Shots that involve significant movement of the environment (primarily camera pans)

10. Consistency Across Multiple Shots

Combinations when using the same character in three or more shots:

Shot Type	Recommended Method
Close-up of face/focus on facial expressions	Act-Two
Full body/walking	Character Reference + I2V
Wide environmental shot	No Character Reference required; Image Reference only

You don’t have to use Act-Two for everything. Keep the overall budget down by dividing up the roles.

11. Common Mistakes and How to Handle Them

Symptom	Cause	Solution
Character face appears distorted	Insufficient character image quality	Retry using a front-facing image with no background
Performance is not transferred	Driving Performance is too dark or at an incorrect angle	Re-record
Character differs from other shots	Reference is not shared	Use the same image for all shots
Clothing color fluctuates	Character image clothing is complex	Regenerate using solid-colored clothing

12. Operational Recommendations for the Practical Training Phase

During the five group projects, each group:

1. Choose one image of the main character
2. Use that image as the Character Reference for all shots
3. Add facial expressions in Act Two for only 1–2 shots during the climax
4. For everything else, use I2V + Character Reference

This way, we can create a cohesive short story featuring the character without going over budget.

13. What’s Next

• Runway Editing — Connect generated shots on the timeline
• Audio — Add dialogue using lip-sync (when combining with Act-Two)