A local theater wants a 20-second scene in which a blue felt puppet asks a question and an orange felt puppet replies. The performers recorded the voices with permission, and the characters are fictional. Runway’s official multi-character Act-Two tutorial describes using separate single-character generations, then compositing them over a common base shot. This workflow is not a one-click two-character animation feature. The panels here illustrate likely editorial risks, not outputs measured from an actual Runway account.
Plan a two-character scene around separate performances
Write a compact dialogue with a speaker label on every line and a timecode outline. Record a clean performance for Puppet A and another for Puppet B, keeping consistent timing and pauses. Runway’s help describes a 30-second maximum for an Act-Two performance and recommends separate video performances for each character. Plan a brief exchange under that ceiling rather than designing a longer sequence that the editor cannot align. Keep the original performer recordings, the approved script, and any release or licensing documentation. Do not use an unconsented person’s face or voice as a shortcut for a fictional character.
Create a stable two-puppet base frame
Prepare one picture showing the blue puppet on the left and the orange puppet on the right. If using Gen-4 Image Reference, define each character’s color, silhouette, clothing, and position. Avoid visually identical puppets: distinct shapes make wrong-speaker errors easier to detect. Include sufficient spacing around arms and hands to allow later masking. Check that the prompt does not introduce human likenesses or third-party logos. If the generated frame contains an inconsistent hat or a missing arm, fix the base image before animating. A flawed source will be repeated in every subsequent composite layer.
Generate background movement before facial performance
The documented workflow uses Gen-4 Video to create a base shot with subtle shared ambient movement. Start with a short segment and inspect lighting, camera motion, and where both puppets sit in the frame. If the full dialogue is longer, Runway describes extending the base video using the prior clip’s last frame as an input. A shared frame can reduce discontinuities, but does not guarantee a seamless transition. Check for repeated background motion, reversed gestures, and brief texture changes where clips meet. Export one base master of the intended duration before processing the individual character turns.
Animate one performer at a time
Act-Two accepts single-character inputs. For the left puppet, crop the character video so only that puppet is visible, pair it with Puppet A’s approved performance, and generate its animation. Repeat with a crop of the right puppet and Puppet B’s distinct recording. Store the results with descriptive speaker names rather than ambiguous file numbers. Runway’s guide explains that sharing a common base helps both animations retain coordinated background movement. Never infer that Act-Two can natively track two faces simultaneously; separate generations and editing are central to this method.
Composite layers without double outlines
In a local editor, put the base shot on the lowest video track and the two performance outputs on separate overlay tracks. Temporarily lower each overlay’s opacity to line up eyes, puppet edges, and the background. Feather masks enough to hide harsh crop borders but not so much that the puppets appear transparent. Check the overlap between a puppet’s moving hand and the other character. A mask can look fine in a still image but show a bright edge during motion. Runway suggests using compositing controls in the editor; there is no guaranteed universal mask setting for all footage.
Audit the speaker-to-mouth correspondence
View the exchange several times with the sound on and muted. When the blue puppet talks, its mouth should move; the orange puppet should not suddenly begin mouthing the same sentence. Compare timecodes against the original recordings and note any premature jaw movements around pauses. In this hypothetical example a wrong track could make the orange puppet react to the blue puppet’s line. That is a compositing or timeline problem to investigate, not evidence of a measured model failure. Adjust track order, trims, or masks and recheck both faces at full playback speed.
Handle transitions, room sound, and captions
Align the two performer audios so dialogue turns do not overlap accidentally. Preserve intentional pauses that signal a response or reaction; removing every gap makes a puppet exchange feel unnatural. If a background soundtrack is added, use a level that keeps dialogue clear. Add accurate captions with speaker identification where confusion is likely, and confirm captions remain legible against moving puppet colors. Review every extended base-clip seam: the curtain may shift, a prop may jump, or a puppet might duplicate for one frame. Correct those issues before claiming the shot is a single continuous scene.
Keep consent, likeness and usage records
Document which actor supplied each voice and performance, what use was approved, whether additional voice transformation occurred, and what files must be deleted after delivery. Make it clear to audiences that the puppets are fictional if context could be misleading. A local theater’s costume or original puppet design may also carry intellectual property restrictions; preserve proof of permission to reproduce it. Runway’s optional Stylize Speech tool does not eliminate a performer’s rights or a need for review. Record changes that alter a voice’s identity, emotional delivery, or script meaning.
Choose a simpler alternative when needed
If you only need an instructional exchange between illustrated characters, still illustrations with separately recorded dialogue and simple mouth frames may be easier to review and maintain. A complex compositing pipeline can be excessive for a 10-second line read. Use Act-Two’s documented separate-character route when facial performance matters and an editor can inspect the seams. The diagrams in this article are original conceptual artwork. They do not imply that a finished video was rendered, downloaded, or user-tested, and results will vary by input quality and feature access.
How We Research
Our research reviewed named official vendor help resources: Runway Help — Creating Multi-Character Dialogues with Act-Two; Runway Help Center. We checked documented feature scope, entry points, limitations, plan or seat requirements, and what the vendor does and does not claim. The example organization, figures, and errors are fictional instructional scenarios, not measurements from a live account. No source datasets, customer materials, actual exports, user interviews, or product-generated outputs were available for hands-on testing. All illustrations are original diagrams, not screenshots of a vendor interface. Before publication or real-world use, an editor should confirm current documentation, privacy and accessibility policies, and review the approved original source against every significant claim.
Frequently Asked Questions
The documented approach uses separate single-character crops and performances, then combines them in a video editor. It is not a direct dual-character input workflow.
Runway’s multi-character tutorial identifies a 30-second limit for Act-Two. Confirm current limits and account entitlements before planning a production.
A shared shot helps keep backgrounds and camera motion consistent while each puppet receives separate lip and facial performance.
Misaligned or oversized overlay masks can reveal both the base puppet and its processed layer. Inspect moving edges at full playback speed.
Only with appropriate consent and rights. Keep a documented agreement for voice, performance and distribution, and avoid portraying a real person as endorsing something they did not approve.