A scene with two people trading lines is a common failure point for single-character animation controls. Runway’s own multi-character tutorial says Act-Two takes a single-character input and builds a conversation through separate captures, generated segments and local video editing. The editorial challenge is matching performances, eye lines and cut points, not asking one model to animate the whole scene at once.
Plan the dialogue before generating
Write the scene as separate lines with timing, pauses and reactions. Record each performance with room to breathe around the line and keep each relevant generation within Act-Two’s documented 30-second maximum. A solo actor can play dialogue back while recording another character, although playback spill might require cleanup later. For multilingual dialogue, preserve the timed reference and obtain proper rights for voices before moving clips into a lip-sync or voice pipeline. AI 05 owns the underlying dubbing and synthesis products.
Separate the characters and shots
Prepare each character’s visual reference and choose whether that reference is a still or a video plate. A still may permit gesture-driven body motion, whereas a video reference retains existing framing and environment motion. Render each speaker independently. For a shot-reverse-shot sequence, this is often simpler than attempting a simultaneous two-character plate; if both must appear together, plan a compositor and consistent negative space for overlays and masks.
Continuity is an editorial task
Record master decisions: costume, focal length impression, lighting direction, camera height and room geometry. Use a timecode or an explicit slate for your raw performances. After generating, assemble a rough cut with the original dialogue audio before spending further credits. Check glance direction, apparent speaker-to-speaker distance and whether each response starts with believable listening behavior. A seamless final cut can be impossible if each independently generated shot invents a different room.
Limitations and verification
Runway’s tutorial is a vendor-described workflow, not an independent validation of multi-person identity fidelity. It requires access to Act-Two, Gen-4 Image, Gen-4 Video and a separate video editor. Availability is product-, tier- and date-specific, so check current model access rather than inheriting a historical tutorial’s prerequisites. Record any additional input image-generation cost separately from performance-capture charges; the number of rejected candidates dominates some productions.
Alternatives and approvals
A traditional live-action composite or puppeted 2D character pipeline provides stronger continuity when absolute control is required. A simple animated dialogue can also use isolated reaction images and standard cuts. Whatever the pipeline, secure performer, visual likeness, footage and soundtrack permissions and approve the finished scene for misleading identity cues. The relevant success criterion is audience comprehension and believable turn-taking, not simply the existence of two speaking faces on screen.
Worked editorial scenario
Imagine a scene where a medic delivers bad news to a colleague. Instead of generating one broad two-person shot, record each actor’s lines and reactions, create separate character shots, and cut on the colleague’s reaction. Maintain matched light direction, camera height and sound bed. The scene works only if the listener appears engaged before speaking, so include a short listening beat. If the assembled scene drifts geographically or emotionally, fix the shot plan before requesting more renders. This workflow is a useful editorial technique even when a different model is ultimately used.
Sources & Further Reading
- Runway and official project documentation — actdialog (first-party source; checked 2026-10-25)
- Runway and official project documentation — acttwo (first-party source; checked 2026-10-25)
Product availability, plans and rights should be rechecked before publication. No hands-on test, independent benchmark or rights-cleared third-party media is claimed in this article.