AI animation tools for two characters interacting in one scene

Choose the workflow around the interaction

Start with Dreamina for reference-led two-character scene concepts. Choose a performance-and-compositing route when each character needs its own acting, or a rigged scene when exact contact matters more than a fast generated shot.

Use the existing material to narrow the choice:

• Two character images and a short scene idea: build a clear shared composition in Dreamina, Kling or Vidu, then generate a brief exchange.

• Two recorded acting performances: Runway documents a common-base, separate-performance and compositing workflow.

• An existing dance or interaction video: Viggle's multi-character replacement route can reuse the source movement.

• A cartoon conversation with prepared artwork: consider Anijam's advertised separate-character dialogue features or Adobe Character Animator's puppet tracks.

• Precise spatial staging: consider Autodesk Flow Studio's 3D scene controls and plan time for animation cleanup.

A longer story adds another requirement: retain the same cast and scene decisions across shots. LTX Studio's reusable Character Elements address that organization job, while the individual shot still needs to work.

Two characters in frame is only the starting point

Judge the interaction by action ownership, reactions and continuity: which character starts, which responds, who holds the prop and whether their identities survive the movement. Shared screen space alone does not answer those questions.

A useful first scene has a readable beginning, a short exchange and a clear ending. For example, the character on the left asks a question; the character on the right looks back and replies. A shared reference frame can establish position and appearance, but the timing and speaker assignment still need direction.

Physical contact raises the difficulty. A handshake needs two hands to reach the same place at the same time. Passing a cup also requires the cup to belong to one person before the exchange and the other afterward. Treat these as production requirements when choosing the route, rather than adding them casually to a general motion prompt.

Compare the production routes

Choose by the inputs and control you need, rather than treating every multi-character feature as equivalent. A saved character, a reference image and an independently recorded performance solve different problems.

1. Dreamina: develop the shared scene and action beat

Dreamina is a useful starting point for building a shared two-character composition and turning it into a short interaction concept. With a supported Seedance workflow, assign each person a distinct reference and describe the order of actions.

Establish the two identities before adding complex motion. A person in a blue jacket on the left and a person in an ochre jacket on the right are easier to direct than two nearly identical figures called “the people.” Supply clear references where supported and make the intended eye lines visible in the shared frame.

Ask for one brief exchange first: the blue-jacketed person nods, then the person in ochre replies with a wave. Keep the camera restrained while judging who did what. Once the action is readable, refine the composition or add a new scene. Use the selected model's actual reference and editing controls; additional references do not act like independent animation tracks.

For commercial work, confirm the permission that applies to the market, model and plan you use. Keep voice references and performer recordings within the permissions you have for those people.

A simple two-person reaction example

A short Dreamina example shows a blue-clothed adult making a small head dip before the adult in ochre raises a hand. It is a useful starting level of movement for a two-person scene: two distinct people, one brief action each and a simple background.

This paid Dreamina Seedance 2.5 example produces a roughly five-second, 720p vertical clip. The original image is 16:9 and the output is 9:16, so the framing narrows around the pair.

Keep a first reaction beat this simple before adding contact, crossing bodies or an object exchange. If your starting material is an existing group image, the guide to animating group photos explains the additional person-by-person checks.

2. Kling AI: generate interaction with character references

Kling VIDEO 3.0 combines character references with native audio and character-specific dialogue instructions. Use distinct subject references and explicit action assignments when you want the two characters generated within the same shot.

Prepare an Element for each character rather than describing both through one vague identity. Create each Element using its available character-reference options, then bind it to the intended subject in the video. Write the lines next to the intended speaker and specify whether you want a continuous shot or a multi-shot exchange.

For physical action, start with the beginning and ending arrangement clearly defined. A short reach-and-reaction is easier to inspect than a prompt that combines a hug, turn, walk and camera orbit. Review the contact itself before approving the take; the presence of two recognizable characters does not establish that their hands or bodies met correctly.

3. Runway: direct each performance, then combine them

Runway's documented multi-character dialogue route builds a common scene, applies one performance at a time with Act-Two, then composites the results. It gives each character a separate acting pass while sharing the same underlying scene motion.

Choose this route when the timing and expression of each speaker matter enough to justify recording and editing. You need access to the relevant image/video tools and Act-Two, which currently requires Standard or higher, plus a local editor and a way to record the performances.

The six-step structure below follows Runway's documented production method. Its common base is especially useful for a conversation in which the camera, lighting and environment must agree across the two acting passes.

Record the dialogue and reactions first

Record a separate performance for each character and establish their timing before generating the scene. Runway's guide keeps the dialogue within Act-Two's 30-second maximum and uses the longer performance to plan the base-video length.

Working alone, record the first role and play it back while performing the second. With two performers, record the exchange together and prepare the separate character recordings and audio afterward. Keep the pauses and reactions that make the conversation feel connected.

Use clear audio and a continuous performance with the face visible. Frame the performer no farther away than waist-up, and avoid cuts within the take. If body gestures will be transferred through the image-input branch, keep the hands visible from the start and begin in a pose compatible with the character.

Build one image with both characters

Create or supply a shared image that places both characters in their intended positions. Runway's guide uses Gen-4 Image References and named references to establish that common scene before motion generation.

For example, assign the driver and passenger to their named references rather than asking for two people in a car. Establish who faces whom, the distance between them and which objects belong to each. The guide also suggests a sketch for composition control when the arrangement is hard to describe.

Check that both faces and the required action areas are visible. This image becomes the foundation for the scene, so fix a confused seat assignment or obstructed face before spending time on the performance passes.

Generate and inspect the common base video

Use the shared image to create the ambient base video, then inspect both characters and the background before transferring dialogue. The guide starts with a 10-second Gen-4 Video clip; the base supplies the scene motion for both later passes.

Keep this motion appropriate to the exchange. Subtle head, hair and background movement may be enough for a seated conversation. A large camera move or a character walking behind the other can make the later crop and composite more difficult.

Review the whole base before proceeding. If one face changes or the scene movement is already distracting, the acting pass will not solve that underlying problem.

Match the base to the performance length

If the dialogue runs longer than the base, extend the scene to a similar duration before the Act-Two passes. Runway's guide reuses a final frame as the next clip's first frame and aligns the shared frame in an editor.

Use the final-frame action, Use frame → Input for video, to start the next 10-second segment. For a longer exchange, repeat with the next segment's final frame. Align the matching frames in the editor and trim a duplicate frame or blend a join only if the motion benefits from it.

Inspect the join in playback. A shared frame helps alignment, but it does not settle whether the surrounding movement feels continuous. Export the assembled base once the length and joins work. A character video shorter than the driving performance can otherwise repeat with a reversed “boomerang” effect in Act-Two.

Apply one Act-Two performance to each character

Crop the same base video so only the first character is visible, pair it with that character's performance and generate. Repeat for the second character with the second recording, keeping both passes tied to the same base timeline.

This is the key production distinction: two acting passes are built from one scene. Keep the crop and timing notes with each output so the layers can be placed back over the matching characters without guessing.

With a character-video input, Act-Two retains the source body, camera and environment motion while changing the facial performance. A raised arm in the driving recording does not become an independently directed body gesture through this route. If the body action needs to change, plan the image-input gesture branch or rebuild the base action before compositing.

At the end of the two passes, retain three videos: the original common base, Character A's performance and Character B's performance. Check each output before treating it as a finished dialogue layer.

Composite both passes over the common base

Place the common base on the lower track and align the two Act-Two outputs above their matching characters. Check timing, scale and edges while viewing the whole shot, then mix the dialogue and intended scene sound.

Temporarily lower an overlay's opacity to see the matching base character underneath. Align the image and restore its opacity. Use restrained edge feathering where needed, then inspect the result during movement rather than judging one paused frame.

Pay special attention when one person crosses in front of the other. A soft edge cannot fix the wrong foreground order or a hand that should pass behind a body. Those moments may require a more precise mask, a different shot or a revised base action.

When you need gesture control instead

Act-Two's character-image route supports gesture control, while the character-video route retains the original body and scene motion and controls facial performance. Separate image-based characters can be keyed and composited, but the extra control also adds cleanup work.

Runway's advanced alternative starts with separate character images on flat green backgrounds. Apply the individual performances, remove the green backgrounds in an editor and assemble the characters over the intended scene. Match their size, eye lines, light and depth order.

Start with the default facial expressiveness setting before increasing it. Higher values transfer more facial motion but can introduce artifacts; lower values can be useful when identity needs more restraint. If you use a later environmental edit through Aleph, recheck the facial performance afterward—Aleph is not a lip-sync model, and another Act-Two pass may be needed.

4. Vidu: combine references for recurring characters

Vidu's Reference to Video route is useful when you already have separate character and scene references. Name each reference's role in the prompt, then describe how the characters approach, react or move within the shared scene.

The current reference guide describes uploading one to seven reference images. Spend those references on clear identities and the scene information the shot actually needs. Two character references and a location can be more useful than several near-duplicate images that leave the action ambiguous.

Use names or tags consistently: the cowboy remains seated while the woman enters from the right, approaches and stops beside him. Review position, appearance and the reaction before adding another action. This is reference-guided generation; it does not give the two characters separately recorded acting tracks.

5. Google Flow: choose the model as well as the scene

Google Flow provides frames and ingredients for building shared scenes, but their availability depends on the selected video model. Veo and Gemini Omni Flash are separate choices within that workflow, with different reference and editing support.

Choose Video, then select the model and add frames or ingredients supported by it. Use named character references or describe each ingredient's role, keeping the two people distinct from the setting. Clean character references with compatible lighting make the intended scene easier to interpret.

The current model guide lists Ingredients/References to Video for Veo 3.1 Lite and Fast at eight seconds, while Veo 3.1 Quality does not support that feature. Gemini Omni Flash 1.1 has its own frames, ingredients and editing capabilities. Check the active model before building the shot around a particular control.

For a speaking scene, write who says each line and inspect the assigned face. Flow's voice-reference controls are also model- and mode-specific; do not assume changing the model preserves every voice or reference option.

6. Anijam: consider separate-character dialogue

Anijam specifically advertises up to four independently animated characters in one scene, each with its own voice track, mouth sequence and expression. That makes it a relevant shortlist option for cartoon conversations with distinct speakers.

Prepare the two designs, a short script and a clear turn-taking plan. Establish the character identities before adding the scene, then refine timing in the platform's editing workflow. Keep each voice associated with the intended character throughout revisions.

For selection, use a short exchange that includes a listener's reaction as well as the speaker's line. Check whether the non-speaking character remains recognizable and appropriately still. If the scene depends on overlapping dialogue or physical contact, include that exact requirement when evaluating the route rather than judging it from a single talking portrait.

7. Viggle: retain an existing interaction's motion

Viggle's Multi-Tracker is a character-replacement route: it identifies, tracks and replaces multiple people in an existing shot. Choose it when the choreography or interaction already exists in source footage.

Start with footage whose movement, framing and contact suit the scene. Keep the intended replacement identities distinct and check the tracking when people overlap, turn away or move out of frame. A useful source performance gives you the action to preserve; the replacement still needs inspection around faces, clothing and interacting hands.

If the scene has not been performed yet, decide whether recording it is easier than generating the choreography. The guide to character animation from a motion reference explains that input decision in more detail.

8. Adobe Character Animator: control prepared 2D puppets

Adobe Character Animator suits prepared 2D characters whose performance needs to remain editable. Put the puppets in a scene, record takes on their separate tracks and build the exchange around controlled expressions, gestures and audio.

The recording workflow lets you arm the puppet track and behavior parameters you intend to capture. Record one character while existing takes play back, then select the other track for its response. This can be useful when a cartoon exchange needs repeatable timing or a gesture that must be revised without regenerating the character artwork.

Allow time for preparing the puppets, triggers and layer structure. The full custom-puppet and performance feature set belongs to the Pro workflow; Starter is a simpler entry with predesigned puppets. For a handover or close overlap, plan the artwork and foreground order as part of the animation rather than expecting face tracking to solve the contact.

9. Autodesk Flow Studio: stage characters in a 3D scene

Autodesk Flow Studio offers a 3D scene route with character assets, transforms, animation and camera motion. Use that structure when character placement and spatial relationships need explicit adjustment before generating the final look.

Create or select characters, complete any required rigging and add them to the scene. The character controls let you position, rotate and scale them; animation can come from existing assets or an AI motion-capture workflow. Keep camera movement and character motion coordinated through the scene's timeline.

This route makes the spatial setup something you can adjust directly. It also requires more preparation than a generated two-shot. Inspect feet, hands, scale and the shared ground before finalizing an interaction, especially when one character reaches toward the other. A clean motion capture or rig does not remove the need to refine contact.

10. LTX Studio: keep a recurring cast organized

LTX Studio's Character Elements give a recurring cast reusable named assets that can be tagged in shots. It is useful for organizing a two-character story across scenes, while each generated interaction still needs its own review.

Create or upload each character, give the Element a clear name and use its tag when constructing a shot. Keep costume variations separately named so a later scene does not accidentally reuse the wrong outfit. Retain the cast's approved references alongside the storyboard.

This is especially helpful when the same pair returns in several locations or episodes. It organizes what the model should reference; it does not specify the exact path of two touching hands. Choose the generation or finishing method for that action separately from the asset-management decision.

Prepare the two-character shot before generating

Give each character a distinct name, appearance and screen position, then define one short action-and-reaction beat. Keep the first composition clear enough to inspect faces, hands and any shared prop.

Build a small shot brief:

Use the same names in references, prompts, recordings and filenames. Avoid switching from “Mira” to “the woman” midway through a crowded instruction. Decide which parts are essential: a readable reaction may matter more than camera movement in a short dialogue.

For physical action, make room for it in the starting composition. A handover requires visible hands and a clear gap between the people. An image that crops out the prop or begins with crossed arms gives you less useful information to work with.

Write action ownership into the prompt

Describe the sequence using the character names and visible positions: one person initiates, the other responds, then state the final pose or prop ownership. Make the camera instruction support that action rather than compete with it.

For a simple proposed reaction shot:

Medium two-shot of Mira in blue on the left and Sol in ochre on the right. Mira turns her eyes toward Sol and nods once. After Mira's nod, Sol looks toward Mira and raises one hand in a brief greeting. They stay in their original positions. Keep both faces visible, with a restrained camera and no cuts.

For dialogue, attach each line to its speaker and specify the listener's reaction. A dialogue brief might say that Mira asks, Sol listens without speaking, then Sol replies. If overlapping speech is essential, plan it explicitly and choose a route that lets you inspect or revise the two parts.

Add more action only after the basic exchange works. A single prompt that asks for talking, a hug, an object transfer, a turn and an orbiting camera makes it harder to isolate a problem and approve a useful take.

Review contact, occlusion and the speaking face

Review the complete shot for identity, action order, contact, prop ownership and dialogue assignment. Give extra attention to the moments when hands meet, one face passes behind another or a prop changes hands.

Watch the whole take at normal speed, then inspect the critical moments more closely. A good beginning and ending can conceal a brief identity swap or impossible contact in the middle.

Choose the repair that matches the failure. A wrong speaker may need a corrected line assignment or individual performance pass. A bad overlap may need a mask or a different composition. A repeated prop may be easier to fix by simplifying the action and regenerating a shorter beat.

Plan for the whole production cost

Budget for the complete shot: preparation, base generation, character passes, rejected attempts, compositing and sound. A route with more separate outputs may offer useful control while requiring more finishing time.

For the Runway route, the current Act-Two guide lists five credits per second with a three-second minimum per generation. Two five-second performance passes therefore total 50 Act-Two credits before the base video, retries, voice changes or other work. Use the selected input's actual quote when planning the shot.

Keep generation cost separate from editing effort. A fast shared-shot result can be a good choice for a simple greeting; a carefully performed conversation may justify recording and compositing. For exact physical choreography, plan the scene and contact work first rather than spending repeated attempts on a prompt that gives too little control.

Frequently asked questions

Use the interaction itself to decide how much generation, performance capture and manual control the shot needs.

Which route should I use for a hug or handshake?

For a quick concept, use a shared reference composition and a short generated shot. For a precise handshake or sustained embrace, use a workflow that lets you correct pose, contact and overlap directly.

Inspect the approach, the contact and the separation. A plausible single frame does not tell you whether the complete action holds together. If the contact must match choreography, a rigged or motion-based scene can provide a more useful starting structure.

Can both characters have separate voices?

Separate voices require correct speaker assignment as well as audio generation. Choose a route with explicit character dialogue controls or separately recorded performances, and check the mouth and voice together.

For an important exchange, keep the script and each character's recording organized separately. Review the listener too: the wrong person reacting or moving their mouth can make an otherwise convincing line confusing.

Will the same two characters stay consistent for a whole film?

Saved references and character assets help organize a recurring cast, but continuity still has to be checked shot by shot. Retain approved references, screen direction, costume details and scene notes across the sequence.

Approve short scenes before assembling the larger story. The guide to consistent locations and characters covers the additional reference and location decisions involved in that next step.

Start with one clear exchange

Start in Dreamina with a clear shared composition and one short interaction beat. Then choose the additional performance or scene-control workflow that the shot actually needs, and approve that exchange before building a longer sequence.

A readable nod and reply, a correctly assigned line or a cleanly staged handover is a stronger foundation than a long clip containing several unfinished interactions. Keep the two identities and the action ownership clear, then develop the scene from there.