How to Film Performance Reference for Character Animation

A useful starting setup

Film one clear, continuous performance with the framing your chosen animation workflow requires.

Put the camera on a stable support, light the performer clearly and rehearse the widest gesture before recording. Keep the relevant face, hands or whole body visible throughout the action. Record the actual line if the character speaks, and leave a little room before and after the performance so you can trim the take cleanly.

The most useful first recording is usually one simple scene: noticing someone, reacting and greeting them, for example. A short action makes it easier to see whether the resulting animation preserved the intended sequence. You can build a longer performance after that relationship works.

Decide what the character must follow

Choose the animation route before deciding how far to stand from the camera.

Write down the result you need: a rendered character video, editable body motion, facial animation or a performance on a prepared 2D puppet. Then identify the details the scene depends on. A shoulder shrug needs visible shoulders; a quiet change of expression needs enough face detail; a step backward needs the feet and floor in view.

An animator using Blender or iClone may want reference beside a rig rather than an automatically solved performance. That changes what is helpful: a convincing shift of weight can matter more than a perfectly unobstructed tracking pose.

Frame the movement, including its widest point

Rehearse the largest gesture before you press record, then leave enough room for it.

Check a raised hand, outstretched elbow, step sideways and deepest lean. If a body-capture workflow needs the complete figure, include the head and feet with a margin around the movement. The opening frame alone is not enough: hands that disappear during a wave remove information at the moment it matters.

For a facial take, move closer so the eyes, eyebrows and mouth remain readable. A distant full-body view can show where the person walked while losing the expression that explains why. For an upper-body performance, frame the hands and face together when the receiving tool supports that composition.

Keep the camera position consistent through the take. Save close-ups as separate recordings when needed, rather than zooming during a motion-capture shot. If the final animation has a specific camera angle, make an additional reference from that view for acting and composition; do not assume every solver accepts that angle equally well.

Make the face, joints and background easy to distinguish

Use steady framing, readable light and clothing that lets the movement remain visible.

Soft, even illumination makes a better starting point than a bright window behind the performer. Check whether the mouth falls into shadow when the head turns. Set focus and exposure so the recording does not repeatedly brighten, darken or blur during the action. Turn off beauty effects that alter facial features.

Choose clothing that follows the body without hiding its shape. For automatic tracking, a little visible texture can help; an all-black outfit against a dark background can conceal joints. Avoid loose sleeves covering the hands and reflective surroundings that produce a second apparent figure. A green screen is unnecessary unless another part of your production needs one.

Make sure the floor and feet remain distinguishable for standing or walking actions. Move clutter away from the performer’s path, and check the actual recorded file rather than relying only on a small camera preview. The goal is readable movement, not an elaborate studio.

Act a small scene rather than a list of poses

Give the character a reason to look, react and gesture.

Instead of recording “smile, wave, point,” imagine that a friend arrives late and you are pleased to see them. Your eyes may move first, your head follows, recognition changes the expression and the hand rises afterward. Those relationships create a performance an audience can understand.

Try this short rehearsal:

1

1. Begin occupied with something simple, such as looking at a note.

2

2. Notice the friend just outside the camera.

3

3. Let recognition change the face before the greeting.

4

4. Raise the hand, say the line and let the gesture settle.

Adjust the length to the receiving workflow. There is no need to freeze between every beat unless calibration or the scene requires it. Record a second interpretation with a different intention—relieved rather than excited, for instance—to discover which acting choice best suits the character.

Record speech with the performance when the character talks

Say the actual line while you act so the mouth, gestures and reactions share the same timing.

Use a quiet room and keep the recording free of background music. Speak at the pace the scene needs rather than rushing to fit an arbitrary duration. If the character says “Over here,” the pointing gesture might lead the words or land on “here”; choose that relationship deliberately.

Preserve the original audio with the original take. If a workflow uses separate audio, export a matching file and keep the start aligned. Avoid changing the video’s speed after recording while leaving the sound untouched.

Check how the destination handles speech. Facial movement, sound generation and lip-sync are separate parts of a production. A tool that follows body movement does not automatically supply the intended dialogue, and a lip-sync feature does not necessarily reproduce the rest of the acting.

Use props differently for acting reference and automatic tracking

A prop can make acting more believable, but an obstruction can make tracking less reliable.

When filming reference for an animator, use an object that gives the action real weight and resistance. Sitting on a chair helps reveal how the hips settle. Picking up a bag changes the shoulder and balance. Acting beside an imaginary heavy object can miss those small adjustments.

For an automatic solver, examine what the object hides. A desk covering the hips and knees or a large hat obscuring the eyebrows can remove useful tracking information. Choose an unobstructed variation when the capture mode needs it, and record an additional acting view if the physical interaction is essential.

Keep the action safe and simple. The reference should explain how the character moves; it should not require the performer to reproduce a dangerous stunt to make the scene convincing.

Review one take, change one choice, then record again

Watch the take before recording another version and choose one specific thing to improve.

First watch it at normal speed. Can you understand the character’s intention? Then watch without sound: does the reaction still read? Finally, inspect the important transition, such as the moment the hand rises or the expression changes.

Give the next take a clear purpose. You might keep the head quieter so the eye movement becomes readable, delay the wave until recognition, or make the step smaller so the feet stay in frame. Changing one choice helps you compare interpretations instead of accumulating nearly identical recordings.

Name useful takes by the difference—such as greeting-relieved-02—and note which one you prefer. Keep a natural version even if you also record a broader alternative for a stylized character.

Use the recording as a performance reference in Dreamina

In Dreamina, give the character image and the performance video different jobs.

Start with a character reference that clearly shows the appearance you want to retain. In a supported Seedance 2.5 reference workflow, add the performance recording for the action and timing it contributes. Keep the reference set focused: several clips with competing movements can make the intended performance harder to specify.

A practical prompt can describe the roles and sequence:

Use the character image for the character’s face, outfit and proportions. Use the performance video for the greeting’s movement and emotional timing. The character notices someone off-screen, recognizes them, smiles and raises the right hand in a small wave. Keep the character’s established appearance and the camera steady.

Add details that distinguish this scene: which side the person looks toward, whether the greeting is hesitant or enthusiastic, and which gesture matters most. For a short shot, a five-to-ten-second reference can be a manageable starting point; choose the actual clip length and settings supported by the selected mode.

Review the first result before adding more actions. If you are still choosing between generated performance and a rig-based workflow, the guide to animating a character with a motion reference explains the next decision.

Match the recording to the tool that will receive it

The same acting idea may need a different camera setup for a generated video, a 2D puppet or a 3D rig.

Use the relevant capture instructions below before recording a full set of takes. These differences affect the footage you need, not just the settings you choose afterward.

MetaHuman Animator: face, body or combined capture

MetaHuman Animator’s current mono-video workflow can solve facial animation, body animation or both.

For body capture, leave room around the complete figure. For facial capture, prioritize readable eyes and mouth. Epic’s guidance recommends HD or 4K footage at 60 fps, with appropriate compression, for its capture workflow; use that as an Epic-specific setup rather than a universal rule for every animation tool.

The body and combined mono-video route uses the experimental Markerless Motion Capture plugin in Unreal Engine 5.8 or later, with Windows-only support for those solves. In the supported workflow, ingest the recording, process the performance, then export and apply the animation. Face and body can be processed from one view, while separate-camera arrangements require their corresponding setup. Live Link Face is another supported capture route on compatible devices; check the mode and device together.

Adobe Character Animator: perform into a 2D puppet

Adobe Character Animator uses webcam-driven behaviors to animate a prepared 2D puppet.

Check the rig before blaming the recording. Body tracking needs a puppet prepared for that behavior, with Camera Input and Body Tracker Input enabled. Step back far enough for the required body view; the recording countdown gives you time to move into position. Adobe recommends using Limb IK alongside Body to help reduce stretching and foot sliding.

A prerecorded MP4 is not a substitute for that webcam Body setup. Prerecorded facial analysis has a separate After Effects-to-Character Animator route, and imported dialogue can be used to compute lip-sync from scene audio. Choose the appropriate input for the part of the performance you want to drive.

Rokoko Vision: prepare a clear body-performance upload

For Rokoko Vision 3.0, record the performance first and upload the video into Rokoko Create.

The current workflow does not record directly in the browser. Use a stable camera around chest height, avoid an extreme wide-angle view and keep the performer’s head and feet visible throughout the action. Choose fitted clothing with distinguishable texture and contrast, and avoid mirrors or obstructions around the body.

This is a useful setup for a body-performance take. If the scene also depends on detailed facial acting, plan how that animation will be supplied rather than expecting the body recording alone to resolve every part of the face.

DeepMotion Animate 3D: follow the selected capture guide

DeepMotion’s single-person capture guide is a useful example of a body-focused recording setup.

The guide recommends a stationary camera, roughly two to six meters from the performer, with the complete figure unobstructed. For movement on the ground, a three-quarter camera angle can help keep joints visible. Clothing texture can assist tracking; hiding the body inside loose or visually indistinct clothing works against the capture.

Its resolution and frame-rate recommendations belong to this workflow, including plan-specific support for higher frame rates. Follow the capture guide for the actual mode you select, especially when using face, hand or multi-person features, rather than treating the single-person instructions as a description of every available mode.

AccuFACE and iClone: prioritize a readable face

AccuFACE’s facial capture workflow needs a clearly framed face and even lighting.

The facial-capture guidance recommends facing forward with the head level, keeping the face inside the view and avoiding strong shadows. Check for accessories that cover important facial features. Follow the calibration instructions, including keeping the eyes open when required.

Keep this facial task separate from the body-animation task in an iClone production. A useful mouth and eyebrow recording does not by itself specify a walk, hand contact or weight shift. Plan those inputs or animation edits alongside the facial performance so the character’s whole reaction remains coherent.

Runway Act-Two: frame a driving performance for the chosen input

Act-Two uses a driving-performance video with a character image or character video.

Choose that second input carefully. With a character image, gesture control can guide body movement. With a character video, Act-Two retains that video’s body movement, camera and environment while transferring facial performance and expressions. The two inputs do not offer the same control.

Runway recommends a single visible face, continuous footage and framing no farther out than the waist. Keep hands visible from the start when using gestures and use a compatible starting pose. Act-Two is available on Standard and higher plans and supports performances up to 30 seconds; confirm those conditions before preparing a longer set of recordings.

Compare the output against the important beats

Judge the animation against the action and emotion the scene needs to communicate.

Watch the result once as a viewer before examining individual details. A greeting that feels warm and correctly timed may be more useful than one with a larger, technically recognizable wave but an unreadable reaction.

Choose the few details essential to the scene and revise those first. For a generated character video, that may mean a more focused prompt or clearer take. For rig-based animation, it may mean adjusting curves, contacts or facial poses after capture.

Common filming problems and better fixes

Fix the missing visual information before trying to solve every problem in the animation settings.

When the source is clear but a rig still bends awkwardly, inspect the rig mapping and animation setup. Refilming is useful for missing information; it cannot repair every character-configuration issue.

Keep the original take and a short handoff note

Save the unedited recording and describe the character, action and timing it is meant to support.

Include the selected take, matching audio if separate, the character reference and a few notes: “Notice the door before the smile,” “Small right-hand wave,” and “Keep the weight on the left foot.” Add timing markers only where they help the animator or model understand the sequence.

For rig-based work, keep the source performance alongside exported animation so later cleanup has a reference. For a recurring generated character, preserve the approved appearance and scene references too. The guide to keeping locations and characters consistent across clips helps organize that next stage without changing the character for each shot.

Frequently asked questions

Match the recording setup to the receiving workflow, then prioritize the details the scene needs to communicate.

Can I do this with one phone?

A single phone can record useful reference, provided its framing and file format suit the receiving workflow. Put it on a stable support and rehearse the action inside the frame. Use a supported single-camera mode for automatic capture; for manual acting reference, choose the view that explains the scene most clearly.

Should I film the face and body separately?

Use separate views when the workflow supports them and one view cannot show the detail you need. A distant shot may capture a walk but lose subtle facial movement. Some workflows combine separate face and body animation; others use a single view. Check the receiving workflow before recording two files you cannot use together.

Do I need 4K or 60 fps?

Use the specifications for the exact capture mode rather than treating 4K or 60 fps as universal requirements. Higher-quality recording can help preserve detail, but readable framing, light and continuous movement still matter. Confirm supported formats and compression as well as resolution and frame rate.

Should I exaggerate every expression?

Make the performance readable while keeping the timing and emotion believable. Start naturally, then record a broader alternative if the character’s style needs it. Extreme motion can hide features or introduce tracking problems. Compare interpretations instead of assuming the largest expression will always animate best.

Will one clip copy every hand gesture and facial detail exactly?

A readable recording improves the input, but the result also depends on the model, capture mode and character setup. Decide which details the scene requires, inspect them in the output and allow for refinement. A rendered video and editable motion-capture data are different deliverables, so select the route that matches the control you need.

Start with one short, readable performance

For a Dreamina character video, begin with a clear character reference, a short performance and one action you can judge. Film the action with enough space and light to see what happens, explain the role of each reference, then compare the result with the scene’s key beats. Once the gesture and expression communicate the right intention, build the next shot from that foundation.