Skip to Main Content
NewKling 4.0Kling 4.0 is here · Seamless outfit swaps & cinematic effectsOutfit swaps & cinematic effectsTry Kling 4.0
Genjutsu AI

Turn a Real Person into an Animated Character

Turn a real person into an animated character, or replace who appears in your video. See what each workflow can preserve, then prepare your clip and reference images.

Original audio can be reused. Movement, expressions and lip sync need separate checks.

What tool turns a real performance into an animated character?

Upload video and reference

For speech, expressions and gestures, Runway Act-Two explicitly documents performance-driven character animation. For replacing a person while following the source motion, camera and timing, Higgsfield Genjutsu offers related editing workflows. These solve different parts of the task.

An animated result alone cannot identify the tool used. Look for the original post, a creator explanation or the editing interface.

Runway Act-Two documentationHiggsfield Genjutsu official page

Keep source audio
0 / 10000
Balance: — creditsCost: —

Generated videos are stored for 7 days after completion. Download yours before it expires.

Model estimates follow your selected duration. Shorter model limits trim the selection; final credit quotes apply after input checks.

Inputs upload automatically for a credit quote. Only clicking Generate reserves credits and creates a task.

Sign in and prepare your inputs to see a credit quote. Generation starts only after confirmation. Video up to 200 MiB, selected clip 4–30 seconds. Each photo up to 30 MiB and at least 300 px per side. Sign in before uploading and generating.

Your Input Preview

Your source video and reference photos appear here after selection. Task progress and results will appear below.

Countryside Duo

Prompt
Edit @video1 using these separate reference mappings. The dancer in the blue jersey, initially on the left: curly-haired woman in a red jacket from @image1. The dancer in the striped green/yellow jersey and cap, initially on the right: brown-haired man in a dark green sweatshirt and gray trousers from @image2. Move the scene to a quiet countryside village street with white cottages and a stone wall. Keep the motorcycle in its original position. Preserve the source movement, timing, camera path, framing, contact points and occlusion order. Track each subject through crossings and cuts; never merge identities or add people. Keep unlisted people and objects unchanged.

Countryside Duo

Original clips and effect examples from Higgsfield, shown as references. Generation uses the original and your photo; results may vary.

Upload only videos, photos and likenesses you have permission to use. Privacy, storage and deletion · Terms of Service

Can I change the person without changing the voice?

Yes. Upload an existing video and a photo, then enable Keep source audio. The selected clip’s speech, music and background sound are added to the result. The voice is not re-recorded; generated mouth movements still need a separate check.

In Genjutsu AI, person replacement requests the original background. Motion Transfer is a separate workflow that can change framing and scenery. Review the output from either workflow.

Related product sources

Related workflow sources: Genjutsu · Viggle Body Swap

Genjutsu AI is an independent service with its own accounts, models and credits. Looking for the official Higgsfield entry?

What does “keeping the performance” mean?

Movement, voice and mouth shapes are separate checks. Compare the official descriptions with what this studio’s current workflow and examples establish.

What does “keeping the performance” mean?
What to preserveOfficial tool descriptionsIn Genjutsu AI
Body movementGenjutsu describes keeping motion and timing. Act-Two uses a performance video to drive the character; body control depends on the character input.Observed in the silent anime case: dancing remains recognizable. Every movement and its timing have not been certified.
Facial expressionAct-Two explicitly describes transferring facial movement and expressions.Not verified for animated speech. The silent dance case does not establish accurate transfer of facial expressions.
Hands and gesturesAct-Two gesture control is available with a character image. With a character video, body motion comes from that video.Not separately verified. Following source movement does not guarantee correct fingers or hand poses; inspect the full clip.
Original soundAct-Two describes transferring performance audio. The reviewed Genjutsu page does not establish source-audio retention.Workflow: Keep source audio reuses the selected recording when an audio track exists. The app-generated case here is silent.
Mouth movementsAct-Two describes speech animation. This is a capability description, not proof of exact lip sync for every input.Not verified for animated speech. Reusing the recording does not repair generated mouth shapes.
Camera and backgroundGenjutsu Motion Transfer describes retaining camera motion and timing; Object Swap aims to preserve the rest of the shot.Person replacement requests the source background; nearby details may change. Motion Transfer can change framing and scenery.

Official descriptions checked October 12, 2026. Studio notes describe the current workflow or the labeled cases below; no new generation test was run for this page.

Higgsfield Genjutsu official pageRunway Act-Two documentationReview the cases below

Choose the workflow for your shot

First decide whether the priority is the existing scene, the body movement or a speaking performance.

  1. Replace the character

    Keep the existing scene and change who appears in it.

    Start with Person replacement in this editor. Add the source video and a character reference, then describe who should change. The model is asked to follow the motion and preserve the background; review nearby details too.

    Upload video and reference

  2. Transfer the movement

    Use the source action to guide a character in a new look or scene.

    Return to the editor and select Motion Transfer. The source supplies movement and the reference supplies appearance. Framing and background may change; choose person replacement when retaining the existing scene is the priority.

    Return to editor to choose Motion Transfer

  3. Drive a speaking character

    Make a character follow a person’s speech, expression and gestures.

    Act-Two takes a performance video plus a character image or video. An image allows gesture control; a character video supplies its own body, camera and scene movement. This studio does not offer an Act-Two control or a verified animated-speaking workflow.

    Runway Act-Two documentation

Genjutsu AI is independent of Higgsfield and Runway. Accounts, credits and support are separate; the official documentation links open those services’ instructions.

Compare the Performance and the New Character

Start with a silent anime result generated by this studio, then listen to a downloaded Higgsfield example. Each case shows its origin and what it can establish.

Anime Times Square · Actual App-Generated Result

Generated by this app on October 9, 2026, using the downloaded Higgsfield Times Square clip and one identity photo.

Observed: the performer becomes an animated woman with short golden hair, an orange top and blue trousers; the street and dance remain recognizable. File decoding and sampled frames were checked. Exact identity, every movement and full-scene animation were not certified.

Original Video
Actual Generated Result
0.0 / 0.0 s

Source and result have no audio track. This pair shows the visual edit only.

Generation settings and result notes

Kling 3.0 Omni transformation via KIE · matching-reference preparation · 720×1280 · source 5.739 seconds / result 5.917 seconds · generated October 9, 2026.

The identity-reference photo for this previous generation is not displayed. The input-guidance frames below are not the reference used for this result.

Waiter Recast · Downloaded Higgsfield Example

This source and effect are downloaded Higgsfield examples, not this app’s output. Both contain audio you can compare. Exact audio equivalence and lip sync have not been certified.

Original Video
Higgsfield Example Result
0.0 / 0.0 s
Choose one audio track

Choose a track, then press Play. Both videos play together; only one track is audible.

Media details and review checklist

Higgsfield Waiter Recast · source and output variant 4 · 1280×720 · approximately 8 seconds · both files contain audio · downloaded October 8, 2026.

What to check in this example

  • Compare music and background sounds at the same moment. For a speaking clip, also check words and pauses.
  • For a speaking clip, listen to the result and inspect mouth openings and closures throughout each sentence. Matching audio timing alone is not proof of matching lips.
  • Compare the character, clothing and background throughout the clip. This downloaded example illustrates an editing direction; it does not establish the quality of this app’s results.
Back to the Generator

The Higgsfield example is audible. This app result is silent and does not demonstrate lip sync.

Original Audio and Lip Sync: What Is Preserved?

Keeping a recording and making a mouth match it are separate checks. Here is what this workflow does, and what to inspect in the output.

Can I keep the original voice, music and background sound?

If your source contains audio, enable Keep source audio in Advanced settings to reuse the selected clip’s speech, music and background sound. Export may re-encode it, so the file is not guaranteed to be byte-identical. Turning this off produces a silent result.

Will the new person’s lips match the original speech?

The editing model follows the source performance, but mouth movements can change. Adding the original recording does not correct them. This page has no verified speaking result generated by this app. Watch your full result with sound before using it.

What if I want the person to say different words?

Keeping source audio keeps the original words. New lines need a separate voice and lip-sync workflow. This tool does not provide that workflow; editing the visual prompt will not rewrite the recording.

From Your Performance to a New Character

  1. Actual generator interface showing the source video upload.Enlarge screenshot
    01

    Upload Your Video

    Choose a clear subject and select 4–30 seconds. MP4, MOV, WebM and OGG are supported, up to 200 MiB. Model-specific limits are checked with your quote.

  2. Actual generator interface with a selected character reference photo.Enlarge screenshot
    02

    Add a Character Reference

    Add a clear reference showing the face, body and outfit you want. Describe the person to replace. Use Person replacement to request the existing scene, or choose Motion Transfer if scene changes are acceptable. Images can be up to 30 MiB each, at least 300 px per side; the quote checks model-specific limits.

  3. Actual Advanced settings with Keep source audio enabled.Enlarge screenshot
    03

    Check Source Audio and the Quote

    Open Advanced settings and confirm Keep source audio is on. Preview the source with sound. Sign in and prepare your inputs, then review the automatic model, clip limits and credit quote.

  4. Actual page showing the newer Higgsfield source and the app-generated template result for review.Enlarge screenshot
    04

    Generate, Listen and Review

    Confirm the quote to generate. Watch the full result with sound, checking lips, face, hands, background and occlusions before downloading. Resume an existing job in task history after a refresh.

Back to the Generator

Choose a Clear Character Reference

Use a clear subject and a reference with a readable face. For an animated character, show the intended eyes, mouth and outfit clearly. For full-body shots, include the body too. Keep the first test short and describe the target at the clip start.

Fast movement, occlusion, turns and interactions between people can increase errors. Extra angles can be assigned to the same identity, but do not guarantee a better result.

Face consistency in AI video
Clear face in a frame from the downloaded Higgsfield Selfie Walk source video.

Clear face and visible outfit

Choose even lighting and a sharp, unobstructed face. Include more of your body and clothing when the source shows a full-body view.

Avoid a covered or unreadable face

Hands, masks, sunglasses, blur and heavy filters hide identity details. Retake the photo so your features are visible before uploading.

These frames come from the newer Higgsfield Selfie Walk source. The obscured version illustrates an input problem; neither is the identity reference for the results above.

Back to the Generator

Replace One Person in a Two-Person Video

Target descriptions, occlusion and checks for two-person clips. Expand to read.

Upload a video and a reference photo, then describe the person you want to replace.

Can AI Replace Just One Person in a Video?

Tools that support directed video editing can attempt to replace a described person using a reference photo. Review both people afterward: the other person may also change. This site uses its person-replacement workflow; keeping the second person unchanged has not yet passed a dedicated two-person test.

Higgsfield describes selective editing in its Genjutsu overview. That is related product information, not evidence of this site’s results. Higgsfield Genjutsu official overview

How to Replace One Person

  1. 01

    Upload the two-person video and a reference photo

    Use a clear photo of the replacement person. Select a clip whose opening frame clearly shows both people.

  2. 02

    Describe the person in the first frame

    Use position plus clothing color or another visible feature. “The person on the left” alone may become ambiguous if they move. For example:

    The person on the left in the first frame, wearing a red jacket.
  3. 03

    Generate, inspect both people, then download

    Check the replacement identity, the other person’s face, outfit and motion, and the background at the beginning, middle, end and any crossing. Download from the completed task after reviewing it.

One Person Replaced. Compare Both People.

No reviewed example of replacing only one person is published here yet. The single-person example below demonstrates a different case. To evaluate your own result, compare the source and output at the same timestamps, alongside the reference photo and your target description.

  • Check the target: does the replacement match the reference throughout the clip?
  • Check the other person: did their face, clothing, body or motion change, or did the two identities switch?
  • Check the scene: are the background, framing and camera movement retained?

Crossings, Occlusion and Other Limits

Similar clothing, fast motion, crossings and occlusion can cause the wrong person to change or affect the other person. Start with a short clip and distinct people. A position description is not a tracking lock.

  • Separated people with stable positions

    Start here. Inspect both faces and outfits at the beginning, middle and end.

  • Both people moving without overlap

    Follow each identity through the motion; check for drift in the person you wanted to keep.

  • Brief crossing or occlusion

    A difficult case. Inspect immediately before, during and after the overlap for identity switches or unwanted changes.

These are review scenarios, not demonstrated successes. Use every result to check both people.

Replacing One Person: Questions

Return to the generator

Frequently Asked Questions

What the current workflow supports, and what to check before generating.

Can Genjutsu replace anyone? Evidence and limits

Read third-party tutorials, conditions and official sources. Higgsfield demonstrations and this app’s results are labeled separately.

What the Higgsfield Genjutsu tutorial actually establishes

The YouTube tutorial by Zubair Trabzada / AI Workshop was published on September 11, 2026. Its description discusses motion transfer, character and product swaps, backgrounds, and Claude MCP. It also discloses Higgsfield sponsorship. Treat its Higgsfield Genjutsu examples as a creator demonstration, not an independent benchmark covering arbitrary footage.

The published chapter list gives useful places to investigate: character swapping at 02:49, results and accuracy at 04:43, and characters plus backgrounds at 05:46. We could read the page description and chapter metadata; subtitles were unavailable during review. This guide therefore cross-checks Higgsfield Genjutsu capabilities against official documentation rather than inventing a transcript or reporting our own generation test.

What to change when Higgsfield Genjutsu misses

If the wrong person changes, make the target description more specific and choose a less ambiguous shot. If identity varies, simplify the references. If the scene changes too broadly, reconsider whether a targeted swap fits better than reconstruction. Change one input at a time so the next Higgsfield Genjutsu attempt answers a useful question.

Do not repeatedly regenerate an unchanged request merely because the first result disappointed you. Identify the mismatch and estimate whether another attempt is worth its cost. Higgsfield Genjutsu uses credits; a usable final edit may require more than one attempt. This guide reports no measured retry count or success percentage.

Does Higgsfield Genjutsu give permission to replace anyone?

Technical capability does not establish permission to use someone's likeness or footage. For a straightforward Higgsfield Genjutsu test, use your own material or references you are authorized to use. Review applicable publishing requirements before sharing a realistic impersonation.

Official upload limits need a separate check

The official product FAQ, checked October 12, 2026, lists reference videos of 4–30 seconds and up to 30 reference images. Public limits can change: follow the validation in the official generator you use. This independent studio has its own upload, clip and model limits, which are checked with the quote.

Sources

Try Your Clip and Character Reference

Return to the same editor to prepare your inputs and review the quote. Your current selections stay in place.

Back to the Generator

Find a related tool, a practical guide, or the credit plan for your next edit.

Face consistency in AI video

Choose a workflow for turns, fast motion and occlusion, then check your own clip.

Watermark-free alternatives

Compare editing tools, export entitlements, credit packs and commercial-use conditions.

Change clothes in a video

Prepare an outfit reference, request a clothing edit and check the result across the whole clip.

Genjutsu beginner tutorial

Follow the video chapters, prepare references, and plan your first edit.

What's the point of Genjutsu?

Understand its purpose, overlapping tools, and when preserving a take is useful.

No monthly subscription

Understand one-time purchases, quotes, credit validity and renewal before generating.

Where to try Higgsfield Genjutsu

Find the official Higgsfield entry points and understand how this independent studio differs.

Editing tutorial

From source to export.

Plans & credits

Compare credit options.