What is a Genjutsu AI video?

A video-to-video edit that keeps a clip's original motion, timing and camera and changes the character's look, outfit and scene to match a reference image. The name comes from Higgsfield's Genjutsu tool; this page offers an independent tool that does the same kind of edit.

Is this Higgsfield Genjutsu?

No. Keepframe (hold-frame.com) is independent and not affiliated with Higgsfield. It uses ByteDance Seedance 2.5 video editing through Volcengine Ark.

How long can my video be?

Each video is about 5 seconds. Upload any MP4 or MOV up to 60 seconds and 100 MB and pick the 5 seconds to use; clips of 4–6 seconds are used whole. The output is about the same length and can be up to 0.4 seconds shorter.

What resolution is the output?

720p. In our test an 854x480 source came back as 1280x720 with the same 16:9 framing.

How much does one video cost?

30 credits per video. Credits come in one-time packs ($6 for 50, $12 for 125) or monthly plans ($9 for 100, $19 for 250).

Can I use footage of a real person?

Partly. The engine declines reference images that look like a real person, so show the outfit, character or place on its own — for an outfit swap, use a product photo of the clothes. Clips that show people can also be declined by the engine's safety checks; if that happens your credits are returned automatically. Only upload people who agreed to it.

What happens if a video fails?

Credits are reserved when you start and returned automatically if the engine rejects the files or the render fails. You only pay for videos that finish.

Do I need to write a prompt?

No. Pick what to change (character, scene or both) and upload the reference. An optional short note can steer details like colors or materials.

Is there a watermark, and can I use the result commercially?

No watermark is added. You can use results commercially if you own or have permission to use the source clip, the reference image and any characters or brands in them.

How long does a video take?

Our 5-second test finished in about 3 minutes. Queue times vary; you can leave the page and come back — recent videos are listed in your account for 24 hours.

New Video · Seedance 2.5 edit · 720p

Genjutsu-Style AI Video: Keep the Motion, Change the Character and Scene

Upload a clip and one reference image. The AI keeps every move, the timing and the camera from your clip, and rebuilds the look of the subject and the scene from the reference.

Motion and timing kept frame by frame30 credits per videoFailed videos refundedNo watermark

Your clip

Result

Keep · motion Source clip MP4 or MOV · up to 60 s · you pick the 5 seconds to use Change · look Reference image JPEG, PNG or WebP · the character, outfit or scene you want

What should change?

Optional note (appearance only, 300 characters)

Tip: the reference image should show the new look on its own — an outfit or product photo, a character, or a place. Reference photos of real people are declined by the engine (credits refunded). Use only footage you have the rights to; uploads are deleted when the video finishes.

Real results: same motion, new look

Every pair below is an actual output from this tool on Oct 1, 2026 — no cherry-picked frames, no editing afterwards. Source clips were generated with Seedance; reference images were made with an image model. Each video cost 30 credits.

Reference: golden retriever in a red cape and goggles on a beach boardwalk

Pet costume + new location

Every stride, ear flop and the pace toward the camera stay the same. The cape, goggles and beach boardwalk come from the reference. 5.0 s in, 5.0 s out, 1280×720.

Reference: flat-lay of a silver bomber jacket, black cargo pants and white sneakers

Outfit swap from a product photo (character only)

Same dancer, same moves, same studio. The silver jacket and white sneakers came through; the cargo pants came out as shorts — check details like this before you post.

Reference: unbranded red patent low-top sneaker on a dark marble pedestal

Product colorway + new set

The turntable spin and camera stay locked; material, colors and the pedestal follow the reference. Our first try used a reference with a brand logo and the logo carried over — use unbranded references for products you don't own.

Measured test: frame by frame

Not a mock-up. This is a real Seedance 2.5 edit run on Oct 1, 2026 with an original illustrated clip. Watch the arm on the left side of the frame: it swings in front of the waist and returns, while the other arm stays still — in both videos.

Source clip. 5.0 s · 854×480 · 24 fps · original illustration

Reference image: gold astronaut suit on a lunar stage with a ringed pink planet — **Reference image.** Gold suit, lunar stage, ringed planet

Result. 4.7 s · 1280×720 (shown scaled) · same arm, same timing, new suit and stage

Frame-by-frame comparison: top row source, bottom row result, sampled every 16 frames; the same arm moves at the same moments

Top: source. Bottom: result. One frame every 16 frames (0.67 s). The screen-left arm reaches across the waist at the same moments; the screen-right arm stays still.

Motion keptPass

Output length4.7 s of 5.0 s

Output size1280×720

Render time≈ 3 min

What wasn't perfect: a caption from the reference image carried into the frame. The generator now tells the engine not to copy text or logos from the reference, but check the result before you publish. One test is one test — results vary with the clip.

How it works

No prompt engineering and no keyframes. You decide what to keep by uploading it, and what to change by showing it.

Upload the motion

Any clip up to 60 seconds — pick the 5 seconds you want. One clear subject works best. The movement, gestures, timing, camera moves and framing are all taken from this clip.

Show the new look

One reference image of the character, costume, material or place you want. Choose whether it replaces the character, the scene, or both.

Generate and download

30 credits are reserved when you start and returned if the render fails. The 720p MP4 is ready in a few minutes; download it within 24 hours.

What works and what doesn't

Other Genjutsu-style tools let you find this out after a failed render. Here it is up front.

Works well

Illustrated, 3D and cartoon characters — VTuber models, game characters, animation tests
Mascots and brand characters performing a trending move
Products — restyle materials, colorways or the set around a product shot
Pets and animals — new costume or setting, same movement
Outfit swaps from a product photo — show the clothes on their own, not worn by a model
One clear subject, full body or upper body in frame, steady lighting

Won't work (or works poorly)

Reference images of real people — declined by the engine (refunded); show the outfit, character or place instead
People without their permission, public figures, minors or sensitive content — not allowed
Clips shorter than 4 s — and only 5 seconds of a longer clip are used per video
Busy scenes with several people or subjects — the model may change the wrong one
Tiny text, logos and captions — can be dropped or copied from the reference
Reference photos with strong lighting — that lighting can carry over into the result
Audio — output is silent; add your sound back in your editor

Specs and pricing

One flat price per video, no resolution tiers, no hidden minimums. Credits are shared with Keepframe's image tools (a standard image uses 5).

Spec	Keepframe Genjutsu-style video
Input clip	MP4 or MOV up to 60 s and 100 MB, 480p or higher. Clips over 6 s: you pick a 5-second segment. Clips of 4–6 s are used whole.
Reference	1 image (JPEG, PNG, WebP), at least 300 px per side
Output	720p MP4, same aspect ratio as the clip, about the same length (up to 0.4 s shorter), no audio, no watermark
Engine	ByteDance Seedance 2.5 video edit (via Volcengine Ark)
Cost	30 credits per finished video. Rejected or failed renders are refunded automatically.
Storage	Your uploads are deleted as soon as the render finishes. Download the result within 24 hours.

Compare all credit plans and usage costs

50-credit pack

$6 once

1 video (+20 credits for 4 images)

No subscription

Best for video · no subscription

125-credit pack

$12 once

4 videos · $3.00 per video

No subscription

Starter

$9 / month

100 credits · 3 videos

Credits carry over. Cancel anytime.

Pro

$19 / month

250 credits · 8 videos · $2.38 per video

Credits carry over. Cancel anytime.

Keepframe vs Higgsfield Genjutsu

Higgsfield created Genjutsu and it is the more complete tool. Here is an honest comparison so you can pick the right one.

	Keepframe (this page)	Higgsfield Genjutsu
Clip length	5 s per video (upload up to 60 s and pick the part)	4–30 s
Reference images	1	Multiple
Output	720p	Up to 1080p (reported by third parties)
Price per video	Flat 30 credits, shown before you start	Credit cost shown in the generator; varies by length and resolution
Failed renders	Refunded automatically	Not stated on the product page
Real people	Reference photos of real people are declined (refunded) — use an outfit or product photo	Not stated on the product page
Watermark	None	On free-plan output (per user reports)
Best for	Short loops, mascots, characters, products, quick tests	Longer shots, higher resolution, multi-reference projects

If you need clips longer than 6 seconds, 1080p or several references, Higgsfield is the better fit. Keepframe is independent and not affiliated with Higgsfield. Competitor details from higgsfield.ai/genjutsu and public reviews, checked Oct 1, 2026.

Ideas that fit a 5-second clip

Mascot does the trend

Film anyone in a full mascot suit doing the move, then swap in your brand character's look.

VTuber and 3D avatar skins

Record one performance with your avatar and try new outfits or worlds without re-rigging.

Product colorways

Turn one spin or reveal shot into a new material, finish or set from a single reference.

Pet costume clips

Keep your pet's exact movement and change the costume or the scene around it.

Animation look tests

Preview how an animated shot reads in a different art direction before committing.

Outfit swaps from a product shot

Use a flat-lay or product photo of the clothes as the reference and keep the same walk or dance.

Genjutsu-Style AI Video: Keep the Motion, Change the Character and Scene