Genjutsu AI is a reference-image-guided video-to-video editing tool. Its core idea is: start from an existing video, and by adding up to nine reference images and a short text description, replace or change a certain element in the frame while preserving the original video's motion, camera movement, and pacing as much as possible.
In terms of use, it is aimed at editing needs that require partial or overall changes to an existing video, such as replacing the character in a dance or trend video with another person or an animated figure, swapping the product in an ad for a different one, putting the same person in a different outfit, redrawing live-action footage in anime or 3D style, or placing the same performance into a new scene.
Main capabilities: replace characters, clothing, products, or backgrounds, or redraw the style of the entire video; upload or use built-in clips; support 480p, 720p, and 1080p resolutions; output MP4 and preserve the source audio; up to nine reference images, and style-type changes can use only a text description.
How to use: Step one, choose the source and clip, upload an MP4 or MOV (up to 5 minutes) or use a built-in clip, slide to select any whole-second range from 5 seconds up to the clip's own length, and drag the window to position it; step two, choose the type of change and add reference images; step three, explain in one sentence what to change and which image to use; step four, choose the model (Seedance 2.5, Seedance 2.0, MiniMax H3, Seedance 2.0 Fast, or Mini, from highest to lowest cost) and resolution; the credit cost is shown before generation, and the MP4 can be downloaded when finished. The entire edit runs in the browser with no download or installation required.
Use cases: swapping people in dance or trending videos, replacing ad products, virtual try-on and styling previews, converting live action to anime or 3D, changing the filming scene, separately replacing two people in a two-person video, and more.
Limitations: results are model-generated and are not pixel-for-pixel identical to the source footage; Seedance 2.5 directly edits the selected range, keeping motion, gestures, cuts, camera, and composition closest to the source, while MiniMax H3 and Seedance 2.0 models generate new shots with reference to the original, allowing more room for change but potentially looser temporal precision; fingers may briefly deform when gripping a product or passing in front of the face; long hair and loose clothing may differ during fast spins and head turns; small logos and packaging text are not guaranteed to match exactly; when there are multiple similar people in the frame, the replacement target must be clearly specified; when the replacement subject's proportions differ greatly from the original performer, drift is more likely. Uploaded footage and reference images undergo the provider's asset review, and rejection will terminatetasks and refund credits, but review does not represent a safety or licensing guarantee. Input clips of 5 seconds to 5 minutes, maximum 50 MB; Seedance 2.5 up to 30 seconds, MiniMax H3 and Seedance 2.0 series up to 15 seconds. Credits are calculated based on the selected model and resolution, and credits are refunded if generation fails; new accounts receive 3 credits, there is no unlimited free tier, videos generated without purchasing a plan carry a moving watermark, and after purchasing any plan or credit pack you can download the same video without a watermark. Users can only edit videos and images they have the right to use, recognizable people must be adults who consent to the editing, content must be non-sexual, and it should be labeled as AI-generated when shared, and must not be used for impersonation or deception.
Regarding pricing, the materials provide some credit examples: Seedance 2.5 at 720p costs 88 credits for 5 seconds, 175 for 10 seconds, 262 for 15 seconds, and 523 for 30 seconds; at 1080p, 10 seconds costs 310 credits; MiniMax H3 at 720p costs 77 credits for 10 seconds. The actual quote is subject to what the generator displays.
