How AI background replacement works on video
Classic background removal cuts the subject out and pastes it on a new plate, which is why edges often flicker. A video-to-video model works differently: it re-renders the frame, keeps the subject and their motion from your clip and draws a new environment around them that follows the original camera path. When the camera pans, the new street pans with it; when the light changes, the scene is lit to match.
Because the whole frame is redrawn, the edit can also change the light, the weather or the season at the same time. That is useful for a night version of a daytime shot, or for a product clip that needs to feel like winter.
When changing the scene is worth it
A new setting is often the cheapest way to reuse a good take.
- Short ads: place the same product demo in a kitchen, an office and an outdoor scene.
- Social clips: move a talking or dancing shot into a city, a beach or a fantasy world.
- Seasonal campaigns: add snow, autumn leaves or evening light to existing footage.
- Clean-up: replace a messy room or a distracting street behind the subject.
- Storyboards: test how a scene would look in different locations before you book one.
Step by step
Genjutsu AI accepts one MP4 or MOV file up to 100 MB, 3–15 seconds long (4–15 seconds for Seedance 2.5 and the Higgsfield workflows), at least 720 pixels on the short side and recorded at 24–60 frames per second.
- Upload a steady clip Shots with a clear subject and moderate camera movement work best. The subject should not leave the frame for long.
- Describe the new world Start from the 'Change the world' preset, for example: 'Turn the setting of @Video1 into a snowy night city street with neon signs. Keep the people, their faces, their motion and the camera path unchanged.'
- Add a place reference if you have one A photo of the location, a mood image or a still from your brand guide can guide the look. Cite it as @Image1 in the instruction.
- Review and download Check edges around hair and hands and any text in the scene, then download the MP4 with the original audio.
Writing a good scene prompt
Be concrete about the place, the time of day and the light: 'a small Tokyo side street at night, wet asphalt, warm shop signs' gives the model more to work with than 'a cool city'. Then say what must not change. Listing the people, their faces, their clothing and the camera path protects the parts of the shot you want to keep.
Keep the new scene plausible for the camera move you filmed. A slow push-in suits an interior; a fast tracking shot along a street suits an outdoor setting. If the edit drifts, shorten the instruction and run it again rather than adding more detail.
Limits to know
Very fast camera moves, people walking in front of each other and tiny subjects in a wide shot can produce artifacts in any video model. Genjutsu AI changes settings and objects; it does not swap a real person's face or put someone's likeness into a performance. Upload only footage and images you have permission to use.