Bernini R Reference Background Video Replacement Process

This is a background replacement video application based on Bernini-R + original video + reference background image + LLM prompt rewriting + BerniniConditioning + two-stage video editing sampling. Users upload an original video, upload a target background reference image, and enter a simple replacement requirement. The system will automatically replace the environmental background of the original video with the scene in the reference image, while trying to keep the character subject, clothing, dance movements, position, camera composition, and original video rhythm unchanged.

The biggest difference between this and ordinary video-to-video is: It does not regenerate a brand new video, but performs background-level video editing based on the original video.
The original video provides the character's actions, motion rhythm, camera composition, and subject position; the reference background image provides the new spatial environment, furniture settings, lighting atmosphere, and scene texture; the background replacement prompt tells the model: only change the background, do not change the character, do not change the actions, and do not destroy the camera structure. The current default case is to replace the video background where the dancing woman is located with a bright, warm, modern living room environment, while keeping the character continue the dance movements from the original video.

The core advantage of this workflow is "reference background + original video action + scene-consistent editing".
If you only use ordinary video repainting, it is very easy to change the character as well; if you only do traditional matting to change the background, the character's edges, shadows, lighting, and spatial relationships tend to look fake. Bernini-R's reference background video editing workflow is more suitable for complete scene replacement: not only replacing the background, but also trying to naturally blend the character with the lighting, shadows, perspective, and tone of the new environment.

Suitable scenarios include: changing scenes for dance videos, short video background replacement, live stream clip environment changing, virtual studio shooting, advertising scene modification, home background replacement, e-commerce character video site changing, brand promotion video remastering, AI model scene migration, and social media video secondary creation.
When using it, it is recommended to focus on controlling three things: the original video determines the character and actions, the reference background image determines the new scene, and the background replacement requirements determine what to keep and what to change. If you want it to be more stable, the prompt should explicitly write: keep the character's appearance unchanged, keep the clothing unchanged, keep the actions unchanged, keep the camera composition unchanged, only replace the background, and let the lighting blend naturally.

When publishing as a RunningHub application, the front-end is recommended to be minimalist: Upload original video + Upload reference background image + Fill in background replacement requirements + Output video. Video frame numbers can be opened as basic parameters; sampling steps and cut points are recommended to be fixed in the backend to prevent user misadjustment from causing the character to be changed, unstable background fusion, or screen flickering.

🎁Claim RH Coins first, then go experience the workflow! Avatar in the upper right corner → Invitation Code → Enter [rh-v1111] to instantly claim 1000 RH Coins, and log in daily to get another 100 coins~ For more ComfyUI workflows, tutorials, and gameplay, welcome to follow the official WeChat account (AIKSK), with simultaneous updates on Douyin / Bilibili / Xiaohongshu / YouTube (AI-KSK) ✨