Bernini R Main Replacement Video Editing Process

This is a subject-replacement video application based on Bernini-R + original video + reference subject image + LLM prompt rewriting + BerniniConditioning + two-stage video editing sampling. Users upload an original video, a reference subject image, and enter a simple replacement requirement. The system automatically replaces the subject in the reference image into the original video while maintaining the motion rhythm, camera angle, background environment, lighting, and composition of the original video as much as possible.

The biggest difference between this and ordinary Image-to-Video is: It does not regenerate a brand new video, but edits the video based on the original video.
The original video provides the motion, posture, rhythm, camera movement, and scene background; the reference image provides the identity and appearance of the new subject; the subject-replacement prompt tells the model who to replace, what to keep, and what not to change. The current default case is to replace the dancing woman in the original video with the woman in the reference image, making the new subject follow the actions and positions of the original video character while preserving the background, lighting, shadows, and camera position as much as possible.

The core advantage of this workflow is "reference subject + original video motion + video consistency editing".
If users only use regular Text-to-Video, it is difficult to stably replicate the motions and camera angles of the original video; if they only do face-swapping or masking composition, it is prone to unnatural edges, body mismatches, lighting breaks, and fake motions. Video editing workflows like Bernini-R are better suited for complete subject replacement: not just face-swapping, but also the character's overall clothing, posture, body movement, and scene blending.

Suitable scenarios include: AI character dance replacement, advertising model replacement, short video character replacement, apparel display videos, virtual model test footage, IP character replacement, plot material remixing, reference image protagonist replacement, batch secondary creation of character advertising materials, and brand promotional video protagonist replacement.
When using, it is recommended to focus on controlling three items: the original video determines the motion and background, the reference subject image determines the replacement target, and the subject-replacement requirements determine the replacement scope and retention rules. For better stability, the prompts should explicitly state: keep the original video background unchanged, keep the camera composition unchanged, keep the motion rhythm unchanged, do not add new characters, and do not change the scene.

When publishing as a RunningHub application, the frontend interface is recommended to be extremely minimalist: Upload original video + Upload reference subject image + Enter subject replacement requirements + Output video. Video frame count can be opened as a basic parameter; sampling steps and cutting points are better fixed in the backend to prevent novices from misadjusting and causing unstable subject blending.

🎁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 claim another 100 RH Coins~ For more ComfyUI workflows, tutorials, and gameplay, welcome to follow our official WeChat account (AIKSK), with simultaneous updates on Douyin / Bilibili / Xiaohongshu / YouTube (AI-KSK) ✨