CIVITAI / Workflows

Bernini-R Reference Video Conditional Editing Workflow

Watch the full video first if you want to understand how this Bernini-R reference video conditioning workflow works in practice. The video shows how a source video and a reference video can be connected into one controlled editing pipeline, where the reference video becomes part of the edited scene while the original video structure remains stable. This ComfyUI workflow is designed for Bernini-R reference video conditional editing. Its main purpose is to take an existing source video and use another video as a visual condition, allowing the workflow to insert, propagate, or integrate the reference video content into the target scene. Compared with simple text-to-video generation, this is a more controlled video editing workflow because it uses both source video structure and reference video content at the same time. The workflow is built around the Bernini-R high-noise and low-noise model route. It uses Bernini_HIGH_fp8_e4m3fn_scaled.safetensors and Bernini_LOW_fp8_e4m3fn_scaled.safetensors as the dual model branches. It also uses UMT5 XXL fp8 text encoding, Wan 2.1 VAE, BerniniConditioning, LoadVideo, GetVideoComponents, KSamplerAdvanced, VAEDecode, CreateVideo, and SaveVideo. The model branches are also patched through PathchSageAttentionKJ, which helps the workflow run the Bernini route with a more optimized attention setup. The key node is BerniniConditioning. In this workflow, BerniniConditioning receives both the source video and the reference video. The source video provides the base scene, motion, camera structure, timing, and audio. The reference video provides the visual content that should be inserted or used as the editing condition. This is different from a simple reference image workflow, because the reference input itself contains temporal motion and video information. The example prompt in this workflow is designed around a reference video appearing as an F1 championship video looping on a billboard in a street scene. This is a very clear use case: the source scene keeps its own environment and camera framing, while the reference video becomes embedded as a dynamic screen, billboard, or video surface inside the scene. This makes the workflow useful for screen replacement, billboard insertion, in-scene video advertising, dynamic poster replacement, and reference-video-driven scene editing. The generation route uses a two-stage KSamplerAdv