Bernini-R in ComfyUI: video editing workflow tutorial

This tutorial explains a Bernini-R workflow in ComfyUI for image and video editing on your own computer. The presenter describes ByteDance's inference code and model weights as open source and walks through separate high-noise and low-noise models, prompts, reference inputs and sampling settings.

The loader includes guidance controls and options intended to reduce VRAM use. Sage Attention requires installation before its patch node can work. Memory needs vary with resolution and task; the presenter points viewers with limited VRAM toward a 1.3B model in the original repository rather than giving a fixed minimum.

Image examples cover cartoon styling and adding a top hat. A 2x upscale VAE produces a 2560x2560 image, though the presenter says it works less well for video. The conditioning node supports up to eight reference images, with four connected in this workflow.

Video examples include hair-color changes at 1280x720 and 24 FPS, face replacement, inserting reference footage into a laptop screen, and changing clothing and backgrounds. The final example uses a 161-frame input and reports 321 frames at 32 FPS after interpolation. These examples illustrate the presenter's results, rather than guaranteeing the same output or speed on other hardware.