nyckm/bruxos-do-vfx-nodes

This is a set of custom nodes for ComfyUI, a visual workflow tool for AI image and video generation. These nodes were built by a Brazilian VFX studio (Bruxos do VFX) for real film production work, including two feature films. The focus is on video editing tasks: removing unwanted objects or people from a scene, face swapping, upscaling footage in batches, and comparing before/after results side by side. The nodes are designed to work with Wan 2.2 and the Bernini AI video model, which can handle video inpainting (filling in areas of a frame). A key concern throughout is memory efficiency: several nodes offer a sequential processing mode specifically for long videos or machines with limited GPU memory, where the video is processed in overlapping chunks rather than all at once. A comparison tool for the two available modes (sequential vs. context window) is documented in the README. The node collection includes tools for loading and saving video files with more codec control than the default ComfyUI options, a built-in A/B comparison player with side-by-side and wipe views, a point-based selection tool for targeting specific regions in a frame (more stable than text prompts for tracking), face upscale stitching that puts a sharpened face back into the original frame, and a variety of mask processing utilities. Installation involves copying the folder into ComfyUI's custom_nodes directory and downloading several large model files from Hugging Face (text encoder, VAE, the base model weights, and a LoRA). The README recommends using a command-line download tool rather than a browser to avoid corrupted downloads. A batch script is included for easier setup on Windows. The documentation is in Portuguese, reflecting its origin as a studio-internal tool. The nodes target people already working in ComfyUI who are doing video compositing and need more control over the object removal and inpainting pipeline than the standard nodes provide.