Stay up to date with the latest AI research, models, tools and behind-the-scenes stories at the intersection of AI and VFX.
Ersan Musa, Previs Supervisor, built Shot-to-Set: give it one shot from a film, and a chain of ComfyUI workflows breaks it down, identifies the characters, the wardrobe, the vehicles and the set, rebuilds them in 3D and assembles the scene in Unreal Engine, so you can put on a VR headset and stand on the set. This article covers how the pipeline works and where a human stays in control.
Read more →Kevin Bertelli, CG Supervisor and real-time specialist, built two Unreal Engine plugins on top of ComfyUI: one turns a text prompt into a 3D asset in the scene, with Z-Image Turbo and Pixal3D; the other captures the camera view, restyles it with Qwen Image Edit and projects the result back onto the geometry, so you can move around a watercolour or comic book version of the scene in Unreal. This article covers how each plugin works and what they are for.
Read more →Julien Chateau, CG Instructor, built a storyboard tool that turns a rough sketch into a clean storyboard frame, with the characters and the framing left where the artist put them. It runs in the browser on ComfyUI workflows, with a LoRA he trained for Qwen Image Edit. It is made for artists who are not comfortable with drawing, and for those who want to iterate faster with the same quality on every frame.
Read more →Rudyard Cretenet, 2D Supervisor, had never used a generative AI tool before the Bootcamp. He came out with three pieces of work: architecture studies for a Paris that never was, made with Z-Image Turbo and Qwen Image Edit 2511; test shots for a film set in Belgium, reworked with Qwen Image Edit 2511; and Goosebumps, a one-minute film made with LTX 2.5 about an AI learning human emotion. This article covers each one, and what a compositor's eye brings to these tools.
Read more →Alexandre Grégoire, Technical Director, built an interactive AI image installation controlled by hand gestures, entirely in ComfyUI: custom nodes read the webcam, the menus are made of nodes, and the images are generated live by image-to-image on an RTX 3060. This article covers how it is built, the one rule that governs the interface, and what the jury said about bias and guardrails.
Read more →