Descriptions:
ComfyUI shipped an update adding basic inpainting support to Z-Image Turbo ControlNet 2.0, and Veteran AI moves quickly to demonstrate the new capability with a working workflow. The update introduces a dedicated node called Z Image Fun Controlnet, which differs from the node used for standard ControlNet 2.0 pose work — it accepts three inputs: a control image (pose map or edge map), the original image to repaint, and a mask defining the region to modify.
The presenter builds the workflow from scratch since no official example was available at time of recording, and walks through each component: loading Z-Image Turbo main model, applying ControlNet 2.0 as a model patch (stored in the model_patches directory, not the controlnet folder — a common point of confusion), drawing a mask in ComfyUI’s mask editor, and configuring sampling with 8 steps, CFG 1.0, Euler sampler, and Simple scheduler. A pixel-level drawing node is included to improve inpainting quality. The demonstration target is adding glasses to a character, and the result shows clean local edits without disrupting the surrounding image.
Viewers are advised to treat this workflow as a community stopgap until an official workflow is released, and to update to the latest ComfyUI commit to access the new node. The video is a direct follow-up to earlier coverage of ControlNet 2.0 that flagged the missing inpaint feature, making it especially useful for anyone who has been waiting for this specific capability before adopting Z-Image Turbo ControlNet 2.0.
📺 Source: Veteran AI · Published December 18, 2025
🏷️ Format: Tutorial Demo
