Note that in these examples the raw image is passed directly to the ControlNet/T2I adapter. Each ControlNet/T2I adapter needs the image that is passed to it to be in a specific format like depthmaps, ...
This repo is the official implementation for the paper MATA: A Trainable Hierarchical Automaton System for Multi-Agent Visual Reasoning, accepted at ICLR 2026. MATA (Multi-Agent hierarchical Trainable ...