I’m trying to build a straightforward image generation workflow that incorporates the latest PULID face identification technology alongside multiple ControlNet models. I’ve been experimenting with different approaches but I’m running into some challenges when combining these components.
The main issue I’m facing is understanding how to properly integrate PULID’s face recognition capabilities while simultaneously using several ControlNet models for different aspects of the generation process. I want to maintain consistent facial features from a reference image while also controlling pose, depth, and other elements through additional ControlNets.
Has anyone successfully implemented a similar setup? What’s the best way to structure this kind of multi-model pipeline? I’m particularly interested in knowing the optimal order for applying these different controls and any potential conflicts I should watch out for when combining PULID with other conditioning methods.