Apply Flux ControlNet

This is the simpler sibling of Apply Advanced Flux ControlNet, and the entire difference between the two is that this one has no start/end step controls - the ControlNet condition applies at a constant strength for the whole sampling run instead of being windowed to a portion of it. If you don't yet know why you'd want that windowing, you probably want this node; if you've already read the Advanced version's article and specifically need to fade the condition in or out, use that one instead.

What it's doing: ControlNet conditions generation on spatial structure pulled from a reference image - edges, a depth map, a pose skeleton - so your prompt handles what appears and the ControlNet handles where it goes. XLabs shipped their own set of Flux ControlNets (separate per-type checkpoints - Canny, Depth, HED - rather than one combined model) right around Flux's August 2024 launch, making this one of the earliest ways to get structural control on Flux at all. Worth knowing before you spend an evening chasing perfect pose control: a community comparison against the other Flux ControlNets available at the time rated depth conditioning as consistently solid but openpose as noticeably weak, and judged the whole 2024 generation of Flux ControlNets as "still a long ways off" the quality SDXL's later union model reached. That's useful context, not a reason to avoid this node - depth and canny work well through it.

The inputs: controlnet is the loaded model from Load Flux ControlNet, image is your reference - already preprocessed into edges, depth, or whatever condition type your loaded checkpoint expects. strength (0–10, default 1) sets how hard the condition pushes; a community baseline from that same comparison thread landed around 0.6 alongside Euler sampling, Normal scheduler, and guidance 3.5, so treat the default of 1.0 as a ceiling to dial back from rather than a target to hit. There's also an optional controlnet_condition input for stacking - feed a previous Apply node's output in here to combine multiple ControlNets (canny plus depth is the common pair) into one conditioning object. The output, also controlnet_condition, only has one place to go: the optional controlnet_condition input on XLabs Sampler. A stock KSampler has no slot for it.

Installing it: search "x-flux-comfyui" in ComfyUI Manager, or cd ComfyUI/custom_nodes && git clone https://github.com/XLabs-AI/x-flux-comfyui and restart. You'll also want Fannovel16's comfyui_controlnet_aux installed if you need to generate your own canny/depth/pose maps from a plain photo rather than supplying an already-processed one - XLabs' own README lists it as a separate requirement, not something this pack bundles.

Where it goes wrong: the ControlNet nodes in this pack have a real-world reputation for rougher VRAM handling than ComfyUI's native equivalents - people report out-of-memory errors even with --lowvram set, and workflows that get stuck or won't free memory after a failed run. If import fails entirely with No module named 'comfy.ldm.flux', that's an out-of-date ComfyUI core rather than a broken node install - update ComfyUI itself first before troubleshooting further.