ControlNet Loader Adv.

This is the grown-up version of the art-venture ControlNet Loader. Same one-box-instead-of-three convenience, but with the two things the basic one leaves out: it takes both your positive and negative conditioning, and it gives you a start/end percent to control when during denoising the condition applies. If you know why that second one matters, this is the loader you want.

Why the step range is the whole point

There's a piece of ControlNet folklore that goes back to February 2023 and still holds up: you usually don't want the condition gripping the image through the entire sample. Let it lock the composition early, then release it so the model can add its own detail. In the community's words at the time, the "Guidance Start" control was treated as a precision breakthrough, and the standing advice for structure-heavy work is still to drop the condition once composition has formed.

That's exactly what start_percent and end_percent do here. Start at 0.0, end at 0.5, and the ControlNet only steers the first half of denoising. Great for "I want this pose but don't fossilize every pixel." The basic loader can't do this - it holds the condition the whole way through.

How it works

You wire in positive and negative conditioning plus your reference image. The node loads the named ControlNet, preprocesses the image into a control map, and applies it to both conditioning streams over the step window you set. Out come two CONDITIONING outputs - positive and negative - that go straight to the matching inputs on your KSampler. One node, both streams, correctly wired.

The inputs that matter

• control_net_name - from your models/controlnet folder; shows None if you've got nothing installed. Match it to your checkpoint's architecture.
• strength - 0 to 10, default 1. Modern union models want 0.65-0.8, not the old 1.0. Overshoot and the image goes rigid and plasticky.
• start_percent / end_percent - the denoising window, each 0.0 to 1.0. Full range (0 to 1) mimics the basic loader. Narrow it to 0 → 0.5 for loose structural guidance.
• preprocessor - needs comfyui_controlnet_aux installed, or the list is just None.

resolution, enabled, control_net_override, and timestep_keyframe are all there too, and all optional. The timestep_keyframe input in particular ties into Advanced-ControlNet if you want keyframed strength schedules on top of the simple start/end window.

Installing it

ComfyUI Manager, search comfyui-art-venture, install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/sipherxyz/comfyui-art-venture

and restart. It's pre-installed on comfy.icu.

Common issues

Loaded a shared workflow and this node is red? That's the art-venture pack not being installed - it's a common silent dependency in community graphs. Manager → "Install Missing Custom Nodes" fixes it.

Empty preprocessor dropdown is the controlnet_aux dependency missing, same as with the basic loader. The pack README notes its old AV_controlnetPreprocessor is deprecated; use the AIO_Preprocessor from controlnet_aux instead.

One thing worth knowing before you reach for this over a plain ControlNet: if you're on a 2026 base and all you need is one condition applied at full strength, an edit model plus a control map does the same job with no checkpoint. Where this node earns its keep is partial strength and a timed window - an edit model has neither dial. That's the honest line on when ControlNet is still the right call.