Load ControlNet Model (Online)

Load ControlNet Model (Online) is the pack's wrapper for ControlNet loading: pick a ControlNet from the catalog, the node downloads it into models/controlnet, and you get a CONTROL_NET object to plug into your conditioning. It maps one-to-one onto ComfyUI's core ControlNetLoader, so if you know that node you know this one - the difference is purely that the model list comes from DoubTech's server and downloads on demand.

The catalog is the biggest in the pack, and it's worth reading before you pick: all 20 entries are SD 1.5-era ControlNets. You get the control_v11 family - ip2p, shuffle, tile, depth, the lineart_anime one - plus the classic openpose pieces (body_pose_model.pth, hand_pose_model.pth), mlsd_large, dpt_hybrid (depth), network-bsds500 (scribble), and a control_v1p_sd15_qrcode model that fits the author's clear fondness for QR-code conditioning. Note what's missing: nothing for SDXL, nothing for Flux. ControlNets are architecture-locked - the KB's ControlNet essay is blunt that no SD-era ControlNet file loads on a modern base, and vice versa.

The one input and where the output goes

control_net_name is the only input (the dropdown). The single output is CONTROL_NET, and it wires into ControlNetApplyAdvanced (or ControlNetApply in older workflows), which takes your positive/negative conditioning plus a control image. Two things that wiring requires:

• A preprocessed control image - for canny/depth/pose you need the matching preprocessor, which lives in the separate ComfyUI ControlNet Auxiliary Preprocessors pack. The KB calls that pack functionally required for any serious ControlNet workflow.
• An SD 1.5 checkpoint to pair with these models. Pick the SD 1.5 entries in the pack's own checkpoint catalog (Deliberate v2, DreamShaper, A-Zovya RPG) - not the SDXL ones.

How it works

Standard pack machinery: the list is fetched from api.aiart.doubtech.com/comfyui/controlnet at startup, the chosen file downloads into models/controlnet on first use, then it's loaded with ComfyUI's stock sd.load_controlnet. Once the file is local, no more network.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/yolanother/DTAIComfyLoaders

or search "DTAIComfyLoaders" ("Comfy UI Online Loaders") in ComfyUI Manager, then restart. Only dependency is requests.

Where people get burned

• Pairing a ControlNet with the wrong base. This is the big one. These are SD 1.5 weights; feeding one to an SDXL or Flux workflow silently produces wrong output or outright errors. Check your checkpoint before you blame the node.
• The preprocessor step. The node loads a ControlNet; it does not generate the edge/depth/pose map for you. No Auxiliary Preprocessors installed → no control image → nothing happens.
• Dropbox HTML trap. The catalog files live on Dropbox, and scripted downloads can come back as an HTML page. A "downloaded" ControlNet that fails to load is usually a few-KB HTML file; grab the real file manually and drop it in models/controlnet.
• First-run download, no progress bar. The node sits there while it fetches; ControlNets are ~1.5 GB each.

This is actually one of the more useful nodes in the pack if you're building an SD 1.5 ControlNet workflow from scratch - the catalog genuinely covers the classics. Just keep your base model on SD 1.5 and bring your preprocessors.