Fal API Flux ControlNet Union Config

This is the sibling to the plain ControlNet Config node in this pack, and unlike its sibling, its default settings actually work: path defaults to InstantX's FLUX.1-dev-Controlnet-Union on HuggingFace, which is a real, Flux-compatible union ControlNet - one checkpoint that covers several conditioning types rather than needing a separate model file per condition. Union packaging like this has been the standard way ControlNets ship since 2024, precisely because it turns "download a dozen files for a dozen conditions" into "download one file, pick a mode at inference time."

That mode-picking is control_mode, a required enum with seven options: canny, tile, depth, blur, pose, gray, and lq. Canny, depth, and pose are the ones you'll reach for most - hard edges, spatial layout, and body position respectively. tile is for detail-preserving upscale-style work, feeding a low-res version of an image back in so the model adds faithful detail rather than reinterpreting it. blur and lq (low quality) lean toward restoration-flavored jobs - deblurring and general quality recovery - and gray conditions on a desaturated version of an image, useful for colorization-style workflows. One important limit: this field takes a single value, not a list, so one Union Config node instance handles exactly one condition at a time. If you want canny and depth active simultaneously, that means two Union Config nodes - but the generation nodes in this pack that accept a CONTROLNET_UNION_CONFIG only expose one such input each, alongside one separate plain controlnet (CONTROLNET_CONFIG) input. In practice that caps you at two structural conditions per generation call, not more.

The rest of the inputs. control_image (required) is your reference image for whichever control_mode you picked. conditioning_scale runs 0.1–2, default 1 - worth dialing down from that default rather than trusting it as-is, since modern Flux-era union ControlnNets tend to recommend scales in roughly the 0.65–0.9 range per condition rather than the SD1.5-era default of 1.0. config_url and variant are optional strings for pointing at an alternate config file or a specific model build if you're using a different union checkpoint than the default.

Output is CONTROLNET_UNION_CONFIG, which plugs into the controlnet_union input on any of this pack's LoRA-and-ControlNet Dev nodes.

Installing it. No node-specific step: ComfyUI Manager (search "ComfyUI-Fal-API-Flux") or a manual clone of https://github.com/yhayano-ponotech/ComfyUI-Fal-API-Flux.git followed by pip install -r requirements.txt. It's a config node and issues no API call on its own, but whatever generation node it feeds still needs the pack's shared config.ini with a [falai] section and api_key to actually run.

Troubleshooting. If conditioning seems absent, first confirm control_mode actually matches what control_image contains - feeding a depth map while control_mode is set to canny will produce nothing useful, since the union checkpoint applies whatever processing that mode expects regardless of what you handed it. If the default path has been swapped for something else and results go strange, revert to the default InstantX union first to confirm the rest of your graph is sound before troubleshooting a different checkpoint.