Flux Union Controlnet Pro workflow

Use this ComfyUI workflow with Flux.1 D to apply Union Controlnet Pro to image generation: provide an input image, choose Depth, Pose, Canny, Tile, Blur, Grayscale, or Low quality, and produce preview and saved-image outputs.

Load it in ComfyUI, provide the input image, and select the mode in the SetUnionControlNetType node above the ControlNetLoader.

To extract control information from your own images, optionally install ControlNet Auxiliary Preprocessors; its openpose, depth, and canny extractors can replace the example image.

Input an image and choose one of seven modes: Depth, Pose, Canny, Tile, Blur, Grayscale, or Low quality.

The graph includes LoadImage, PreviewImage, and SaveImage nodes for loading, previewing, and saving images.

Update ComfyUI, download the Union ControlNet model to models/controlnet, and then load the workflow.

The listed model files are clip_l.safetensors, FLUX.1-dev-ControlNet-Union-Pro/diffusion_pytorch_model.safetensors, flux1-dev-fp8.safetensors, and t5xxl_fp8_e4m3fn.safetensors.

Named nodes include BasicGuider, BasicScheduler, CLIPTextEncode, ControlNetApplySD3, ControlNetLoader, DualCLIPLoader, EmptySD3LatentImage, FluxGuidance, KSamplerSelect, and LoadImage.

Other named nodes include ModelSamplingFlux, OpenposePreprocessor, PreviewImage, RandomNoise, SamplerCustomAdvanced, SaveImage, SetUnionControlNetType, UNETLoader, VAEDecode, and VAELoader.

In SetUnionControlNetType, map canny to openpose, tile to depth, depth to hed/pidi/scribble/ted, and blur to canny/lineart/anime_lineart/mlsd.

Map pose to normal, gray to segment, and low quality to tile in SetUnionControlNetType.

To tune the stated control influence, adjust strength and end_percent in ControlNetApply; the sample values are 0.4 and 0.6, and the stated strength range is 0.3–0.8.

Sugerencia · sin verificar

Before running, confirm that your ComfyUI setup includes the named nodes used by the workflow.

Sugerencia · sin verificar