📸 Bawk Image Loader

Bawk Image Loader is the img2img gateway of the pack: click-to-browse image loading with preprocessing that gets your reference image ComfyUI-ready. The whole pitch of Bawk Nodes is "fewer nodes, fewer clicks," and this is where that shows up most - you never hand-type a path again. It's the node you reach for when you want to start from a photo, a phone pic, or a previous render instead of a blank latent.

How it works

PIL loads the file. auto_orient reads EXIF orientation and rotates phone photos correctly - the thing a default loader quietly gets wrong (most phone shots are landscape-in-EXIF, actually rotated). target_size (Original/512/768/1024/1536/2048) resizes while keeping the aspect ratio, using resize_method - Lanczos by default, the sharp choice for downscaling. pad_to_square letterboxes with black borders if you need exact squares for a particular model. normalize_colors maps values to the 0–1 range for FLUX compatibility and is on by default.

The inputs that matter

• image - the picker itself. Click it and the file browser opens; no typing.
• target_size - resize big reference photos so your sampler doesn't try to work at 4000px and melt your GPU.
• create_mask - generates a mask output from the image's alpha channel (white = opaque, black = transparent), handy for inpaint-style or masked img2img work.
• auto_orient - leave on unless you have a pipeline that expects the raw EXIF.

Outputs and wiring

It outputs image, mask, filename, width, and height. In the pack's img2img chain you wire image into Bawk Sampler's input_image socket and set denoise below 1.0. The README's cheat sheet: 0.3–0.5 keeps the original structure, 0.6–0.7 is a real style change, 0.8–0.9 is a major transformation, and 1.0 is a full replacement (effectively text-to-image). width/height are handy if you want to force the sampler's resolution to match the source.

Install

Same pack install - ComfyUI Manager → search "Bawk Nodes" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/juddisjudd/ComfyUI-BawkNodes.git

It needs Pillow, which stock ComfyUI ships, so no extra dependencies. The pack's requirements.txt (torch, torchvision, safetensors, numpy, requests) adds nothing you don't already have.

Gotchas

• Resizing happens before mask generation, so the mask always matches the resized image. That's the behavior you want, but know it.
• On a load error the node returns a 64×64 placeholder with the error message in the filename output - you'll spot it immediately, and it won't hard-crash your workflow.
• normalize_colors is tuned with FLUX in mind, but since ComfyUI normalizes image tensors anyway, leaving it on won't hurt an SD1.5/SDXL chain either. No real downside.

One genuinely nice touch: this is the only loader in the pack, and it slots into both the FLUX and Z-Image chains unchanged - same four sockets, same behavior. Load a reference, pick your denoise, go.