PVL FluxPro Fill (fal.ai)

Flux Fill was BFL's answer to the oldest problem in image editing: you've got a spot you want gone - a person, a watermark, a crack in the wall - and you need the model to repaint what's behind it. This node runs fal-ai/flux-pro/v1/fill from inside ComfyUI: give it an image, a mask marking the region to regenerate, and a prompt describing what should be there, and it returns a filled version. Same job as a local inpainting pipeline, minus the model download and the VRAM.

It's part of the ComfyUI Assistant Node pack from pvlprk. The mechanism is the pack's standard fal flow, with one addition: here you're sending two images.

How it works

The node converts your image and your mask into base64 data URIs - the image as an RGB PNG, the mask as a grayscale PNG - and posts both to the fill endpoint. In the standard Flux Fill convention, the mask marks the area to regenerate, and the model fills that region guided by the prompt. The node just passes your ComfyUI mask through as-is, so whatever pixels your mask marks are what the API treats as the region to repaint.

The rest of the controls are familiar: seed (-1 for random), num_images (1–4), output_format, sync_mode, and safety_tolerance (1–6, default 2), which is how hard fal's filter pushes back.

The inputs that matter

• image - the photo you're fixing.
• mask - the region to repaint. This is the input that decides everything, so paint it carefully. The fill happens where the mask is; vague or sloppy masks give you sloppy fills.
• prompt - what should be in the filled area. "Remove the street sign and continue the brick wall" works a lot better than an empty prompt.
• seed - set it if you want to re-roll the same fill deterministically or compare variants.

Output: one IMAGE - the filled result, at the same resolution as the input.

Installing it

Part of pvlprk/comfyui-pvl-api-nodes. ComfyUI Manager (search "ComfyUI Assistant Node") or:

cd ComfyUI/custom_nodes
git clone https://github.com/pvlprk/comfyui-pvl-api-nodes

Set your fal.ai key as FAL_KEY (or FAL_API_KEY / FAL_CLIENT_KEY), restart, and make sure you have credits.

Where people get burned

The classic fill failure is a mask that's too tight or too loose. Too tight and the model has nothing to sample from at the edges; too loose and it repaints things you wanted kept - feathering the mask a couple of pixels usually fixes the seams. Don't expect object-level precision: Flux Fill is excellent at texture continuation and object removal, less magical at reconstructing a face or a logo it never saw. And as with every node in this pack, if you get a blank placeholder result instead of an image, the console is where the actual error message lives.