Flux Kontext Max (BFL)

Flux Kontext Max (BFL) is the heavier sibling of Flux Kontext Pro in the ComfyUI-FLUX-BFL-API pack. Same instruction-based editing - image plus a sentence, model edits the whole frame - but "max" is BFL's top-tier rendering for the Kontext line. Think of Pro as the everyday editor and Max as the one you pull out when the output quality matters more than the credit cost. BFL's tiering pattern is consistent across everything: the best quality never leaves the API, and Max is that for Kontext.

Functionally, the two nodes are near-identical. The request shape is the same, the input sockets are the same, the output is the same single IMAGE. The difference is which endpoint it hits (flux-kontext-max instead of flux-kontext-pro) and what BFL returns for the extra money - higher-fidelity rendering, which is worth it for client work, hero shots, or any edit where you'll notice a smoother gradient.

How it works

Standard pack mechanics: you build a prompt and up to four base64 reference images, the node POSTs to the endpoint, polls get_result?id=... every 5 seconds, and hands back a finished IMAGE tensor. The whole image goes into context and a whole new image comes out - so unmasked regions come back close but not pixel-identical, same as Pro. If you need a specific region untouched at the bit level, that's mask-based editing, not Kontext.

The inputs

• prompt - the edit instruction, the whole ballgame.
• input_image + input_image_2/3/4 - up to four references for multi-image edits.
• aspect_ratio - none (default) or a fixed ratio if you want to change the frame.
• prompt_upsampling - let BFL expand the prompt internally. Usually leave off.
• safety_tolerance - 0–6, default 2. Moderation strictness.
• output_format - jpeg/png, default png.
• Optional seed, webhook_url/secret, and a config override.

Installing it

One repo, one key, same as every node in the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/gelasdev/ComfyUI-FLUX-BFL-API.git

Restart, put your key in config.ini or wire a Flux Config (BFL) node. No model downloads; the pack's only dependency is torch.

Where people get burned

• Picking Max for everything. You're paying a premium tier for quality you might not need. For throwaway exploration, use Pro or one of the Flux 2 nodes and keep Max for the shots that matter.
• Expecting a mask or a region control. Doesn't exist. Kontext is whole-frame regeneration.
• Black output. That's the pack's failure signal - moderation, API error, or the ~200s poll ceiling. Watch the [BFL] logs; every call prints a curl equivalent you can replay to see the raw API response.

The honest framing: if you're comparing Max against Pro and you can't tell the difference in your use case, you're paying for nothing. But for final deliverables - print, client proofs, anything where rendering quality is the point - the Max tier is the reason the node exists.