Flux.1 Kontext [pro] Image

If you've spent any time around ComfyUI, you know Flux.1 Kontext as the 12B open-weights editor that made "change the clothing, swap the background, delete the watermark" a one-sentence operation. This node is the pro tier of that same model - the API-only version that never got weights, and never needed them. BFL keeps the good stuff on the server, and ComfyUI wraps it as a built-in partner node so you can edit from inside your graph instead of tabbing over to the BFL playground.

Why reach for the API tier when the open Kontext dev weights exist? Three reasons people actually do: it runs on nothing but your internet connection (zero VRAM, zero model download), it's the un-distilled Pro base rather than the guidance-distilled dev, and it slots into the same workflow as your local nodes - generate a character locally, push a Kontext edit pass through this node, send the result onward. The flip side is it costs money per run and the filter is BFL's.

How it works

Kontext is an in-context editor: the model sees your image and your sentence together and rewrites the pixels to match, rather than diffusing from a masked area. That's what gives it its famous character consistency across successive edits. The node sends your prompt, settings, and (optionally) an input image to BFL's API and returns a finished IMAGE.

Note: despite being marketed as an editor, this node works without an input image too - leave input_image unconnected and it's a plain text-to-image generator on the Kontext base.

The inputs that matter

• prompt - say what and how to edit. "Make it rain" is a prompt; "keep her jacket and face identical, replace the background with a neon city at night" is the kind of sentence this model loves.
• input_image - optional. Feed it and you're editing; leave it empty and you're generating.
• guidance - default 3, ranges 0.1–99. This is the Flux-style guidance scale, not CFG; nudge it up for stricter prompt adherence.
• steps - default 50, up to 150. More steps cost more time (and the API bill is per run, not per step, so within reason this is free to crank).
• aspect_ratio - a string like 16:9; anything between 1:4 and 4:1 is accepted.
• prompt_upsampling - BFL's prompt auto-embellisher. On, the prompt gets rewritten for "more creative generation" and output becomes nondeterministic (same seed ≠ same image). Defaults off for a reason.

Output is a single IMAGE - wire it to Preview/Save or downstream into anything image-shaped.

Cost, setup, and gotchas

It's a partner node: ships with ComfyUI core, needs a logged-in Comfy account with credits, works best on a current ComfyUI. Runs cost about 8.4 credits each, and there's no charge attached to steps or guidance, so tune freely. Two practical notes: the API polls for completion, so long runs show up in the UI as "waiting" - that's normal, not a hang. And because it's a hosted model, BFL's content moderation applies to everything you send it; the "request moderated" failure is a real state, and there's no local override. If you want the unfiltered, own-the-weights version, that's the Kontext dev checkpoint and a local workflow - not this node.