FLUX.1 Kontext [pro]

If the FLUX.1 Kontext [dev] node is the open-weights tier with every dial exposed, [pro] is its adult sibling: same instruction-editing model - feed an image, say "make it snowing," get a new image - but with the sampling knobs taken away and a couple of hosted conveniences added. It's Black Forest Labs' flagship editing model running entirely on Runware's cloud, and the node's widget list reads like a statement about who the audience is.

What you get instead of the knobs

Required: positivePrompt, width, height. Then the shortened optional list:

• referenceImages - the image you're editing. This is the whole game: Kontext edits from a reference rather than from scratch.
• providerSettings.bfl.promptUpsampling - a hosted convenience. Flip it on and BFL's server rewrites and expands your prompt before generation. Worth trying once; some people love the quality lift, others find it drifts from their intent.
• providerSettings.bfl.safetyTolerance - BFL's moderation tolerance, 0–5, default 2. Lower is stricter, higher is more permissive. This is the knob you'll actually touch when a legitimate prompt keeps getting refused - but know that BFL's moderation is famously firm, and this is a tolerance dial, not a bypass.
• seed and numberResults (up to 20) - reproducibility and variation.

No steps, no scheduler, no CFG. The pro tier decides those for you, which is honestly fine for editing - the model is tuned well and most people never needed the sampling knobs on Kontext anyway. Output is one image.

Why pro and not dev

The [dev] tier in this family ships open weights under a non-commercial license; pro is the commercial, hosted flavor. If you're producing client work or anything you might sell, pro removes the license question entirely - you're renting the model, not redistributing weights, so the non-commercial clause never gets a chance to bite you. That's the cleanest reason to pick pro over its dev sibling, and it's the same argument that pushed a lot of the editing world toward this pack.

Install and gotchas

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

Restart, or install "Runware" from ComfyUI Manager, and set your API key (Settings → Runware API key or RUNWARE_API_KEY). No weights download.

The shared pack gotchas apply: every run bills the account (title bar shows cost), and - the one that catches Kontext users specifically - remember this model edits the whole image, so unchanged regions come back close-but-not-identical. If bit-exactness matters, composite the edited region back over the original with a mask. Also, promptUpsampling changes outputs between runs in ways you can't fully control; leave it off if reproducibility matters more than prompt polish.