arkennemasis fal · Flux Pro Kontext Max Text to Image · $0.08/image

Kontext is usually described as an editor, which makes this node look odd: it's the Max tier with no image input at all. What you're buying is generation with Max's prompt adherence and typography, at $0.08 per image. If your prompts are long, specific, and full of things most models quietly ignore - exact counts, spatial relations, text that has to be legible - this is the endpoint built for that.

Inputs

• prompt - a real prompt. Flux-lineage models want sentences rather than tags, and the whole point of the Max tier is that it pays attention to the ones you write.
• aspect_ratio - default 1:1 here (unlike the multi nodes, which default to (not set)). Options run 21:9, 16:9, 4:3, 3:2, 1:1, 2:3, 3:4, 9:16, 9:21. Change it to 16:9 before you generate a wallpaper.
• guidance_scale 3.5 (1–20), seed (-1 lets the model pick), num_images 1–4, output_format (jpeg or png).
• safety_tolerance default 2, options 1–6, strictest to most permissive. Moderation, not quality.
• enhance_prompt, default off. When on, fal expands and rewrites your prompt before generating. Useful for short prompts; actively unhelpful when you've spent ten minutes choosing words.

Plus the pack's four shared widgets: max_cost_usd (default $20, checked before upload - nothing is sent if the estimate is over, and 0 disables it), reuse_identical_run (on by default; identical settings and media return the paid-for files instead of billing again), extra_json (JSON merged last into the request, for fal fields the node has no box for), max_concurrent (default 1 - one paid fal call at a time across every fal node in the run).

Outputs

images is what you wire into preview or save. image_paths gives the saved locations in output/fal/fal-ai_flux-pro_kontext_max_text-to-image/. seed returns the seed actually used, prompt echoes the prompt, timings and info are diagnostics - and info carries the request id plus the cost estimate and fal's full answer. has_nsfw_concepts appears here too, which is your first thing to check if an image comes back black.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Hishamahmer/comfyui-arkennemasis

Manager route: search comfyui-arkennemasis, or Install via Git URL. Restart ComfyUI afterwards; the node shows up under arkennemasis/fal/Image/Flux. No weights, no extra Python packages for the fal provider.

# .env, in the folder with run_nvidia_gpu.bat (or your ComfyUI folder)
FAL_KEY=your-key-from-fal.ai/dashboard/keys

There's no key box on the node, so nothing to leak into a workflow file.

The unglamorous parts

Typography is better, not solved. Max generates legible text where other models produce hieroglyphs, but long strings and tight letterforms still come back with a wrong character sometimes. Spell-check the output before you deliver it.

It's not cheap per idea. Eight cents an image means a prompt-hunting session is a real spend. The acceleration-style shortcuts other Flux nodes have don't exist here - there's no speed dial, just price and count. Keep num_images at 1 until a prompt is earned.

No steps, no guidance-based recovery from a bad prompt. You get guidance_scale and your words. If output is stiff or over-baked, rewrite the prompt rather than hunting for a setting, because there isn't one.

Black frames mean the safety checker, not a crash. Ease safety_tolerance up a step and re-read the prompt.