arkennemasis fal · Flux Pro Kontext Max · $0.08/image

Kontext Max is the same job as the Pro node - image plus instruction, edited image out - at $0.08 per image instead of $0.04. The difference fal sells is prompt adherence and typography: Max follows fiddly instructions more literally, and it renders text in the image without the garbled letterforms that plague most models.

Whether that's worth double is genuinely use-case dependent, and you should test both before deciding. For "swap the shirt colour" the cheap one is fine. For "change the sign to read OUT OF OFFICE and keep the typeface" Max is the one doing the work.

The inputs

• prompt - the edit instruction.
• image - the reference the model edits, in context.
• aspect_ratio - (not set) by default, with 21:9 … 1:1 … 9:21 available when the output shape has to be forced.
• guidance_scale 3.5 (1–20), seed, num_images 1–4 (each one billed), output_format (jpeg/png).
• enhance_prompt, default off - fal rewrites your prompt before sending, which helps with three-word instructions and hurts when the exact wording was deliberate. safety_tolerance, default 2, options 1–6, strictest to most permissive.

Then the shared four that every fal node in this pack carries: max_cost_usd (default $20, and the check happens before anything is uploaded, so a mis-set size or count can't quietly bill you), reuse_identical_run (on by default - identical inputs and settings return the already-paid files, which is also why a repeat run after a crash is free), extra_json (a JSON object merged into the request last, for any fal field this node has no box for), and max_concurrent (default 1: one paid fal call at a time across the whole run).

Outputs

images, image_paths (under output/fal/fal-ai_flux-pro_kontext_max/), seed, has_nsfw_concepts, prompt, timings and info. The info string is where the request id and cost estimate live - worth grabbing when you want to look up a run in fal History later, for free.

Install

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

Manager users: search comfyui-arkennemasis. Either way, restart ComfyUI and look under arkennemasis/fal/Image/Flux. No weights to download, and no extra pip packages for the fal provider - requirements.txt's only real entry is httpx, which the pack's Codex nodes need.

# .env, in the folder holding run_nvidia_gpu.bat
FAL_KEY=your-key-from-fal.ai/dashboard/keys

Honest notes

Text rendering still isn't free. Max is better at typography, not reliable at it. Short words and clean sans-serif lettering work; long strings, curved surfaces and intricate faces still come back with a letter or two subtly wrong. Check every character before you ship.

Instrument clutter survives. Kontext-era editing is strong at swapping one region and weak at tidying a whole scene. If your instruction implies "and also remove the cables", it'll likely do half of it.

Anatomy at full-body scale is the family's known weakness - heads and hair suffer when a whole person fits the frame. Crop, edit, composite.

Four images means four times the bill, and every iteration bills again. At eight cents a pop, keep num_images at 1 while you're still deciding what you're asking for. The live badge on the node is the number to trust, not the title.