arkennemasis fal · Flux 2 Pro · $0.03 + $0.015/MP

Flux 2 Pro is the tier BFL actually sells to production users: released alongside Dev on 25 November 2025, API-only, and still the reference point people compare new image models against. You cannot download it. This node is the text-to-image endpoint, at $0.03 for the first megapixel of output and $0.015 for each additional megapixel of input and output, rounded up.

First, a warning about the description

The node's own blurb reads "Image editing with FLUX.2 [pro] … style transfer and sequential editing workflows", which is fal's copy for the edit endpoint of the same model - the descriptions on the two pages are effectively swapped. Trust the schema instead: this node has no image socket at all. It's text-to-image. If you need the editor, that's the ArkFal_fal_ai_flux_2_pro_edit node, which does have image_1 through image_6.

The inputs

prompt, and here it's worth writing properly. Flux 2's encoder is an LLM reading a chat turn: sentences with explicit ownership and spatial relations beat keyword stacks, and CLIP-era weighting syntax is disabled. image_size defaults to landscape_4_3 with the usual presets plus custom for explicit image_size_width and image_size_height. Because billing is per megapixel rounded up, a 512×512 output still costs you a whole megapixel - the pricing example is blunt about it: it's $0.03 whether you ask for 0.25MP or 1MP.

seed at -1 lets the model decide. output_format defaults to jpeg; switch to png for anything you'll re-edit. safety_tolerance is a 1–5 combo - 1 strictest, 5 most permissive, default 2 - with enable_safety_checker next to it, and disabling the checker requires authorisation on your fal account.

The shared options behave as on every fal node in this pack: max_cost_usd (default $20, 0 disables) is checked before anything is uploaded; reuse_identical_run hands back an identical already-paid result rather than charging twice; extra_json merges raw JSON for fal fields the node doesn't expose; max_concurrent (default 1) limits simultaneous paid calls across the graph. Note there's no num_images on this node either, so one run is one image - batch by looping.

Outputs

images and image_paths are the result, seed comes back for reproducibility, and info is the JSON receipt with the request id, cost estimate, saved files and fal's full response. No prompt passthrough, no timings - Pro and Max have leaner output sets than the Klein and [dev] nodes in this pack.

Install

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

Restart, then:

FAL_KEY=your-key-from-fal.ai/dashboard/keys

in the .env next to run_nvidia_gpu.bat. The fal provider needs no Python packages, so there's nothing to install for this node specifically.

Where people get burned

The tier confusion, first: if your image socket is empty and the node still complains, you may be on the wrong one of the Pro pair - the descriptions are swapped on fal's pages, and the schema is the only reliable guide. Second, price shape. $0.03 for the first megapixel then $0.015 after means 1920×1080 is $0.045 and 2MP-ish outputs stay cheap; but you're still paying per attempt with no batch discount, so fifty exploratory renders is a couple of dollars.

Third, the honest question of whether you need Pro at all. For most work Klein at $0.011–0.015 a megapixel is the better buy, and the local Klein models are free after electricity. Pro earns its rate on prompt comprehension and composition quality - and it's the tier whose weights will never be downloadable, which is exactly why renting is the only option. Where a hosted filter would refuse the job, stay local; no API tier is going to route around its own moderation.