Flux Pro Kontext Text-to-Image (fal)

Kontext is best known as an image editor - feed it a picture and a sentence, get the edit. This node is the other mode: pure text-to-image, no input picture required. FluxProKontextTextToImage_fal calls Kontext's generation endpoint on fal, so you get the same model's aesthetic and prompt-following, generated from scratch. Type a prompt, get a native IMAGE.

Why have both? Because sometimes you want to start a project in Kontext's style rather than edit an existing frame - generate a base with this node, then hand it to FluxProKontext_fal for iterative, sentence-driven edits. The pair is a natural workflow: create, then refine. This one is the "create" half, and unlike its editing sibling it takes no image input at all.

The inputs that matter

• prompt - natural language, like all of Flux. This is the only required field; everything else is a default you can leave alone.
• aspect_ratio - defaults to 1:1, with the usual spread from 21:9 down to 9:21. This is your framing control since there's no explicit width/height here.

Optional: guidance_scale (3.5, range 1–20), max_quality for a higher-fidelity pass, num_images (up to 4), safety_tolerance (1–6), output_format (jpeg/png), seed, and sync_mode. Single IMAGE output, straight to Preview or Save Image.

guidance_scale is your main quality lever here. The 3.5 default is a reasonable middle; raise it if the model is drifting off your prompt, lower it if results look over-baked or contrasty. max_quality trades speed and cost for a cleaner render - leave it off while you're iterating on the prompt, flip it on for the final. And since there's no width/height, aspect_ratio is doing double duty as both framing and rough resolution, so pick it deliberately for the target format.

How it works

Nothing runs on your machine - it's a hosted Pro-tier endpoint, so there's no steps slider and no local model. fal renders and returns the image. The pack caches identical calls, so re-running the same prompt and settings is free and instant; force_rerun forces a fresh roll.

Installing it

ComfyUI Manager → search ComfyUI-fal-API → install → restart. Manual:

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

No weights to download. Set your key (config.ini, or export FAL_KEY=your_key; fal.ai/dashboard/keys), restart, find it under FAL.

Where people get burned

The main confusion is picking this node when you actually wanted to edit something. If you have a source image to modify, you want FluxProKontext_fal (the one with the image input); this text-to-image variant ignores any picture and generates fresh. Grab the right sibling for the job.

Beyond that: it's Kontext, so it carries Kontext's content policy - no NSFW, enforced at the endpoint, surfaced as a readable fal error if you trip it. And it's a paid call per generation; lean on the result cache while you're iterating on a prompt, and if you just want plain fast Flux text-to-image without the Kontext framing, FluxSchnell_fal or FluxPro11_fal may be cheaper picks.