Fal API Flux Pro v1.1 Ultra

This is the odd one out in the whole pack, and the schema gives it away immediately: no width, no height, no num_inference_steps, no guidance_scale. Every other node here hands you the usual Flux dials; Ultra takes them away and gives you aspect_ratio, raw, and output_format instead. That's not a missing feature, it's a different endpoint - FLUX1.1 [pro] Ultra is Black Forest Labs' higher-resolution, higher-throughput Pro variant, and it's built around picking a shape and a look rather than tuning steps and CFG by hand.

What you set. prompt still does the usual work - full sentences, not tags, since Flux's T5 encoder reads natural language. aspect_ratio is a seven-option enum (16:9, 4:3, 21:9, 1:1, 3:4, 9:16, 9:21) that replaces manual pixel dimensions entirely - pick the shape, fal decides the exact resolution on their end. raw is a boolean, defaulting to true, and it's the parameter this node exists for: Ultra's raw mode is BFL's own answer to Flux looking too polished, aiming for something closer to an actual photograph than the airbrushed "AI glossy" look Flux defaults to. If your outputs feel too clean or too perfect, that's the first thing to toggle, not the prompt. output_format picks between jpeg and png for what fal renders and returns - png if you want a lossless file to build on, jpeg if file size matters more than the last bit of fidelity. num_images (1–4) and enable_safety_checker behave as everywhere else in the pack, and safety_tolerance is the same 1–6 content-filter dial (strictest to most permissive) that shows up on every Pro-tier node here - worth knowing it exists, and worth knowing BFL has quietly pulled this exact parameter from other endpoints before without warning. seed is optional, 0 for random.

Output is IMAGE, same as everywhere else.

What's missing on purpose. No LoRA input, same as the plain Pro nodes - Pro-tier weights aren't published, so there's nothing to fine-tune. And notice there's no strength or image input either: this is text-to-image only, not an img2img or editing node.

Installing it. No difference from the rest of the pack: ComfyUI Manager, search "ComfyUI-Fal-API-Flux," install, restart - or clone the real repo manually (the README's own git clone example has a stale placeholder URL, so copy the actual link from the GitHub page rather than the README verbatim) and pip install -r requirements.txt. Then the part that actually gates everything: a config.ini next to the node's Python files, a [falai] section, and your api_key. No key, no image, and the failure usually shows up as a console error rather than a helpful node message.

Troubleshooting. Given this node's own GSC numbers are effectively zero, it's rarely used and even more rarely written about, so lean on the general pack advice: confirm your fal.ai account is actually provisioned for the Ultra endpoint specifically, not just base Pro, since fal tends to gate newer variants separately. If aspect_ratio seems to produce a different final size than you expected, that's expected behavior - Ultra picks the exact pixel dimensions itself, it isn't taking your width/height from somewhere else, because there's nowhere else to take them from.