Fal API Flux Pro Redux

Redux is Flux's answer to the "image prompt" job that IP-Adapter did for SD1.5 and SDXL: instead of extracting edges or depth, it reads what's actually in a reference image and conditions generation on that content, producing variations rather than structural copies. It's one of the official Flux Tools Black Forest Labs shipped in late 2024, and this node is fal.ai's hosted Pro-tier version of it.

There's a real difference worth flagging between this and the native ComfyUI Redux workflow. People running Redux locally quickly learned it needs careful weight-tuning to be usable with a prompt at all - one well-known community writeup found the default local workflow just produced near-copies of the input, and that anything above roughly 0.04–0.07 image-prompt strength would "overwhelm the prompt and just produce a variation" of the source rather than something new. This fal-hosted node doesn't expose that strength dial at all - there's no weight or blend parameter in its inputs. So if you find the output leaning too heavily on your reference image and ignoring your prompt, you don't have that specific lever here; your options are a more assertive, more specific prompt, or lowering guidance_scale to give the prompt relatively more pull. It's a real limitation next to the native node's finer control, and worth knowing going in rather than discovering by trial and error.

Inputs. image (required) is the reference to condition on. prompt still matters here and should read as a sentence - this node isn't purely image-in, image-out the way some Redux setups are; text genuinely steers the result alongside the reference. width/height default to 1024×1024 in steps of 8, num_inference_steps defaults to 28 (1–100), guidance_scale defaults to 3.5 (0.1–40), num_images batches 1–4, enable_safety_checker toggles the content filter, and safety_tolerance is a 1–6 enum running strictest to most permissive on fal's documentation - a dial BFL has pulled from other endpoints without warning before, so treat it as advisory rather than permanent. seed is optional.

Output is IMAGE.

What's absent. No LoRA input, consistent with every Pro-tier node in this pack - Pro's weights aren't published for anyone to fine-tune. If you need Redux-style variation plus a custom LoRA, that combination doesn't exist anywhere in this pack; the Dev-tier nodes here don't expose a Redux-equivalent image-conditioning input at all, only img2img via strength, which is a related but distinct job (img2img denoises from your image; Redux conditions on your image's content while starting fresh).

Installing it. ComfyUI Manager, search "ComfyUI-Fal-API-Flux," is simplest. Manual installers clone the real repo - https://github.com/yhayano-ponotech/ComfyUI-Fal-API-Flux.git, not the placeholder URL in the README's own example - and run pip install -r requirements.txt, then create a config.ini with a [falai] section and api_key in the pack's folder. Every node in this pack fails the same way without it.

Troubleshooting. If output looks like a near-identical copy of your input, that's the known Redux behavior described above, not a bug - try a more forceful prompt before anything else. Beyond that, the usual suspects: a missing config.ini, an account without Pro-tier access, and the ComfyUI console for fal's real error text.