Flux Pro 1.1 (TOGETHER)

Want FLUX.1.1 Pro - the model you literally cannot download, the one that's only ever existed behind an API - inside your ComfyUI graph? That's this node. "Flux Pro 1.1 (TOGETHER)" calls black-forest-labs/FLUX.1.1-pro on Together.ai's servers and hands you back a finished IMAGE tensor. Your GPU does nothing but display the result. That's the whole pitch, and it's a good one if you've hit the wall of trying to run FLUX locally on a 8GB card.

Pro is the undestilled tier of the FLUX.1 family - the base model before guidance distillation - and 1.1 was the quality bump on top of the original. This is where the hands work, the text renders, the anatomy mostly behaves. The KB's FLUX notes call it the baseline newer models get measured against, and while mid-2026 local models have overtaken it, you still can't beat it on convenience: one node, no 12B stack, no fp8 gymnastics.

The one-step thing

Here's the part that surprises people: the steps input on this node has min and max of 1. It's not a slider you tune - it's a lock. FLUX.1.1 Pro is distilled to run in exactly one step on the API, so the node doesn't even pretend to give you a choice. There's also no guidance knob, for the same reason: this model doesn't take a CFG-style scale the way the original Pro did. What you do get:

• prompt - multiline. Pro parses natural language beautifully; write sentences, not tag soup.
• prompt_upsampling - boolean, on by default. Lets the API silently expand your prompt. It's usually a free quality win, and people who turn it off do so to keep their exact wording.
• safety_tolerance - 1–6, default "2". Together's moderation dial; lower is stricter.
• width / height - 256–1440, but the code hard-checks that both are multiples of 32. 1024×1024 is the safe default.
• seed - optional, default -1 (random). Set it and you can reproduce a specific image, which matters when you're paying per call and finally got a keeper.

Output is a single IMAGE tensor - save it, feed it to an upscaler, run it through an img2img node, whatever your workflow needs.

How it works under the hood

Same shared plumbing as the rest of the pack (which is worth knowing if you touch more than one of these nodes): it reads your API key from config.ini into TOGETHER_API_KEY at import, instantiates the together Python SDK, and calls images.generate() with your prompt and a hardcoded model string. The result comes back base64-encoded, gets decoded to a PIL image, then a numpy array, then a standard torch IMAGE tensor. If the request fails, the node returns a black 512×512 image instead of erroring - that's the pack's silent-failure habit, so a black render means "check the console for Error generating image:," not "your workflow is broken."

Install

Shared with the whole pack:

cd ComfyUI/custom_nodes
git clone https://github.com/BZcreativ/ComfyUI-FLUX-TOGETHER-API.git
cd ComfyUI-FLUX-TOGETHER-API
pip install -r requirements.txt

Then write config.ini with [API] and API_KEY = your_key_here - note the README's config.ini.example uses the wrong key name (together_api_key), while the code reads API_KEY. Restart ComfyUI after; the key is loaded once at startup. ComfyUI Manager works too if you search "ComfyUI-FLUX-TOGETHER-API". No models download; the only real dependency is the together SDK.

What actually gets you

Two honest warnings. First, this is the paid node - unlike the pack's "Flux Dev" node (which secretly runs the free FLUX.1-schnell tier), every Pro 1.1 call burns Together.ai credits. Check their pricing before you batch-generate a hundred images. Second, the pack is small and barely maintained, with community presence essentially zero - this is a "it works until it doesn't" kind of node, not a community darling. Treat it as a convenient gateway to Pro 1.1, and keep your expectations about polish in check.