Flux Pro (TOGETHER)

Here's a fun stat: this node has zero Google impressions. Zero. Nobody is searching for it, which is a shame, because "Flux Pro (TOGETHER)" is the one node in this pack that gives you the original FLUX.1-pro - the pre-1.1, pre-distillation-to-one-step Pro - with actual knobs to turn. It's the middle child of ComfyUI-FLUX-TOGETHER-API: the pack's free "Dev" node gets all the attention because it costs nothing, and the 1.1 node gets all the hype because it's the newest. This one just sits there being the original.

Why you'd reach for it

FLUX.1-pro is BFL's base model before the guidance distillation that produced Dev and Schnell - the uncompromised version, available only through APIs like Together's. Compared to its younger sibling in the same pack, this node exposes real generation controls: a working steps range, a guidance scale, and an interval parameter the 1.1 node doesn't offer at all. If you like to actually tune your generations instead of accepting a locked one-step pipeline, this is the one you want.

The inputs that matter

• prompt - multiline, natural language. Pro is a strong T5-driven model; write sentences.
• steps - 1–40, default 4. This is a real range, unlike Pro 1.1's locked 1. The original Pro behaves more like a conventional diffusion model here; more steps buys quality, but it's still a paid call, so find the floor that looks good to you.
• guidance - 0.1–10, default 2.5. This is the closest thing to a CFG knob on a FLUX API node, and the KB's community consensus is that FLUX guidance defaults skew too high - start around 2 and only push up if adherence is weak.
• interval - 1–10, default 2. The Together API's CFG interval. Honestly: if you don't know what you're doing with it, leave it at 2. It's the parameter nobody explains in the README, and messing with it tends to cause more confusion than benefit.
• prompt_upsampling - boolean, default on. Lets the API expand your prompt; leave it unless you need exact wording.
• safety_tolerance - 1–6, default "2". Together's moderation strictness.
• width / height - 256–1440, and the code throws if they aren't multiples of 32. Stick to 1024×1024.
• seed - optional, default -1 (random). Set it to reproduce an image you're paying for twice.

Output is a single IMAGE tensor, same as every node in the pack - wire it into SaveImage or pass it onward.

How it works

Same shared machinery as its siblings: the node reads your API key from config.ini into TOGETHER_API_KEY at import, spins up the together SDK, and calls images.generate() against black-forest-labs/FLUX.1-pro. The base64 response is decoded to a PIL image, converted to numpy, then to a torch tensor. On failure it returns a black 512×512 image - the pack's signature silent failure - so a black output means check the console, not your workflow.

Install

Standard for the 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

Create config.ini with [API] and API_KEY = your_key_here. Watch the trap: the README's config.ini.example names the key together_api_key, but the code reads API_KEY - copy the format above. Restart ComfyUI after editing (the key is loaded once at import). ComfyUI Manager finds it as "ComfyUI-FLUX-TOGETHER-API" too. Nothing heavy downloads; the together SDK is the only real dependency.

Honest verdict

It's a paid node (every render burns Together credits), the pack is tiny and basically unmaintained, and - as those zero impressions suggest - you'll be using it entirely on your own initiative. But if you want the original Pro with a steps slider and a guidance dial, this is the only place in ComfyUI you'll find it wired up this simply. It works, it's simple, and it costs you a few cents a shot. Sometimes that's the right trade.