FLUX 1.0 [pro]

FLUX 1.0 [pro] is the one model in the FLUX.1 family you cannot download, and this node is the normal way to reach it. Pro is the uncompromised base model - the version before guidance distillation, before the shortcuts that make Dev and Schnell fast. When you want the full-strength generation and you're willing to pay per image and wait for it, this is the endpoint.

That's the real pitch. Local FLUX users spend their whole lives quantizing Dev to fit a consumer card; Pro never had that problem because it never left BFL's servers. Every generation here runs on their infrastructure at whatever quality the full model produces. It's the closest thing to "just throw compute at it" in this pack.

How it works

The node POSTs your prompt (and optional reference image, base64-encoded) to the v1/flux-pro endpoint, polls until the result is ready, and returns a standard ComfyUI IMAGE. Like the dev node, prompt_upsampling defaults to true, so the API enriches your wording server-side - write short and let it expand. The one big difference from dev: because Pro isn't guidance-distilled, you get a lot of rope - steps up to 100 - and it actually uses it.

Inputs that matter

• prompt (required) - natural language. Pro rewards detail.
• guidance (default 2.5, range 1.5–5) - this is the pre-distillation model, so guidance behaves the way it does on an undistilled base: the community's "3.5 is too hot" advice from the dev era applies even harder. Start at 2.5.
• width / height (up to 1440, step 32) - explicit control, unlike the dial-less ultra node.
• steps (10–100, default 50) - the headroom is real. For hero shots people push well past 50; for tests, 30 is plenty.
• interval (1–10, default 1) - guidance interval, leave it alone.
• image_prompt (optional IMAGE) - remix an existing image.
• prompt_upsampling (default true) - off if you want your wording untouched.
• safety_tolerance (0–6, default 2).

Output is one IMAGE tensor - Save Image or wire it onward.

Install

Same one-time pack setup: ComfyUI Manager → Black Forest Labs API Nodes, or git clone https://github.com/black-forest-labs/bfl-comfy-nodes into custom_nodes, restart, then set BFL_API_KEY (env var, bfl_api_key.txt beside the pack, or the api_key_override field). Key from api.bfl.ml. No model downloads, no extra dependencies - the pack is pure API glue.

Where to be careful

Cost compounds fast here, because you have more dials than the ultra node and every step bills. The classic mistake is leaving steps at 100 out of curiosity and discovering the credit meter after one image. Also note the licensing: Pro is API-only by design and stays that way - this node is the sanctioned access path, not a loophole. If your workflow needs less control and a fixed bill, the 1.1 generation's pro node is faster; if it needs zero dials, that's ultra. Pro 1.0 is for when you want the base model's full quality and you're OK paying for every bit of it.