Replicate black-forest-labs/flux-pro
The API-only top Flux tier in ComfyUI
- IMAGE
Flux Pro is the top tier of Black Forest Labs' Flux family - the uncompromised base model, before the guidance distillation that produces Dev and Schnell. It is API-only: you can't download the weights and run it locally, so a node like this is genuinely the only way to use it from ComfyUI. This one calls it through Replicate's cloud, so you need a Replicate account and a REPLICATE_API_TOKEN, and you pay per image - but there's nothing to install and no hardware requirement.
Why bother over Dev or Schnell? Quality, mainly, and one specific thing our Flux notes call out: the distilled tiers (Dev, Schnell) have real facial-variety problems - the "same face" look - and Pro doesn't, because it's the undistilled model those two were derived from. If you want Flux at its best and cost isn't the deciding factor, this is it.
How it works
Because Pro is the full base model, it has proper step and guidance controls, unlike the distilled tiers. You still don't get a negative prompt (that's a Flux-wide thing), but you do get to trade steps for quality and tune how hard the guidance bites. It's the most "dial-able" of the Flux tiers exposed in this pack.
The inputs and outputs that matter
prompt(required) - natural language; Flux parses full sentences and complex scenes well.aspect_ratio(1:1) - pick a ratio like16:9,2:3,9:16.steps(25, up to 50) - more steps, more refinement (and more cost). 25 is a good balance.guidance(3, range 2–5) - prompt adherence. Higher sticks closer to the prompt but can over-bake; our Flux notes lean toward the lower end for natural results.safety_tolerance(2, range 1–5) - how permissive the content filter is. Higher is more lenient, but it's capped - you can't fully disable it.
Output is a single IMAGE. interval (2) nudges variety/guidance behavior, seed fixes the result, and force_rerun bypasses Replicate's cache when identical inputs would return the same image.
How to install it
One install covers every Replicate node.
- ComfyUI Manager: search ComfyUI-Replicate, install, restart.
- Manual:
then restart.cd ComfyUI/custom_nodes git clone https://github.com/replicate/comfyui-replicate cd comfyui-replicate pip install -r requirements.txt
No weights download - Pro is closed and runs on Replicate. Only the replicate pip package installs. Set your token first: export REPLICATE_API_TOKEN="r8_..." (replicate.com/account/api-tokens).
Common issues & troubleshooting
No negative prompt field. Right - Flux has no CFG-style negative on any tier, Pro included. Steer entirely with the positive prompt. This trips up people coming from SDXL.
It costs more per image than Dev or Schnell. That's the tier. Pro is the premium option; if you're doing volume or drafts, generate on flux-dev (or flux-schnell) and reserve Pro for finals. And if you're weighing Pro at all, look at flux-1.1-pro in this same pack - it's the newer, faster, generally cheaper successor and usually the better pick.
A prompt got filtered. safety_tolerance raises the ceiling but is capped, and Flux models are censored at the weights level too, so some prompts won't go through regardless.
Over-baked, contrasty results. Ease guidance down toward 2–3. High guidance on Flux tends to crunch the image; lower guidance with a well-written prompt usually looks better. Same image on a re-run means the Replicate cache fired - change the seed or set force_rerun.
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | — | |
| aspect_ratioopt | COMBO | 1:1 | 7 options: 1:1, 16:9, 2:3, 3:2, 4:5, 5:4, +1 |
| stepsopt | INT | 251–50 | — |
| guidanceopt | FLOAT | 3.002–5 | — |
| intervalopt | FLOAT | 2.001–4 | — |
| safety_toleranceopt | INT | 21–5 | — |
| seedopt | INT | — | |
| force_rerunopt | BOOLEAN | false | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |