😃 SML Flux Pro Ultra 1.1 (Replicate)

Same pack as the plain Flux Pro node, same trick, one tier up. This node calls FLUX 1.1 Pro Ultra - Black Forest Labs' higher-end API-only model - through Replicate, so nothing heavy runs on your machine. Think of it as the "money shot" button: better fidelity on fine detail, faces, and in-image text than the standard Pro tier, at a higher price per image. If the standard node covers your volume work, Ultra is what you pull out when one render really has to land.

Like its sibling, it's a thin remote wrapper. Queue the node, it sends your prompt to black-forest-labs/flux-1.1-pro-ultra, waits for the cloud run, downloads the returned PNG, and hands you a normal IMAGE tensor. Your GPU does nothing. Everything about install, config, and failure behavior is shared with the Pro node - one clone gives you both.

What's different from the Pro node

Fewer knobs, not more. Ultra drops width/height, guidance, and prompt_upsampling entirely - you get aspect ratio, steps, quality, and two flags. The things a beginner actually sets:

• prompt (multiline) - full sentences, not tags; that's how you get the most out of Flux's text encoder.
• aspect_ratio - twelve presets, and this is the fun part: it includes ultrawide 21:9 and tall 9:21 that the Pro node doesn't offer. Pick custom if you need something else.
• raw - the one genuinely interesting control. Off by default, which means the model applies its own finishing pass (upsampling plus its default aesthetic polish) before you ever see the image. Flip it on for the raw model output - useful if you want to run your own upscaler or a specific LoRA-style post-process instead of trusting BFL's.

steps defaults to 30, output_quality to 100, and there's the same optional seed (default -1 = random). Output is a single IMAGE tensor - straight into Save Image, no VAE decode.

Install

This ships in the same pack as SML_FluxPro_Replicate_Standalone, so if you already installed that one, you already have this node. Otherwise:

cd ComfyUI/custom_nodes
git clone https://github.com/smlbiobot/ComfyUI-Flux-Replicate-API
cd ComfyUI-Flux-Replicate-API
pip install -r requirements.txt

Then create your key at replicate.com/account/api-tokens and:

cp config.ini.example config.ini
# edit config.ini, set REPLICATE_API_TOKEN = your-key-here

That's it - no model files, no heavy deps (just torch and replicate==1.0.4). ComfyUI Manager users can search the pack title instead of cloning.

Gotchas

Everything that bites the Pro node bites this one harder, because each Ultra call costs more:

• Black image = silent failure. On any API error the node catches it, prints Error generating image with FluxPro to the console, and returns a blank 1024×1024 frame so the graph keeps running. A black square is your cue to read the console, not to re-queue blindly - check the token in config.ini, billing on the account, and rate limits.
• config.ini must sit inside the pack folder. Easy to lose track of when you have custom nodes everywhere.
• The node blocks while the cloud runs. A single Ultra generation takes tens of seconds, and the queue waits.
• raw=True means the finishing pass is on you. The polished default is a big part of why Ultra looks premium; skip it and you may need your own upscale step to get back to poster quality.

Use Pro for the batch, Ultra for the one that matters - and remember it's a paid API call either way, so it's a convenience for people who want top-tier cloud output without the VRAM stack, not a replacement for local generation.