Flux 1.1 [pro] Ultra Image

Flux 1.1 Pro Ultra is BFL's top-shelf text-to-image API - the "Ultra" here refers to quality and speed tier, not resolution, and it was one of the very first partner nodes to ship in ComfyUI core back in May 2025. If you've ever wanted the best Flux render money can buy without running the 12B stack locally, this is the node: prompt in, finished image out, no sampler settings, no VRAM.

The shared rules for the whole partner/image/BFL family apply, so let's get them out of the way in one go: this is a paid remote call through Comfy's proxy (api.comfy.org → BFL), billed from the credits on your Comfy account, with no bring-your-own-key and no model files to download. The node posts your prompt, polls a status endpoint while you watch the progress bar, and pulls the finished PNG back into a normal IMAGE tensor. Nothing about that loop is under your control, which is the point - the API is a black box and that's the deal you're making.

What you do control is refreshingly small, and that's a feature. prompt is the whole game. aspect_ratio takes a string like 16:9 and validates it's between 1:4 and 4:1 before it ever leaves your machine. seed is best-effort determinism. raw (default off) is the one genuinely interesting switch: flip it on and the model skips its heavy aesthetic post-processing, giving you flatter, more "naturally lit" images - handy when you're going to re-style or upscale downstream and want raw material rather than a finished look.

There's also an optional image_prompt with an image_prompt_strength (default 0.1) that blends your reference image with the prompt - an image-to-image mode in the same node. It's the least-used input and the most surprising to find on the "Ultra" text-to-image tier, but it works the same as you'd expect from Flux: low strength means the prompt dominates, higher means the reference does.

The one input to leave alone is prompt_upsampling - it rewrites your prompt for "more creativity" and explicitly makes results nondeterministic. On a premium-priced model where you presumably wrote a decent prompt, letting the server rewrite it is usually a downgrade.

Where does it fit in a workflow? Honestly, as a source node for everything else: a prompt + Ultra render feeding an upscaler or an inpainting pass gives you a quality ceiling that most local stacks can't touch on anatomy, hands, and in-image text. The cost is real credits per image, and BFL's prices aren't fixed in stone - the badge on the node is the current number, and it changes. For hero assets, campaigns, or anything where the render is the product, that's often a fair trade. For churning out a hundred variants, you'll want to think twice and probably land on a local model instead.