FLUX 1.1 [ultra] Finetuned
Your finetune at the no-dials flagship
- image_prompt
- IMAGE
FLUX 1.1 [ultra] Finetuned is the strangest combination in this pack on paper and one of the most useful in practice: the dial-less flagship model - no steps, no guidance, no resolution boxes - now running a model you trained. You get ultra's "type a sentence, get a magazine cover" behavior, except the cover features your subject. For anyone who's trained a BFL finetune, this is the "wow" endpoint.
Same rule as every Finetuned node here: you train through BFL's fine-tuning API at api.bfl.ml, not in ComfyUI. Training happens on their GPUs, you get a finetune_id, and this node consumes it. The interesting part is how much of ultra's interface survives the finetune - almost all of it, because the finetune is a full model replacement rather than a nudge. Your character inherits ultra's quality ceiling, not the other way around.
Inputs that matter
- prompt (required) - describe the scene; the finetune pulls it toward your subject.
- finetune_id (required) - your trained model's ID. Nothing works until this is valid.
- finetune_strength (0–2, default 1.2) - slightly hotter than the 1.0 finetuned nodes' 1.1 default. Raise it when your subject comes out generic; ease it down when it overwhelms the prompt.
- aspect_ratio (default
"1:1") - ultra's replacement for width/height. Set"16:9","9:16", whatever the shot needs. - raw (default off) - skip BFL's extra post-processing for a more natural look.
- image_prompt + image_prompt_strength (0–1, default 0.5) - remix a reference image; strength controls how much survives.
- output_format (default
png), safety_tolerance (1–5, default 5), seed.
Output is one IMAGE tensor.
The catch you should hear twice
Ultra is already the most expensive tier in the pack, and adding a finetune on top doesn't make it cheaper - training bills separately, and every generation through this node bills at flagship rates. It's also the least controllable finetuned node: no guidance, no steps, nothing to tune when a result comes out wrong except finetune_strength and the prompt. You're betting that ultra's baked-in defaults serve your subject well. For most character-consistency work they do - this is genuinely the nicest-looking output in the pack. But if you need dials, the 1.0 pro Finetuned node is the controllable sibling.
Install
Part of the official Black Forest Labs API Nodes pack, one-time setup: ComfyUI Manager → search the pack title, or git clone https://github.com/black-forest-labs/bfl-comfy-nodes into custom_nodes, restart, and set BFL_API_KEY (env var, bfl_api_key.txt beside the pack, or api_key_override). No local models, no heavy dependencies. Note this one also lives under the Flux category, not "Flux Finetuned," same as the fill-finetuned sibling - search for "ultra" if you can't find it in the menu.
Inputs (12)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | — | |
| finetune_id | STRING | — | |
| finetune_strengthopt | FLOAT | 1.200–2 | — |
| image_promptopt | IMAGE | — | |
| aspect_ratioopt | STRING | 1:1 | — |
| rawopt | BOOLEAN | false | — |
| safety_toleranceopt | INT | 51–5 | — |
| seedopt | INT | 00–4294967294 | — |
| image_prompt_strengthopt | FLOAT | 0.500–1 | — |
| output_formatopt | STRING | png | — |
| api_key_overrideopt | STRING | — | |
| regionopt | COMBO | EU1 | 2 options: EU1, US1 |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |