FLUX 1.0 [pro] Finetuned

FLUX 1.0 [pro] Finetuned is what people usually mean when they pay for BFL's fine-tuning: the flagship base model, now skewed toward a subject you trained. This is the endpoint you reach for when a LoRA isn't cutting it - when you want a full model that genuinely knows your character or product, running at Pro quality, and you're fine with the bill that implies.

Here's the context that makes this node make sense. Locally, the Flux ecosystem runs on LoRAs - small add-ons that nudge a base model - because training a full 12B finetune is out of reach for consumer rigs. BFL's fine-tuning API does that full-model training on their GPUs. It costs more, but the result is a different beast: a dedicated model, not a nudge. This node is the consumption side of that. You train at api.bfl.ml, get a finetune_id, paste it in, and generate.

Inputs that matter

• prompt (required) - content; the finetune pulls it toward your subject.
• finetune_id (required) - your trained model's ID.
• finetune_strength (0–2, default 1.1) - the influence dial. This is the one you'll actually tune: generic output → raise it; subject overpowering the prompt → drop it.
• guidance (default 2.5, range 1.5–5) - same as base Pro.
• steps (1–50, default 40) - note the range is tighter than base Pro's 100; the finetuned endpoint tops out at 50.
• width / height (256–1440, step 32) - explicit resolution control.
• image_prompt + image_prompt_strength (0–1, default 0.1) - remix an image; the low default strength keeps the reference subtle.
• output_format (default jpeg), prompt_upsampling, safety_tolerance (0–6, default 2).

Output is one IMAGE tensor.

Install, key, and the honest cost picture

Same pack, once: ComfyUI Manager → Black Forest Labs API Nodes, or git clone https://github.com/black-forest-labs/bfl-comfy-nodes into custom_nodes, restart, authenticate via BFL_API_KEY, bfl_api_key.txt, or api_key_override. No local models anywhere in the path.

The cost picture is the thing to keep front of mind: training the finetune bills, every generation bills, and the flagship base model doesn't bill like a discount tier. This is a production-tool node, not a toy. Where people get burned: treating finetune_strength as a set-and-forget (it defaults to 1.1 and some subjects need a real nudge toward 1.5+), and assuming the API's moderation cares about your fine-tune's training set - safety_tolerance still gates what you can prompt. If you're experimenting, the canny/depth/fill finetuned siblings are cheaper places to validate a finetune before you spend flagship credits.