FLUX 1.0 [depth] Finetuned

FLUX 1.0 [depth] Finetuned is the depth node for people who've trained a model of their own. Same geometry-preserving trick as the plain depth node - feed a reference image, keep the spatial layout, regenerate the content - but the content now runs through a model you trained on BFL's servers, so a specific face, character, or product rides along into the scene.

Like every Finetuned node in this pack, the rule is: train elsewhere, consume here. The training happens through BFL's fine-tuning API at api.bfl.ml - their console, their GPUs, their pricing. What you get back is a finetune_id, a short string you paste into this node. There is no training UI in ComfyUI, no dataset folder to point at. If you haven't trained anything yet, this node will just error on the empty finetune_id until you have one.

Inputs that matter

• prompt (required) - the content you want; the finetune pulls it toward your subject.
• control_image (required) - the reference whose depth structure is preserved.
• finetune_id (required) - your trained model's ID.
• finetune_strength (0–2, default 1.1) - how hard the finetune dominates. Raise it when your subject comes out generic; lower it toward 1.0 when the finetune is overpowering the prompt.
• guidance (default 15, range 1–100) and steps (15–50, default 50) - the same levers as the plain depth node.
• output_format (default jpeg) - these finetuned endpoints default to JPEG; switch to png for lossless.

Output is a single IMAGE tensor. It's otherwise identical in shape to everything else in the pack, so it drops into any workflow that expects a standard image.

Install, key, and the usual warnings

Same pack, same one-time setup: ComfyUI Manager → Black Forest Labs API Nodes, or git clone https://github.com/black-forest-labs/bfl-comfy-nodes into custom_nodes, restart, and authenticate with BFL_API_KEY, a bfl_api_key.txt beside the pack, or api_key_override. No local models - both the base model and your finetune live server-side.

Two cost layers to keep straight: training the finetune bills separately, and every generation through this node bills per image on top. Add BFL's moderation on the input side (safety_tolerance 0–6, default 2) and you have a capable but not cheap pipeline. It earns its keep when consistency across scenes matters more than the bill - same character, any depth layout, no local training stack to maintain.