FLUX 1.0 [depth]

Depth conditioning is the "geometry stays, everything else may change" control. Where canny pins down edges, FLUX 1.0 [depth] pins down spatial layout - the foreground stays in front, the background stays behind, the perspective doesn't move. It's the API version of FLUX.1 Depth, BFL's official depth-conditioned model from the Flux Tools lineup, and it's about as low-friction as structural control gets.

The thing that surprises people coming from local ControlNet: you feed it a normal photo, not a depth map. The node sends your control_image to BFL's server, the server estimates depth from it, and generates within that depth structure. No MiDaS or Depth Anything model to download, no preprocessing chain to wire up. The pack is one file of API glue, and it shows.

How it works

The node base64-encodes your image and POSTs it to the v1/flux-pro-1.0-depth endpoint alongside the prompt, then polls until the result is ready and decodes it back into a ComfyUI IMAGE. There's no local inference anywhere in the path - your GPU stays idle and BFL's servers do the depth estimate and the diffusion. That's the trade you're making: zero VRAM cost, per-image API cost.

Inputs that matter

• prompt (required) - what you want placed into the preserved spatial structure.
• control_image (required) - the reference whose depth layout constrains the result.
• guidance (default 15, range 1–100) - lower than the canny node's default 30, because depth structure is already strong and needs less shoving. If the output feels rigidly welded to the source layout, ease it down.
• steps (15–50, default 50) - quality vs. spend.
• safety_tolerance (0–6, default 2) - moderation dial, with the usual impact on prompt freedom.
• seed - reproducibility, API-caveats apply.

Output is one IMAGE tensor - Save Image, Preview, or wire it onward.

What it's actually good for

Relighting a scene, moving or swapping objects while the camera stays put, re-texturing a room, or generating a clean render that respects a real photo's perspective. Depth is the forgiving structural control: it doesn't demand crisp contours the way canny does, so organic subjects work fine. If you've ever wanted the "before" of a 2D-to-3D parallax shot or a scene rebuild without rebuilding the geometry yourself, this is the node.

Install

Part of the official Black Forest Labs API Nodes pack - install once, all twelve nodes come along. ComfyUI Manager, search the pack title, or:

cd ComfyUI/custom_nodes
git clone https://github.com/black-forest-labs/bfl-comfy-nodes

Restart, then authenticate: export BFL_API_KEY=..., a bfl_api_key.txt in the pack folder, or the api_key_override field. Get a key at api.bfl.ml. No downloads, no heavy dependencies - the whole pack runs on what ComfyUI already ships.

The honest caveat is the same one every node here carries: it's a paid call, so budget test images, and if the depth layout comes back wrong there's no local fallback to tweak - you're trusting BFL's depth estimator. For most scene-structure work it's solid, and you'll never fight a dependency error to get there.