ComfyUI Node
BD Lotus-2 Predict
Run a loaded Lotus-2 model on an image. For depth: outputs a 3-channel depth map (the model produces RGB-encoded depth — channels are nearly identical), a [0,1] normalized linear depth, and a colorized preview. For normal: outputs the normal map directly (R=X, G=Y, B=Z, mapped to [0,1]). process_res caps the longest input edge before inference (1024 is the model's training resolution — going higher costs VRAM with marginal quality).
BD Lotus-2 Predict
- model
- image
- mask
- map
- raw_linear
- colorized_preview
◄num_inference_steps10►
◄process_res1024►
◄guidance_scale3.5►
◄timestep_core_predictor1►
◄bg_fillauto►
Category🧠BrainDead/Depth
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| model | LOTUS2_MODEL | — | |
| image | IMAGE | — | |
| num_inference_steps | INT | 101–50 | Detail sharpener denoising steps. Default 10 from upstream. Quality plateaus around 10-15 for most images. |
| process_res | INT | 1024512–2048 | Cap longest input edge at this resolution. Output is resized back to input size after inference. |
| guidance_scale | FLOAT | 3.50–10 | — |
| timestep_core_predictor | INT | 10–1000 | Core predictor stage's effective timestep (default 1). |
| maskopt | MASK | Optional subject mask (e.g. from BD Remove Background). BEST PRACTICE: feed the FULL original image to `image` (so the model keeps scene context + detail and avoids silhouette halos), and pass the subject mask here to clean the background AFTER. Subject=1, background=0. | |
| bg_fillopt | COMBO | auto | How to fill the background where mask=0: auto — DEPTH→black (far), NORMAL→#8080ff (flat, facing camera) black/white/normal_neutral — force that fill transparent — output RGBA (alpha=mask) keep — don't mask. (NEVER use white for depth bg, or black/white for normal — they're invalid far/flat values.) |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| map | IMAGE | — |
| raw_linear | IMAGE | — |
| colorized_preview | IMAGE | — |