Nodes/Dobidop ComfyStereo/StereoDiffusion
ComfyUI Node

StereoDiffusion

A ComfyUI node in image/stereo with 17 inputs and 3 outputs.

By Dobidop·Created 2 years ago·Updated 3 months ago· 53
StereoDiffusion
  • image
  • depth_map
  • model
  • clip
  • vae
  • stereo_pair
  • left_image
  • right_image
scale_factor5.0
directionuni
deblurfalse
pipeline_modeFast (Warp + Inpaint)
guidance_scale3.0
num_inference_steps20
seed1337
null_text_optimizationtrue
denoise_strength0.60
model_idrunwayml/stable-diffusion-v1-5
inpaint_model_idrunwayml/stable-diffusion-inpainting
prompt
Categoryimage/stereo

Inputs (17)

NameTypeDefaultDescription
imageIMAGE
depth_mapIMAGE
scale_factorFLOAT5.01–20Controls the strength of the stereo effect (disparity)
directionCOMBOuniuni: unidirectional attention, bi: bidirectional attention
deblurBOOLEANfalseAdd noise to unfilled regions to avoid blurring
pipeline_modeCOMBOFast (Warp + Inpaint)Standard: DDIM inversion (high quality, slow). Fast: Warp image with depth, then AI-inpaint only the gap regions (fast, works with turbo/LCM models)
guidance_scaleFLOAT3.00–20CFG scale. Standard mode: 3-10. Turbo models: 0.0. LCM: 1.0-2.0
num_inference_stepsINT201–100Number of inference steps. Standard DDIM: 30-100 (default 50). Fast inpainting: 20-30. Turbo/LCM: 1-8
seedINT13370–18446744073709550000Random seed for reproducible results
null_text_optimizationoptBOOLEANtrueEnable null-text optimization for better reconstruction (Standard mode only)
denoise_strengthoptFLOAT0.600.1–1How much noise to add before denoising (Fast mode). Lower = preserve original more, Higher = more model creativity for filling gaps
modeloptMODELComfyUI MODEL input. Fast mode: use inpainting model (9ch UNet). Standard mode: use any SD1/SD2 model.
clipoptCLIPCLIP from Load Checkpoint
vaeoptVAEVAE from Load Checkpoint
model_idoptSTRINGrunwayml/stable-diffusion-v1-5Fallback HuggingFace model ID (Standard mode, when no ComfyUI model connected)
inpaint_model_idoptSTRINGrunwayml/stable-diffusion-inpaintingFallback inpainting model ID (Fast mode, when no ComfyUI model connected)
promptoptSTRINGOptional text prompt to guide inpainting (Fast mode). Describe the image content for better gap filling.

Outputs (3)

NameTypeDescription
stereo_pairIMAGE
left_imageIMAGE
right_imageIMAGE