Nodes/Forbidden Vision/Forbidden Vision 💎 Refiner
ComfyUI Node

Forbidden Vision 💎 Refiner

A ComfyUI node in Forbidden Vision with 20 inputs and 2 outputs.

By luxdelux7·Created about a year ago·Updated 23 days ago· 97
Forbidden Vision 💎 Refiner
  • latent
  • vae
  • image
  • refined_latent
  • refined_image_preview
enable_upscalefalse
upscale_modelSimple: Bicubic (Standard)
upscale_factor1.20
neural_correctortrue
corrector_tone1.00
corrector_color0.70
depth_relightfalse
depth_relight_stringth0.30
depth_dof_enablefalse
depth_dof_strength0.40
depth_dof_focus0.75
depth_dof_modelV2-Small
maintain_aspect_ratiotrue
enforce_mod32_boundariestrue
use_tiled_vaefalse
tile_size512
offload_models_to_cputrue
CategoryForbidden Vision

Inputs (20)

NameTypeDefaultDescription
enable_upscaleBOOLEANfalse
upscale_modelCOMBOSimple: Bicubic (Standard)1 options: Simple: Bicubic (Standard)
upscale_factorFLOAT1.201–8
neural_correctorBOOLEANtrue
corrector_toneFLOAT1.000–1
corrector_colorFLOAT0.700–1
depth_relightBOOLEANfalse
depth_relight_stringthFLOAT0.300–1
depth_dof_enableBOOLEANfalse
depth_dof_strengthFLOAT0.400–1
depth_dof_focusFLOAT0.750.5–0.99
depth_dof_modelCOMBOV2-Small2 options: V2-Small, V2-Base
maintain_aspect_ratioBOOLEANtrueIf Enabled: Crops edges slightly to fit grid without stretching.
enforce_mod32_boundariesBOOLEANtrueForces output resolution to be a multiple of 32.
use_tiled_vaeBOOLEANfalse
tile_sizeINT512256–2048
offload_models_to_cpuBOOLEANtrueMove neural corrector and depth models from VRAM to RAM after processing. Frees GPU memory for other nodes (e.g. checkpoint swaps in queued workflows). Disable if you're chaining many refiner passes and want to skip the small reload overhead.
latentoptLATENT
vaeoptVAE
imageoptIMAGE

Outputs (2)

NameTypeDescription
refined_latentLATENT
refined_image_previewIMAGE