ComfyUI Node

🖼️ LDM VAE Decode

Decodes a latent tensor back into pixel space using the f8 VAE decoder. The latent is divided by the training scale factor (0.18215 for f8 VAE) before decoding. If the output appears too dark or washed out, adjust vae_scale. Output: IMAGE tensor in [0, 1] range, BHWC layout.

By Winlensky·Created a day ago·Updated about 12 hours ago· 1
🖼️ LDM VAE Decode
  • samples
  • vae
  • IMAGE
vae_scale0.1822
CategoryLDM

Inputs (3)

NameTypeDefaultDescription
samplesLDM_LATENTLatent tensor from the Sampler node.
vaeLDM_VAEVAE output from Load LDM Checkpoint.
vae_scaleFLOAT0.18220.01–1Latent scaling factor. Standard f8 VAE value is 0.18215. Adjust if output brightness looks incorrect.

Outputs (1)

NameTypeDescription
IMAGEIMAGE