What It Does
LatentUpscale resizes the latent tensor before it passes to the sampler or decoder. This is a key component of hi-res fix workflows: generate at a lower resolution first (faster, better composition), then upscale the latent and run a second sampling pass at the target resolution.
Because the upscaling happens in latent space rather than pixel space, the subsequent sampling pass can add genuine detail rather than just interpolating existing pixels.
Inputs
samplesLATENTInput latent to resize.
upscale_methodSTRINGInterpolation method.
widthINTTarget width (in latent pixels, 8× for image pixels).
heightINTTarget height.
cropSTRINGCrop mode.
Outputs
LATENTLATENTResized latent.
What Numonic Captures
- Target latent dimensions
- Upscale method
Known Gaps
- Source latent dimensions
- Effective pixel-space dimensions (8× latent)
Related Nodes
Capture ComfyUI metadata automatically
Numonic extracts workflow metadata from every ComfyUI generation — models, samplers, seeds, prompts, and custom nodes. Track provenance, maintain compliance, and never lose a workflow.