Extensions/Muse-Upscale-Advanced
ComfyUI Extension

Muse-Upscale-Advanced

Advanced upscale-model node for ComfyUI — tiled, OOM-safe processing with an exact target resolution (multiplier or fixed), no third-party dependency.

By muse-collective-26·Created about 17 hours ago·Updated about 17 hours ago· 0
muse-collective-26/Muse-Upscale-Advanced
Nodes1
On cloudLocal install
CategoryMuse/Image
Stars0
Updatedabout 17 hours ago
Readme

Muse Upscale Advanced

One node: Upscale Model Advanced (Muse) — load an upscale model, run it (tiled and OOM-safe), then resize the result to an exact target resolution (a multiplier of the original input, or a fixed width/height).

No third-party dependency — only ComfyUI's own core internals.

Features

  • Any model from models/upscale_models
  • Exact output size: either a multiplier of the original input, or a fixed width/height
  • Tiled, memory-safe processing — blended tile edges, and automatically shrinks tile size further on a CUDA out-of-memory error instead of failing outright
  • Precision control (auto / fp32 / fp16 / bf16) and batch-size control for VRAM
  • Model offloading after use, and an optional result cache for repeated calls with identical inputs

Inputs

| Widget | What it does | |---|---| | upscale_model_name | Model file from models/upscale_models | | use_fixed_resolution | On: use fixed_width/fixed_height. Off: use output_multiplier. | | output_multiplier | Final size relative to the original input (1.0 = same size, 2.0 = double) | | fixed_width / fixed_height | Exact target size when use_fixed_resolution is on | | tile_count | Starting tile grid per side (1 = single pass) | | precision | Processing precision for the model pass | | batch_size | Images processed together per pass | | offload_model | Offload the model to CPU after use | | disable_cache | Skip the result cache and always reprocess |

Outputs: IMAGE, width, height.

Install

Drop this folder into ComfyUI/custom_nodes/ and restart ComfyUI.