ComfyUI Node
🐇 Upscale w/ Model (Advanced)
Based on Comfy's native "Upscale Image (using Model)", with controls exposed to tune for large batches, avoid slow OOM fallbacks, and create opportunities to optimize for speed. Defaults - Behaves about the same as the original node. Controls - max_batch_size > 0: process images in chunks to keep VRAM steady and reduce fallback slowdowns. - tile_size: choose a starting tile; original node defaults to 512. 0 = auto (falls back 512 → 256 → 128 on OOM). - channels_last: try ON for a speedup on some systems. - precision: lower (fp16/bf16) can be faster; may impact quality depending on the model.
🐇 Upscale w/ Model (Advanced)
- upscale_model
- image
- IMAGE
◄max_batch_size0►
◄tile_size0►
◄channels_lastfalse►
◄precisionfp32►
Categoryimage/upscaling
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| upscale_model | UPSCALE_MODEL | Pick your ESRGAN model (e.g. 2× / 4×). | |
| image | IMAGE | Images to upscale. Accepts a batch: frames×H×W×C with values in [0–1]. | |
| max_batch_sizeopt | INT | 00–4096 | How many images to process at once. 0 = all at once. Set >0 if you hit OOM. |
| tile_sizeopt | INT | 00–2048 | How big each tile is. 0 = auto (starts at 512 and halves on OOM). Bigger is faster; smaller is safer. |
| channels_lastopt | BOOLEAN | false | Try this ON for a small speed boost on some GPUs. If you see no gain, leave it OFF. |
| precisionopt | COMBO | fp32 | Math mode. fp32 = safest. fp16/bf16 can be faster on many GPUs, may impact image quality. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |