This is a ComfyUI wrapper for Andrew DalPino's SuperCool upscaler, enabling its use directly within ComfyUI's node workflow. No extra dependencies required—just drop in the models and go.
The original SuperCool repo (https://github.com/andrewdalpino/SuperCool) provides a fast and high-quality image upscaler. This wrapper integrates that functionality into ComfyUI using its native Python environment—no external pip installs needed.
All upscaling models are fixed at 4x internally—despite the file naming for compatibility with ComfyUI conventions (see below).
- Memory Management: Automatic model unloading and VRAM cleanup
- Internal Batching: Process large image sequences without external batching nodes
- Mixed Precision: Reduced VRAM usage with torch.autocast
- Better Error Handling: Comprehensive error reporting and recovery
- Progress Feedback: Visual progress for batch processing
unload_after_run: Automatically unloads model after processing (default: True)force_gc: Forces garbage collection for thorough memory cleanup (default: True)batch_size: Process images in chunks to prevent OOM errors (default: 10)
No more complex batching setups! Process hundreds of video frames directly:
- Load your video frames
- Set appropriate batch size for your VRAM
- The node handles everything automatically
Credit: Memory optimisations contributed by Maelstrom2014 - cheers for making this production-ready!
git clone https://github.com/DragonDiffusionbyBoyo/BoyoSupercoolWrapper
So it ends up like:
ComfyUI/custom_nodes/BoyoSupercoolWrapper/
| Model Size | Download Link | Rename To |
|---|---|---|
| Small (fast) | https://huggingface.co/andrewdalpino/SuperCool-4x-Small | supercool_2x |
| Medium | https://huggingface.co/andrewdalpino/SuperCool-4x-Medium | supercool_4x |
| Large (slow, high qual) | https://huggingface.co/andrewdalpino/SuperCool-4x-Large | supercool_8x |
Note: All models upscale by 4x. The naming (2x, 4x, 8x) is just used for node dropdown compatibility—they do not reflect actual scale factors.
ComfyUI/models/upscale_models/SuperCool/
If the SuperCool folder does not exist, create it manually.
Once everything is in place, launch ComfyUI and look for the SuperCool Upscaler (Memory Optimised + Batching) node under the upscale category. You can select from three quality/speed presets via dropdown.
- Batch Size: Start with 10, adjust based on your VRAM:
- RTX 3060 (8GB): Try 5-8
- RTX 4070/4080 (12-16GB): Try 10-15
- RTX 4090 (24GB): Try 20-30+
- Unload After Run: Keep enabled to free VRAM for other nodes
- Force GC: Keep enabled for thorough memory cleanup
- Interpolation First: Use RIFE or similar interpolation nodes before upscaling
- Batch Appropriately: Lower batch sizes if you're running other memory-intensive nodes
- Monitor Progress: Watch the console for batch processing updates
This wrapper uses only ComfyUI's standard packages. No additional installations are required.
Original SuperCool: Andrew DalPino
Memory Optimisations: Maelstrom2014 - massive thanks for the production-ready improvements!
Wrapper & Integration: Dragon Diffusion UK
Maintained by Dragon Diffusion UK: https://www.dragondiffusionuk.co.uk We build practical, bleeding-edge AI integrations for actual humans.
Pull requests, feedback, and banter welcome. PS please note I am a vibe coder so if you ask me something that is not on sport I will struggle haahahahah