In this video I show my ComfyUI Ultimate Upscaler workflow and explain how I use it to upscale images with tiled generative detail enhancement. I go through both versions of the workflow: one built around SDXL and one built around FLUX. SDXL is usually my preferred option because it has a smaller footprint, runs faster, and uses much less VRAM, while FLUX can sometimes produce better results but is heavier and slower.

I also explain the core principle behind generative upscaling in ComfyUI: splitting the image into tiles, preparing the correct tile size, adding padding to avoid visible seams, controlling noise for detail, using captioning for tile prompts, and combining everything back into a high-resolution final image.

This workflow includes custom helper nodes, tile planning, preview tools, and options for different magnification values. I also show how padding, target resolution, model selection, and VRAM settings affect the final result.

GitHub repo:
https://github.com/GeekatplayStudio/ComfyUI-UltimateUpsacaler.git

Clone into this folder:
ComfyUI\custom_nodes

Then paste this:

git clone https://github.com/GeekatplayStudio/ComfyUI-UltimateUpsacaler.git

cd ComfyUI-UltimateUpsacaler
install.bat --comfyui ..

If that does not work, use your full ComfyUI path instead:

git clone https://github.com/GeekatplayStudio/ComfyUI-UltimateUpsacaler.git

cd ComfyUI-UltimateUpsacaler
install.bat --comfyui D:\ComfyUI

Full install with FLUX:

git clone https://github.com/GeekatplayStudio/ComfyUI-UltimateUpsacaler.git

cd ComfyUI-UltimateUpsacaler
install.bat --comfyui D:\ComfyUI --with-flux-assets

After install:
Restart ComfyUI
Load this workflow:
workflows\hyper_tile_sdxl.json

Final folder should be:
ComfyUI\custom_nodes\ComfyUI-UltimateUpsacaler

#ComfyUI #SDXL #FLUX #AIUpscaler #ImageUpscaling #StableDiffusion #GenerativeAI #AIArt #ComfyUIWorkflow #Upscaler #Geekatplay #UltimateUpscaler

⚠️ AFFILIATE DISCLOSURE: We may earn a commission from purchases made through the links below at no extra cost to you.
My recommendations:
Hitem3d: https://www.hitem3d.ai/?utm_source=aff&utm_medium=cps&utm_campaign=geekatplay
TripoAI: http://studio.tripo3d.ai?via=geekatplay
Meshy 3d AI: www.meshy.ai?via=geekatplay
i10X. :https://i10x.ai?fpr=vladimir24
Topaz AI Video and Photo processing: https://topazlabs.com/ref/1514/

My Patreon webpage - https://www.patreon.com/geekatplay
Tutorials and packs - https://gumroad.com/geekatplay
Tutorials website - https://www.geekatplay.com
Photography - https://www.chopinephotography.com

00:00 Why AI images still need upscaling
00:08 My upscaling workflow and custom nodes
00:15 How tiled upscaling actually works
00:22 Feel free to use and improve the code
00:30 Two workflows: FLUX and SDXL
00:38 Why SDXL is faster and lighter
00:57 Image preload and tile preview
01:13 Why 512x512 tiles work well for SDXL
01:25 How image tiling works for upscale
01:57 Scaling logic and tile sizing
02:17 Why padding is important
02:34 What happens without padding
02:50 Planner, ratios, and resolution handling
03:19 Tile size, denoise, and automation
03:33 How denoise affects details
03:46 Target width, height, and magnification
04:10 VRAM limits and large upscale warnings
04:38 Florence captions for tile prompts
04:52 Sharpening and preview tools
05:16 Choosing checkpoints and models
05:37 Custom upscaler and sampler settings
06:00 Tile padding, masks, and seam control
06:17 Final preview and saved output
06:35 FLUX workflow overview
06:57 Ongoing node improvements
07:11 Subscribe for new workflows and updates
07:20 Changing magnification to 3x
07:38 Tile captions and prompt compositing
07:52 Generating and assembling the tiles
08:19 Comparing detail to the original
08:33 Why I like generative upscaling
08:40 Final thoughts and resources