Best ComfyUI Upscaler Workflow | SDXL vs FLUX for Sharp AI Images

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/C... ---------------- Clone into this folder: ComfyUI\custom_nodes Then paste this: git clone https://github.com/GeekatplayStudio/C... cd ComfyUI-UltimateUpsacaler install.bat --comfyui .. If that does not work, use your full ComfyUI path instead: git clone https://github.com/GeekatplayStudio/C... cd ComfyUI-UltimateUpsacaler install.bat --comfyui D:\ComfyUI Full install with FLUX: git clone https://github.com/GeekatplayStudio/C... 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=af... 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 -   / 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