ComfyUI Model & Task Routes

How ComfyUI Model & Task Routes works on Windows

Ready-to-follow node recipes pairing top diffusion models with specific generation tasks and Windows GPU memory tiers. Exact loader, sampler, and resolution settings for every combination of model architecture and workflow goal.

Every page in this directory focuses on practical local execution on Windows 10 and Windows 11, keeping your checkpoints, LoRAs, and generated outputs entirely on your own SSD.

Start by grabbing the current Windows build from /download, or browse the guides and node blueprints linked below.

Key workflow principles

• Keep large `.safetensors` and `.gguf` weights inside organized `models/` subfolders or map shared directories through extra_model_paths.yaml.
• Match your model precision (FP16, FP8 e4m3fn, or GGUF Q4/Q8) to your card's VRAM tier so sampling stays in GPU memory without paging to disk.
• Drag any generated PNG directly onto the ComfyUI Studio canvas to reload its exact node wiring, seed, and sampler values.

Featured pages in ComfyUI Model & Task Routes

• Flux.1: Text-to-Image Graph in ComfyUI — Blueprint for clean base generation from positive and negative prompts with Flux.1 (8 GB–16 GB VRAM) using EmptyLatentImage → KSampler → VAE Decode → Save Image.
• Flux.1: Image-to-Image Refinement in ComfyUI — Blueprint for restyling an input image at 0.45–0.65 denoise with Flux.1 (8 GB–16 GB VRAM) using Load Image → VAE Encode → KSampler (denoise 0.55) → VAE Decode.
• Flux.1: Inpainting & Mask Editor in ComfyUI — Blueprint for replacing or repairing masked regions seamlessly with Flux.1 (8 GB–16 GB VRAM) using Load Image (MaskEditor) → SetLatentNoiseMask / InpaintModelConditioning → KSampler.
• Flux.1: Multi-LoRA Character & Style Stack in ComfyUI — Blueprint for combining character and lighting LoRAs without overcooking with Flux.1 (8 GB–16 GB VRAM) using Loader → Load LoRA #1 (0.75) → Load LoRA #2 (0.50) → KSampler.
• Flux.1: ControlNet Depth & Pose Lock in ComfyUI — Blueprint for locking camera composition and human pose from a reference frame with Flux.1 (8 GB–16 GB VRAM) using DepthAnything / DWPose Preprocessor → Apply ControlNet (Advanced) → KSampler.
• Flux.1: 8 GB Low-VRAM Optimization in ComfyUI — Blueprint for running heavy weights on 8 GB RTX 3060 / 4060 cards without CUDA OOM with Flux.1 (8 GB–16 GB VRAM) using FP8 / GGUF Q4 Loader → KSampler → VAE Decode (Tiled).
• Flux.1: 4K Tile Upscale Pipeline in ComfyUI — Blueprint for turning a 1024px render into a sharp 4096px master with Flux.1 (8 GB–16 GB VRAM) using Load Upscale Model (4x-UltraSharp) → Ultimate SD Upscale (denoise 0.28).
• Flux.1: Automated Seed Batch Queue in ComfyUI — Blueprint for queuing 20–100 prompt variations overnight on Windows with Flux.1 (8 GB–16 GB VRAM) using Primitive Seed (increment) → Instant Queue / Batch Count → Save Image with date prefix.
• SDXL 1.0: Text-to-Image Graph in ComfyUI — Blueprint for clean base generation from positive and negative prompts with SDXL 1.0 (8 GB+ VRAM) using EmptyLatentImage → KSampler → VAE Decode → Save Image.
• SDXL 1.0: Image-to-Image Refinement in ComfyUI — Blueprint for restyling an input image at 0.45–0.65 denoise with SDXL 1.0 (8 GB+ VRAM) using Load Image → VAE Encode → KSampler (denoise 0.55) → VAE Decode.
• SDXL 1.0: Inpainting & Mask Editor in ComfyUI — Blueprint for replacing or repairing masked regions seamlessly with SDXL 1.0 (8 GB+ VRAM) using Load Image (MaskEditor) → SetLatentNoiseMask / InpaintModelConditioning → KSampler.
• SDXL 1.0: Multi-LoRA Character & Style Stack in ComfyUI — Blueprint for combining character and lighting LoRAs without overcooking with SDXL 1.0 (8 GB+ VRAM) using Loader → Load LoRA #1 (0.75) → Load LoRA #2 (0.50) → KSampler.
• SDXL 1.0: ControlNet Depth & Pose Lock in ComfyUI — Blueprint for locking camera composition and human pose from a reference frame with SDXL 1.0 (8 GB+ VRAM) using DepthAnything / DWPose Preprocessor → Apply ControlNet (Advanced) → KSampler.
• SDXL 1.0: 8 GB Low-VRAM Optimization in ComfyUI — Blueprint for running heavy weights on 8 GB RTX 3060 / 4060 cards without CUDA OOM with SDXL 1.0 (8 GB+ VRAM) using FP8 / GGUF Q4 Loader → KSampler → VAE Decode (Tiled).