ComfyUI Models & Checkpoints

How ComfyUI Models & Checkpoints works on Windows

Where to place safetensors, GGUF, UNet, LoRA, ControlNet, and VAE weights for SDXL, Flux.1, and Hunyuan Video on Windows. Organizing models/checkpoints, models/unet, and extra_model_paths.yaml so ComfyUI shares weights cleanly across local tools.

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 Models & Checkpoints

• Flux.1 Dev in ComfyUI — 12B rectified-flow transformer delivering top prompt adherence and typography when loaded via DualCLIPLoader (clip_l + t5xxl) and ae.safetensors. Requires 12 GB+ (8 GB with GGUF Q4 / FP8) VRAM and loads from models/unet/ or models/checkpoints/.
• Flux.1 Dev Low-VRAM Settings for ComfyUI — Step-by-step memory tuning for Flux.1 Dev (12 GB+ (8 GB with GGUF Q4 / FP8)) so 8 GB and 12 GB RTX cards render cleanly without CUDA Out of Memory errors.
• Flux.1 Schnell in ComfyUI — Apache-2.0 distilled 4-step transformer built for rapid local prototyping and sub-second iterations on consumer RTX cards. Requires 8 GB–12 GB VRAM and loads from models/unet/ or models/checkpoints/.
• Flux.1 Schnell Low-VRAM Settings for ComfyUI — Step-by-step memory tuning for Flux.1 Schnell (8 GB–12 GB) so 8 GB and 12 GB RTX cards render cleanly without CUDA Out of Memory errors.
• Stable Diffusion XL 1.0 in ComfyUI — Native 1024×1024 dual-encoder architecture with the widest ecosystem of LoRAs, ControlNets, and IPAdapter weights. Requires 8 GB+ VRAM and loads from models/checkpoints/.
• Stable Diffusion XL 1.0 Low-VRAM Settings for ComfyUI — Step-by-step memory tuning for Stable Diffusion XL 1.0 (8 GB+) so 8 GB and 12 GB RTX cards render cleanly without CUDA Out of Memory errors.
• Stable Diffusion 1.5 in ComfyUI — Lightweight 512×512 baseline checkpoint ideal for ultra-fast batch runs, AnimateDiff motion modules, and low-VRAM laptops. Requires 4 GB–6 GB VRAM and loads from models/checkpoints/.
• Stable Diffusion 1.5 Low-VRAM Settings for ComfyUI — Step-by-step memory tuning for Stable Diffusion 1.5 (4 GB–6 GB) so 8 GB and 12 GB RTX cards render cleanly without CUDA Out of Memory errors.
• Stable Diffusion 3.5 Large in ComfyUI — 8B MMDiT model combining triple text encoders (CLIP-L, CLIP-G, T5-XXL) for rich artistic styles and complex multi-subject prompts. Requires 12 GB–16 GB (FP8 / GGUF supported) VRAM and loads from models/checkpoints/.
• Stable Diffusion 3.5 Large Low-VRAM Settings for ComfyUI — Step-by-step memory tuning for Stable Diffusion 3.5 Large (12 GB–16 GB (FP8 / GGUF supported)) so 8 GB and 12 GB RTX cards render cleanly without CUDA Out of Memory errors.
• Hunyuan Video in ComfyUI — 13B open-weights text-to-video and image-to-video foundation model producing coherent 720p motion sequences inside native ComfyUI video nodes. Requires 12 GB–24 GB (with GGUF / FP8 offload) VRAM and loads from models/diffusion_models/.
• Hunyuan Video Low-VRAM Settings for ComfyUI — Step-by-step memory tuning for Hunyuan Video (12 GB–24 GB (with GGUF / FP8 offload)) so 8 GB and 12 GB RTX cards render cleanly without CUDA Out of Memory errors.
• Wan 2.1 Video in ComfyUI — High-temporal-consistency video suite offering both a consumer-friendly 1.3B model for 8 GB GPUs and a 14B flagship for studio shots. Requires 8 GB (1.3B) to 16 GB+ (14B) VRAM and loads from models/diffusion_models/.
• Wan 2.1 Video Low-VRAM Settings for ComfyUI — Step-by-step memory tuning for Wan 2.1 Video (8 GB (1.3B) to 16 GB+ (14B)) so 8 GB and 12 GB RTX cards render cleanly without CUDA Out of Memory errors.