//
High-resolution AI image generation
SDXL (Stable Diffusion XL) generates stunning 1024x1024 images natively, a big step up from SD 1.5. It needs more VRAM but produces significantly better results.
ComfyUI is the best interface for SDXL.
git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI
pip install -r requirements.txtGet the base model and optionally the refiner.
# Download from Hugging Face or CivitAI:
# - sd_xl_base_1.0.safetensors (~6.5GB)
# - sd_xl_refiner_1.0.safetensors (~6.5GB, optional)
# Place in: ComfyUI/models/checkpoints/š” The refiner improves details but doubles generation time.
Use the built-in SDXL workflow or download one.
# ComfyUI includes SDXL workflows
# Load from: Load Default > SDXL
# Or download workflows from:
# https://github.com/comfyanonymous/ComfyUI_examplesSet your prompt and generate!
Settings for SDXL:
ā¢ Resolution: 1024x1024 (native)
ā¢ Steps: 25-50
ā¢ CFG: 7-8
ā¢ Sampler: DPM++ 2M Karrasā Out of memory on 8GB GPU
ā Enable VAE tiling and fp16 mode. Use smaller batch sizes. Consider upgrading to 12GB+ card.
ā Images look blurry
ā Use the refiner model for sharpening. Or use a sharpening LoRA. Increase steps to 40+.