Juggernaut XL

Version 9 of this SDXL fine-tune, built by KandooAI with RunDiffusion and folding in the RunDiffusion Photo v2 training, targets realistic photography: skin texture, natural light and contrast, and subjects like portraits, architecture, cars, food, interiors and wildlife. It is a checkpoint, not an application. The Hugging Face repo provides a single safetensors file for SDXL front ends such as ComfyUI, InvokeAI, Forge or SwarmUI, plus a diffusers-format fp16 tree for loading with DiffusionPipeline in Python. The VAE is baked in. Suggested settings are about 832 by 1216 pixels, DPM++ 2M Karras, 30 to 40 steps and a CFG of 3 to 7, with light or empty negative prompts. Since it stays on the SDXL architecture, existing SDXL LoRAs, ControlNets and IP-Adapters generally work, and the card says 8 GB of VRAM is enough. Licensing is CreativeML Open RAIL-M with an added condition: personal and creative use is fine, but serving it behind a paid API needs a separate deal with RunDiffusion. The card also pushes RunDiffusion's hosted service hard. A dependable pick for local photoreal SDXL work.