Hotshot-XL

What Hotshot-XL is

Hotshot-XL is a free, open-source AI model for turning text prompts into GIFs alongside Stable Diffusion XL. It can use existing or fine-tuned SDXL models and user-provided SDXL LoRAs for personalized subjects. SDXL ControlNet can guide composition and layout, with one ControlNet model supported at a time. The model is trained for one-second GIFs at 8 frames per second and supports trained aspect ratios from 320×768 through 768×320. Changing duration or frame rate is experimental, and the project warns that outputs may become unstable or jittery because training used one-second videos at 8 FPS. Local-use instructions cover dependency installation, model-weight downloads and inference. A repository script supports fine-tuning with additional text and video pairs; the project recommends SDXL LoRAs rather than fine-tuning Hotshot-XL for personalized subjects. Validation GIFs can be sent to Weights & Biases as an optional workflow. The model card notes limits with photorealism, legible text, compositional prompts, and rendering faces and people. The model has 3B parameters. Natural Synthetics Inc. is identified as developer. GitHub lists Apache-2.0, while Hugging Face identifies Open RAIL++-M.