Local Video Generation Models: Alternatives to Wan2.1 and Wan2.2 on RTX 3090
A Reddit user is seeking recommendations for local video generation and editing models that perform better than the recently released Wan2.1 and Wan2.2 models on a consumer-grade RTX 3090 GPU. As open-source video generation models like Wan2.1 and Wan2.2 advance, users are looking to optimize their workflows on local consumer hardware, highlighting the demand for efficient, high-quality local AI video tools. Wan2.1 and Wan2.2 support text-to-video (T2V) and image-to-video (I2V) generation, with model sizes like 1.3B and 14B that can run on local setups using tools like ComfyUI. Running larger models or newer iterations on an RTX 3090 (24GB VRAM) requires careful optimization of model weights and precision formats like FP8.
## BACKGROUND
Wan2.1 and Wan2.2 are open-source, large-scale video generative models developed by Wan-AI, capable of text-to-video and image-to-video tasks. The RTX 3090 is a popular NVIDIA GPU with 24GB of VRAM, making it a benchmark hardware choice for running demanding local AI models. ComfyUI is a node-based graphical user interface commonly used to run these diffusion models locally.