ByteDance Open-Sources Lance: A Native Multimodal Model That Runs Locally on 40GB VRAM

ByteDance open-sourced Lance, a compact multimodal AI model for image and video tasks.

The release includes model weights, inference code, and quantization guidance for local deployment.

Lance is designed to run on about 40GB VRAM, with community quantized versions targeting 24GB GPUs.

The launch quickly sparked derivative releases on Hugging Face, signaling strong interest in local multimodal AI.