LongCat-Image is a text-to-image model with about 6 billion parameters. It's open-source and made by the team behind Meituan, a Chinese tech company, as part of their LongCat models series. It was dropped in December 2025.
The LongCat-Image's built to be compact but still punchy. They say it beats many older, bigger models in things like showing text clearly in Chinese and English, creating realistic images, and running smoothly on lower-end setups.
LongCat-Image focuses on keeping things small but still works well. It can handle realistic pictures, show clear multi-language text, and doesn’t need top-tier hardware.
It makes pictures from text, can render text clearly in English and Chinese.
With 6B parameters it’s easier to run than the 70B+ giants. It gives high-quality images, either from scratch or edited.
Keep in mind, it’s not perfect. At just 6B parameters, there’ll be limits. Super complex prompts or rare stuff might not come out perfect. It aims for a balance, not for matching the top-tier art styles.
LongCat-Image could be good for making realistic pictures from Chinese or English prompts or using in places where you can’t run huge models. It’s also handy if you want to build on something open and light without starting from scratch. You could use it in apps, edge devices, or other low-resource setups.
Two more versions exist: Edit and Dev. Edit is meant for image editing, and Dev is a mid-training snapshot for developers who would like to finetune the model to their own needs.
If you'd like to access this model, you can explore the following possibilities:
Weights