Recently, the AI field has received some groundbreaking news: the release of HiDream-I1, an open-source image generation model with 1.7 billion parameters. This model not only excels in generating high-quality images but even surpasses the highly acclaimed GPT-4o in certain aspects. HiDream-I1's release allows ordinary users to experience advanced image generation and editing technology without needing a subscription.
Developed by the Chinese company Zhixiang Future, HiDream-I1 has successfully entered the leaderboard of the Artificial Analysis benchmark platform, quickly becoming a new benchmark for open-source models. Within 24 hours of its launch, it topped the charts, becoming the first domestically developed generative AI model from China to achieve this feat. Comparative tests against GPT-4o and Flux1.1 show HiDream-I1 performing well in terms of realism, detail, and understanding user instructions.
For example, with a prompt like "a realistic photo of a horse galloping across the sea from right to left," HiDream-I1 accurately renders the splashing water, the vastness of the ocean, and the dynamic movement of the horse. Furthermore, HiDream-I1 excels in color gradation and detail processing, generating rich and vivid images that provide users with a superior visual experience.
Even more exciting is Zhixiang Future's plan to release another model called HiDream-E1 in the near future. This model will support interactive image editing, allowing users to freely modify the style and content of images, further enhancing the flexibility and practicality of AI image generation.
The release of HiDream-I1 not only provides users with a convenient image generation solution but also propels the development of open-source AI technology, offering a solid foundation for developers looking to enter this field.
HiDream-I1 Model: https://huggingface.co/HiDream-ai/HiDream-I1-Full
HiDream-I1 Code: https://github.com/HiDream-ai/HiDream-I1
