DeepSeek, a Chinese AI startup, has recently launched Janus-Pro, a multimodal text-to-image AI model, after its prior success with R1. Janus-Pro, available as open source, claims to outperform popular image generators like Stable Diffusion and Dall-E in benchmark tests. With various model sizes, Janus-Pro is noted for its autoregressive framework and reportedly lower training costs, as well as energy efficiency compared to US competitors. Initial reactions are promising, but user testing will ultimately determine its standing against other models.
DeepSeek's new Janus-Pro model, after outpacing ChatGPT in downloads, highlights a significant leap in multimodal AI capabilities and competitive performance against established image generators.
The Janus-Pro model showcases an autoregressive framework that DeepSeek claims 'surpasses' unified models, addressing various aspects of image creation and analysis.
DeepSeek's Janus-Pro, released after the successful R1 model, is garnering initial positive impressions and outperforms existing generators on key benchmarks.
Nvidia praises Janus-Pro as an 'excellent AI advancement,' signaling a shift in the competitive landscape among AI startups.
Collection
[
|
...
]