Tencent officially released its latest Hunyuan Image 2.0 model (Hunyuan Image2.0), marking the entry of AI image generation technology into the "millisecond response" era.
The new model has made significant improvements in speed. Compared to its predecessor, the parameter scale of Hunyuan Image 2.0 has increased by an order of magnitude. By combining an efficient image codec and a novel diffusion architecture, it can achieve millisecond-level fast responses at speeds where most commercial products typically require 5 to 10 seconds for inference. When users generate images, they can receive real-time image outputs while inputting text or giving voice commands, greatly changing the traditional "draw - wait - redraw" mode and enhancing user interaction experiences.
Super-realistic image quality
Besides the breakthrough in speed, Hunyuan Image 2.0 has also made significant progress in the quality of image generation. The model uses reinforcement learning and introduces a large amount of human aesthetic knowledge to effectively avoid the common "AI flavor" in AI-generated images. The generated images not only have strong realism and rich details but also have high usability. In the authoritative GenEval benchmark test, Hunyuan Image 2.0's accuracy in understanding and generating complex text instructions exceeds 95%, far surpassing other similar models.
Innovative real-time painting board function
This upgrade also introduces a real-time painting board function, leveraging the real-time image generation capability of the new model. When users draw sketches or adjust parameters, the preview area can synchronously generate coloring effects. This feature breaks through the traditional "draw - wait - modify" process, greatly facilitating the creative process for professional designers. Additionally, the real-time painting board supports multi-image fusion. Users can upload multiple sketches, and the AI will automatically coordinate perspective and lighting according to the user's prompts to generate a fused image, further enriching the interactive experience of AI image generation.
Tencent also revealed that the native multimodal image generation large model is currently under development. The new model will perform excellently in areas such as multi-round image generation and real-time interaction experiences, hoping to bring users a richer creation experience.
Product access: https://hunyuan.tencent.com/