Tencent HunyuanImage 2.0 model (Hunyuan Image2.0) was officially released on May 16th, on the official website of Tencent Hunyuan, and opened to the public for registration and experience.
The Tencent Hybrid Image 2.0 model has two main features:Real-time graphic generation, ultra-realistic image quality.
Compared to its predecessor, Tencent's Hybrid Image 2.0 model has an order of magnitude higher parameter count, and thanks to its ultra-high compression rate image codec and new diffusion architecture, its image generation speed is "significantly faster than the industry's leading models," with similar commercial products requiring 5 to 10 seconds of inference per image.Tencent hybrid claims to achieve millisecond response, support users can type or talk while out of the pictureIt's a change from the traditional "draw card - wait - draw card" approach.
In addition to being fast.Tencent Hybrid Imaging 2.0 Model Image Generation Quality Improved SignificantlyThe images generated are said to avoid the "AI flavor" of AIGC images through algorithms such as reinforcement learning and the introduction of a large amount of human aesthetic knowledge alignment.
On GenEval (Geneval Bench), an evaluation benchmark in the field of image generation that specializes in testing the ability of models to understand and generate complex textual instructions.Tencent's hybrid image 2.0 model accuracy exceeds 95%, far surpassing other similar models.
This model upgrade also brings the release of theReal-time drawing board functionThe model-based real-time graphic ability allows users to synchronize the coloring effect in the preview area when drawing line drawings or adjusting parameters, which breaks through the traditional linear process of "drawing-waiting-modifying" and helps professional designers to create their own work.
Real-time drawing board support for multi-image fusionAfter uploading multiple images, users can superimpose multiple sketches onto the same canvas to create freely, and after AI automatically coordinates perspective and light and shadow to generate fused images according to the content of the prompts, which further enriches the interactive experience of AI-generated images.
Tencent Hybrid Native Multimodal Image Generation Large Model is on the Way, Tencent said the new model excels in multi-round image generation and real-time interactive experience.