Tencent’s Hunyuan team has unveiled the Hy Image3.5 preview model, marking a significant advancement in AI-powered image creation. This new model is engineered for professional-grade image generation, boasting capabilities in text-to-image, image-to-image, and multi-turn conversational creation.
Professional-Grade Enhancement
According to Tencent, Hy Image3.5 preview underwent rigorous internal blind testing involving hundreds of professional designers. The results indicate a 30% improvement over its predecessor, Hy Image3.0, performing on par with Seedream 5.0 pro and demonstrating a slight edge over Nano-Banana Pro and Qwen-Image-3.0 Pro.
The GSB blind testing methodology, used for evaluation, ensures that testers judge image quality without knowing the source model, providing an objective assessment of performance.
Real-World Application and Adoption
Hy Image3.5 preview has already seen progressive, phased gray-scale release and internal testing within the Yuanbao App for over two months, reportedly leading to a more than 50% increase in overall image generation requests.
The model is now integrated into a suite of applications, including Yuanbao, WorkRally, OnSolo, Miora, WorkBuddy, and ima. It is also accessible via Tencent Cloud’s Media Processing (MPS) and Video on Demand (VOD) platforms.
For users, Yuanbao offers free access to features like portrait editing, intelligent photo retouching, travel planning visualizations, and knowledge infographics within its ‘Creation’ section. The ima application leverages copilot functionality to transform knowledge bases into visual web pages, PPTs, and images. Miora, WorkRally, and OnSolo are targeting professional aesthetic and film material creation, with a limited-time two-week free trial available.
Advanced Capabilities and Features
Hy Image3.5 preview supports both text-to-image and image-to-image generation. Users can provide up to five reference images simultaneously for image-to-image tasks. The model excels in multi-turn conversational creation, allowing for continuous refinement and editing based on historical context. It also offers flexibility in output dimensions, supporting various aspect ratios and generating images up to 2K resolution.
Tencent highlights the model’s versatility, stating it can cater to everyday needs such as photo editing, portrait beautification, and knowledge card creation. Simultaneously, it is equipped to handle demanding professional applications, including commercial posters, TV series storyboards, film production materials, UI design, creative illustrations, brand visuals, and e-commerce assets.
Iterative Development and Future Outlook
Tencent Hunyuan emphasizes its commitment to rapid development. Since rebuilding its infrastructure in February, the Hunyuan large model has seen major version updates approximately every two months. The strategy of releasing ‘preview’ versions ahead of formal launches allows for the integration of real-world user feedback directly into the research and development cycle, ensuring continuous improvement and adaptation to user needs.
Pricing and Access
For developers and enterprises, access to the Hy Image3.5 preview model is available through Tencent Cloud’s TokenHub API. The pricing for a 2K image is set at 0.15 Yuan per image via Tencent Cloud API, and 0.024 USD per image on the international version. Charging is solely based on the output images produced.
References:
- Hunyuan Official Website: aistudio.tencent.com
- Tencent Cloud TokenHub: console.cloud.tencent.com
