Tencent's Hunyuan team recently launched the Hy Image3.5 preview model, aimed at professional-grade image generation and editing, with support for text-to-image, image-to-image, and multi-turn conversational creation.

Tencent said Hy Image3.5 preview has already been rolled out for early gray-scale testing and internal trials on the Yuanbao app. Following the upgrade, overall image-generation requests on Yuanbao rose by more than 50%.

Blind Tests by Professional Designers Show Gains

According to Tencent, Hy Image3.5 preview went through GSB blind testing involving several hundred professional designers.

In head-to-head tests, the model's overall performance improved 30% over the previous Hy Image3.0, matched Seedream 5.0 Pro, and came in slightly ahead of Nano-Banana Pro and Qwen-Image-3.0 Pro.

GSB blind testing is a paired-comparison evaluation method. Evaluators compare two sets of generated results and judge which is better without knowing which model produced them, placing more weight on the actual creative experience.

These results come from Tencent's internal testing, and actual performance may vary depending on factors such as prompts, reference images, generation tasks, and evaluation criteria.

Support for Text-to-Image, Image-to-Image, and Multi-Turn Creation

Hy Image3.5 preview supports both text-to-image and image-to-image. Users can generate images directly from text descriptions, or upload reference images to adjust characters, products, scenes, or visual style.

The model supports uploading up to five reference images at once and can combine multiple assets in a single creation. For example, users can upload images of a person, clothing, a product, and a scene together and ask the model to generate visuals in a unified style.

The model also supports multi-turn conversational creation. Users can build on the previous round's output to request further changes, continuously adjusting characters, composition, color, background, and details without having to restate all their requirements each time.

On sizing, Hy Image3.5 preview supports a range of aspect ratios and resolutions up to 2K, suitable for social media images, posters, e-commerce assets, and professional visual design.

Covering Everyday Editing and Professional Creation

Tencent said Hy Image3.5 preview can be used for both everyday image editing and commercial content production.

For personal use, the model supports features such as portrait editing, smart photo retouching, travel planning graphics, and knowledge infographics.

For professional use, the model can be applied to commercial posters, episode storyboards, film and TV materials, UI design, creative illustration, brand visuals, and e-commerce assets.

Through multi-turn dialogue, creators can first generate a base image, then progressively adjust character poses, clothing, camera angles, background, and text layout, cutting down the time spent repeatedly reworking prompts and regenerating.

API Priced Per Output Image

On the commercial side, Tencent Cloud's TokenHub already offers API access to Hy Image3.5 preview.

According to IT Home, the domestic Tencent Cloud price is 0.15 yuan per 2K image, while the international version is $0.024 per image. Billing applies only to output images, with no extra charge based on input text or the number of reference images.

Enterprises and developers can access the model through Tencent Cloud's TokenHub and integrate it into workflows such as content production, e-commerce design, marketing materials, and visual office work.

Specific pricing and service scope may change according to Tencent Cloud platform policies, and the console display should be treated as authoritative at the time of actual onboarding.

Already Integrated Into Multiple Tencent and Partner Apps

Hy Image3.5 preview is now live in apps including Yuanbao, WorkRally, OnSolo, Miora, WorkBuddy, and ima, and has been integrated into Tencent Cloud's Media Processing Service (MPS) and Video on Demand (VOD) platforms.

Yuanbao offers experiences such as portrait editing, smart photo retouching, travel planning graphics, and knowledge infographics, which users can access through its "Creation" feature.

ima, through its Copilot feature, can turn knowledge-base content into visual web pages, slide decks, and images, helping users convert text materials into more intuitive content.

Miora, WorkRally, and OnSolo focus mainly on professional aesthetic creation and film and TV material production, and Tencent is offering a two-week free trial for the related features.

Hunyuan Keeps Up a Rapid Iteration Pace

Tencent Hunyuan said that since rebuilding its infrastructure in February 2026, the Hunyuan large model has maintained a pace of shipping a major version roughly every two months.

For product releases, Tencent uses a "preview first, official version follows" approach, putting the model into real products and user scenarios first, improving its capabilities based on actual usage feedback, and then moving forward with subsequent official upgrades.

This release of Hy Image3.5 preview focuses on improving image quality, multi-image referencing, multi-turn editing, and professional creation capabilities. It also reflects how competition among image models is shifting from single-shot output quality toward sustained creation, contextual understanding, and commercial efficiency.

As the model is integrated into Yuanbao, office tools, media processing platforms, and professional creation apps, Tencent is further expanding the use cases for its Hunyuan image models and pushing image generation from a personal entertainment tool toward enterprise-grade content production infrastructure.