Tencent Unveils Hy Image 3.5 for Generation and Editing

Tencent's Hunyuan team has introduced a preview of Hy Image 3.5, a unified model that handles both text-to-image and image-to-image tasks through a single endpoint. By consolidating these capabilities, the model allows developers to streamline their applications, eliminating the need to route generation and editing requests to separate systems. The new model can produce high-quality outputs at resolutions up to 2K, which helps teams maintain visual consistency across large batches of product and campaign assets.

The update represents a significant leap over its predecessor, Hunyuan Image 3.0. Tencent reports that Hy Image 3.5 achieved a 30 percent win-rate improvement over version 3.0 in human evaluations. While the previous generation relied on a separate 3.0-Instruct variant for reference-based editing, the 3.5 version integrates these workflows. The technology builds on the open-source Hunyuan Image 3.0 base, which features an 80-billion-parameter Mixture of Experts architecture containing 64 experts and 13 billion active parameters. This architecture reduces computation by activating only a subset of the network for each request.

For developers integrating the technology, Tencent has priced the hosted model at $0.024 per generated image via the Tencent Cloud API. Notably, the company does not charge for uploaded reference images, which lowers the cost for projects requiring multiple style, layout, or character guides. The preview is currently accessible under the identifier hy-image-v3.5-preview. Additionally, users can test the model for free for two weeks through partner design platforms Miora and OnSolo, though full API-specific testing for latency and concurrency still requires direct cloud access.