Tencent Releases Hy Image 3.5 Preview With Multi-Round Editing and 4K Output

Tencent has released Hy Image 3.5 Preview, an image-generation model that can create or revise pictures over multiple turns and produce output up to 4K. The company is integrating it into its Yuanbao chatbot and film-editing and design tools, putting the release directly into products as well as a developer API.

The timing shows how Tencent is moving its Hunyuan AI effort beyond text-based systems. The company describes the model as an expansion of its AI applications beyond text, while the product integrations give its image work a route to users inside Tencent’s existing software.

From a cloud listing to a product rollout

Tencent Cloud’s documentation, updated September 21, lists the model under the ID hy-image-v3.5-preview. It accepts text and reference images, supports text-to-image and image-to-image generation, and lets an application send earlier conversation history back with a new instruction for multi-turn editing.

That interaction model is a meaningful shift from treating image generation as a one-shot prompt. A user can provide a reference image, generate a result, then request a targeted change while carrying the prior image and conversation context into the next round. The service returns an image synchronously, rather than making the caller submit a job and poll for completion.

What the API exposes

• Text-to-image, image-to-image and multi-turn editing through a chat-style messages format.
• Text plus reference-image input, custom image sizing and optional external-search enhancement.
• Output up to 4096 by 4096 pixels. Developers can explicitly request that 4K size.

The details matter most for teams building an editing flow around the model. Tencent’s API preserves a prepared history that developers can append to the next request, including the prior generated image. That makes sequential changes part of the interface design, rather than an undocumented behavior users must recreate themselves.

Tencent’s quality case remains an internal one

Tencent says it tested Hy Image 3.5 Preview with hundreds of in-house designers. In those tests, it says the model matched ByteDance’s Seedream 5.0 Pro and performed slightly better than Google’s Nano Banana Pro and Alibaba’s Qwen-Image-3.0 Pro.

Those results are Tencent’s own assessment, and the company did not quantify the quality claims. The next test is therefore practical rather than rhetorical: whether the model’s revision workflow holds up across the Yuanbao and creative-tool integrations, and whether outside evaluations reproduce Tencent’s competitive picture.