Skip to content

Hy Image 3.5: Tencent releases a model for image generation and editing

2 min read

In short

On September 22, 2026, Tencent Hunyuan released Hy Image 3.5 Preview, a model that both generates images from text and edits them from photos. In Tencent's tests it wins 30% more often than Hy Image 3.0, renders multilingual text more accurately and keeps people and products consistent across edits. The weights are closed; it's available via API.

Contents
Art director's table covered with campaign prints: the same young woman and white sneakers in a studio, on a street and on a beach

On September 22, 2026, the Tencent Hunyuan team released Hy Image 3.5 Preview, the successor to Hy Image 3.0. The headline change: text-to-image generation and image editing now run in a single model. Before the official launch, the model was tested for two months in Tencent’s Yuanbao chatbot.

What Hy Image 3.5 can do

  • Generation and editing. Text-to-image and photo editing go through one model, and edits can continue over several turns, like a conversation.
  • Text in images. Tencent focuses on legible multilingual text and layout: posters, infographics and UI mockups.
  • Consistent people and products. When the background, outfit or style changes, the model keeps a person’s features and a product’s look.
  • Styles: photorealism, cinematic, comic, commercial photography and illustration.
  • Short prompts. The model rewrites and expands the prompt on its own.

How much better than Hy Image 3.0

According to Tencent, in a blind comparison by the company’s designers, Hy Image 3.5 wins 30% more often than 3.0. Tencent also says it’s on par with Seedream 5.0 Pro and slightly ahead of Nano Banana Pro. These are internal company tests: the model wasn’t on any independent leaderboard at launch.

Specs and availability

Hy Image 3.5 Preview
Developer Tencent Hunyuan
Release September 22, 2026
Modes text-to-image, image editing, multi-turn editing
Resolution up to 2K natively, 4K via upscaling
References up to 5 in ComfyUI, up to 20 on OpenRouter
Aspect ratio from 1:4 to 4:1
Weights closed

The model is available through Tencent’s cloud API, in Tencent apps (Yuanbao and others), in ComfyUI partner nodes and, since October 5, on OpenRouter. Hy Image 3.0 shipped with open weights, but 3.5 is so far offered only as a cloud service.

How to try it in AI2Face

Hy Image 3.5 is available in the AI2Face generator next to the models Tencent itself compares it with — Seedream and Nano Banana Pro — plus GPT Image. For ad shoots with one model and one product, see our guide to putting clothes on a model. The cost is shown before each generation.

FAQ

How is Hy Image 3.5 different from Hy Image 3.0?

Generation and editing are merged into one model, with better text in images and better consistency for people and products across edits. In a blind test by Tencent's designers, 3.5 wins 30% more often than 3.0.

Can I download Hy Image 3.5?

No. Unlike the open-weight Hy Image 3.0, version 3.5 Preview is only available through Tencent's cloud API and partner services.

What resolution does Hy Image 3.5 support?

Up to 2K natively. There's a 4K tier, but according to the ComfyUI docs it's upscaled from 2K.

Sources

3
  1. https://x.com/TencentGlobal/status/2102308689378259221
  2. https://docs.comfy.org/tutorials/partner-nodes/tencent/hy-image-3-5
  3. https://openrouter.ai/tencent/hy-image-v3.5-preview

More news