Tencent Hunyuan Image 3.5 is here: more accurate text drawing, stronger image editing, a 2K image costs just 1.5 cents

On September 22, Tencent Huan Yuan officially released the preview version of Image 3.5 (Hy Image3.5preview). The new version has comprehensively upgraded core capabilities such as image quality, layout, and photo editing, while significantly reducing costs, with the main slogan "cheap and easy to use."

According to the official introduction, Hy Image3.5preview has improved by about 30% compared to the previous generation. In terms of text rendering, the model can now more accurately reproduce text content, avoiding awkward situations like "missing limbs." Layout capabilities have also been significantly enhanced, allowing the model to better understand users' requirements for visual arrangements. In photo editing, the model supports inputting up to five reference images at the same time, enabling users to combine and adjust styles or elements from multiple images, with a maximum output resolution of 2K.

In terms of pricing, Hy Image3.5preview offers a highly competitive price—generating a 2K image via the Tencent Cloud API costs only 0.15 yuan, and it is charged only for the final product, with no charge for failed images. In internal blind tests, the model tied with Seedream5.0pro, and performed better than two other similar competitors.

In application scenarios, Hy Image3.5preview covers various needs such as daily photo editing, poster design, and storyboard drawing. The model is currently available on products like Yuanbao and ima within the Tencent ecosystem, and developers can also directly call it through the Tencent Cloud API.

Notably, during the gray-scale testing period, the demand for drawing increased by more than 50%. A film studio reported that using Huan Yuan Image 3.5 for batch output significantly shortened the pre-production time, leading to a noticeable improvement in overall efficiency.

Tencent Huan Yuan stated that it will continue to optimize image generation capabilities, further improving the model's performance in complex scenarios while maintaining high cost-effectiveness.

Hy Blog:

https://hy.tencent.com/research/hy-image35-preview

Model Experience:

https://aistudio.tencent.com/

Official Prompt Manual:

https://page.weixin.qq.com/smartpage/p/b1_ALAAAwZ2ACcbt4quDAMT0i1obS0

Tencent Hunyuan Large Model Upgraded: 770B Flagship New Version Hy4preview Shockingly Open-Source, Directly Enters the First Tier of Open-Source Models

Tencent Hunyuan released the preview version of its flagship model, Hy4preview, with a total parameter count of 770B and an activated parameter count of 49B. It supports a context length of 1M. This model performs exceptionally well in real productivity tasks such as coding, office work, and science, ranking among the first-tier open-source models. Its comprehensive performance in internal blind tests exceeds that of similar competitors, and it has already been applied in multiple fields within Tencent.

Tencent Hunyuan launches open-source flagship model Hy4preview with 770B total parameters and 1M context

Tencent Hunyuan released the open-source flagship large model Hy4preview on August 28th, with a total of 770B parameters and 49B activated parameters, and a context length of 1M, positioning it as real productivity. The model has been open-sourced on HuggingFace, GitHub, ModelScope, and is also available on Tencent Cloud TokenHub and OpenRouter. It can be experienced through WorkBuddy/CodeBuddy, Yaobao, and IMA. It was built using expert data from fields such as software engineering, games, finance, and security.

Compressed from 3.3GB to 440MB! How does Tencent Hunyuan's ultra-low bit translation model achieve near-zero loss?

Tencent Hunyuan compressed a 1.8B edge-side translation model to hundreds of megabytes, achieving near-zero loss in translation quality, and has been applied to real-time translation of B站 live chat messages, supporting high-concurrency real-world applications. The model supports mutual translation among 33 languages, and its translation quality leads among models of the same size.

Tencent Hunyuan Launches Hy ASR 3.0 Preview: Speech Recognition No Longer Just Converts Words, but Truly Understands the Context

Tencent Hunyuan launches the speech recognition model Hy ASR 3.0 preview, which leverages the language comprehension capabilities of the large language model Hy3, combining high-precision recognition with deep semantic understanding. It has been significantly improved in general recognition, context awareness, multi-scenario robustness, and dialect coverage. It can provide more accurate and coherent transcriptions by combining context, achieving an evolution from word-by-word transcription to semantic understanding.

Google Integrates Nano Banana2 into Google Earth to Support Online AI-Generated Geospatial Images

Google has integrated the AI image generation model Nano Banana2 into the web version of Google Earth, making it available to users worldwide. This feature combines satellite imagery, aerial photos, and 3D topographic data, allowing users to input a scene description at any location to generate high-precision visualized images.