The Strongest Open-Source Image-to-Image Model in the World! Tencent Hunyuan Image 3.0 Officially Open Sourced, 80 Billion Parameters Enhance AI Creation

Following the successful launch of the "Yuanbao Spring Festival 1 Billion Red Packet" event, the Tencent Hunyuan team officially announced on January 28th the open-sourcing of its latest Hunyuan Image 3.0 image-to-image model (HunyuanImage3.0-Instruct). This model, featuring a mixture of experts (MoE) architecture with a total parameter size of 80 billion (about 13 billion activated parameters), ranked seventh in the global image editing ranking on LMArena, placing it among the top tier and becoming the strongest open-source image-to-image model globally.

Core technological breakthroughs: "Think first, then edit." Hunyuan Image 3.0 adopts a native multimodal architecture, deeply integrating text and visual understanding. The model learns to analyze image content and user intent before performing tasks by using data at the scale of tens of millions and injecting chain-of-thought (CoT), thus generating accurate editing instructions. Combined with the self-developed MixGRPO algorithm, the model performs exceptionally well in following instructions and maintaining consistency in non-edited areas.

Comprehensive editing experience and multi-scenario empowerment. This model supports various features such as adding, deleting, modifying, style transformation, modification of characters and text, restoration of old photos, and fusion of multiple images. For ordinary users, it is a powerful tool for creating memes and sharing on social media; for professional designers, it significantly improves the efficiency of e-commerce poster creation and character customization for games.

Full commitment to the open-source community. Currently, Tencent Hunyuan has hosted the model's weights and complete code on GitHub and Hugging Face. In addition to the standard version, it also released a distilled version of the model to lower the entry barrier for developers. Developers and researchers can now directly access these resources and conduct application development and secondary development based on this powerful foundation platform.

Official website experience: https://hunyuan.tencent.com/chat/HunyuanDefault?from=modelSquare&modelId=Hunyuan-Image-3.0-Instruct

Tencent Hunyuan releases version 1.5 of its open-source translation model: Performance on edge devices improves significantly, with results comparable to large closed-source models

Tencent Hunyuan open-source translation model version 1.5 introduces 1.8B and 7B models, focusing on efficiency and high-quality translation with optimized cloud-device synergy. The 1.8B model is designed for mobile devices, requiring only 1GB memory for offline operation, enabling on-device deployment and excellent performance.....

Tencent Hunyuan New Technology Makes Large Models 'Less Oily' to Make AI-Generated Images More Realistic!

Recently, the Tencent Hunyuan team released their latest research findings on their official WeChat account —— SRPO (Semantic Relative Preference Optimization), aimed at improving the realism of AI-generated images, especially addressing the 'oily' issue in the skin texture of the open-source text-to-image model Flux. This innovative technology is expected to bring about revolutionary changes in the image generation field. In today's era where digital art is becoming increasingly popular, the quality of AI-generated images has become particularly important. The Flux model, as a popular foundation model in the open-source text-to-image community, is often criticized for its

Tencent HunyuanImage 2.1 Makes a Strong Debut! Open-Source 2K Text-to-Image Champion, Turns into High-Resolution Art Master in an Instant?

Recently, the Tencent Hunyuan team officially open-sourced HunyuanImage2.1, this 17B parameter DiT (Diffusion Transformer) text-to-image model quickly topped the Artificial Analysis Image Arena ranking, surpassing HiDream-I1-Dev and Qwen-Image, becoming the new leader in open-source weight models. The model supports native 2048x2048 resolution output, and significantly improves

Tencent Hunyuan Releases New Theory on Floating Point Quantization Training, Revealing the Limits of Large Model Training

In today's rapid development of Large Language Models (LLM), the training and inference costs of models have increasingly become a focus of research and application. Recently, the Tencent Hunyuan team released an important study that delves into the 'Scaling Laws' of low-bit floating point quantization training, which refers to the principles governing the scale of floating point quantization training. The core of this research lies in exploring how to significantly reduce computational and storage costs without sacrificing performance by lowering the precision of the model.

The Strongest Open-Source Image-to-Image Model in the World! Tencent Hunyuan Image 3.0 Officially Open Sourced, 80 Billion Parameters Enhance AI Creation

Related Recommendations

Tencent Hunyuan releases version 1.5 of its open-source translation model: Performance on edge devices improves significantly, with results comparable to large closed-source models

Tencent Hunyuan New Technology Makes Large Models 'Less Oily' to Make AI-Generated Images More Realistic!

Tencent HunyuanImage 2.1 Makes a Strong Debut! Open-Source 2K Text-to-Image Champion, Turns into High-Resolution Art Master in an Instant?

Tencent Hunyuan Open-Sources End-to-End Video Sound Effect Generation Model HunyuanVideo-Foley

Tencent Hunyuan Releases New Theory on Floating Point Quantization Training, Revealing the Limits of Large Model Training