Tencent's Hunyuan DiT large model (HunyuanDiT) continues to be iteratively updated, bringing users more powerful image generation capabilities. The editor of Downcodes learned that HunyuanDiT and the community recently released three new controllable plug-ins, ControlNet, namely tile (high-definition amplification), inpainting (image repair and expansion) and lineart (line drawing), which significantly enhances the model's performance. Range of applications and creative freedom. The addition of these plug-ins allows Hunyuan DiT to show stronger application potential in the fields of art, creativity, architecture and other fields, providing more accurate and convenient image generation services to developers and creators around the world.
Tencent's HunyuanDiT large model (HunyuanDiT) recently teamed up with the community to release three new controllable plug-ins, ControlNet, namely tile (high-definition amplification), inpainting (image repair and expansion) and lineart (line drawing), to further expand Its ControlNet matrix. The addition of these plug-ins enables the Hunyuan DiT model to cover a wider range of application scenarios, including 80% of cases and scenarios such as art, creativity, architecture, photography, beauty and e-commerce, providing global enterprises and individual developers and creators with Provides more accurate image generation and greater creative freedom.
The Tile plug-in can expand information for the picture and achieve ultra-clear amplification, even reaching 4K to 8K resolution, which is suitable for scenes that require the ultimate pursuit of picture details. The Inpainting plug-in can fill in the smeared and mottled parts of the picture according to the needs of the creator, achieve effects such as background replacement and character subject change, and handle large-area image redrawing. The Lineart plug-in uses different line types to create real-life, animation and architectural pictures, and is suitable for generating architectural renderings and coloring manuscripts.
In addition, Tencent Hunyuan DiT has previously released ControlNet models with canny (edge), depth (depth), pose (human posture) and other conditions to support developers in reasoning, and has open sourced the ControlNet training program to enable developers and creators to Ability to train custom ControlNet models.
Since announcing a comprehensive upgrade and open source in May, Hunyuan DiT, as the industry's first Chinese-native DiT architecture open source graph generation model, has continued to build a developer ecosystem and released an exclusive acceleration library to improve reasoning efficiency and shorten graph generation time. And further open sourced the inference code. In July, Hunyuan DiT was upgraded to version 1.2, and a small video memory version was open sourced. It only requires 6G of video memory to run, making it more friendly to developers deployed locally on personal computers.
Currently, Hunyuan DiT has more than 3.1k stars on Github, making it the most popular domestic DiT open source model.
Official website
https://dit.hunyuan.tencent.com/
code
https://github.com/Tencent/HunyuanDiT
Model
https://huggingface.co/Tencent-Hunyuan/HunyuanDiT
paper
https://tencent.github.io/HunyuanDiT/asset/Hunyuan_DiT_Tech_Report_05140553.pdf
All in all, Tencent Hunyuan DiT's continuous updates and open source strategy provide developers and creators with powerful tools and resources, and promote the progress and development of Wenshengtu technology. It is worth looking forward to more innovations and breakthroughs in the future.