Tencentorigin of the universeVincent Figure Large Model(hybrid DiT model) announced a comprehensiveOpen SourceTraining code, including LoRA plug-in and ControlNet plug-in.
LoRA is a technique for fine-tuning large language models to train models with specific features using a small amount of data without modifying the original model or increasing its size.
ControlNet is a controlled chemical generation algorithm that allows users to better control image generation by adding additional conditions. Tencent Hybrid provides three premiere ControlNet models that can extract and apply image conditions such as edges, depth, and human pose.
In addition, DiT has released a proprietary acceleration library to improve reasoning efficiency and simplify usage. DiT has been widely used in various fields such as material creation, product synthesis, game graphics, etc., including Tencent's AdMyth platform and a number of media outlets using DiT models for content generation.
Official website.
https://dit.hunyuan.tencent.com/
Code:
https://github.com/Tencent/HunyuanDiT
Model:
https://huggingface.co/Tencent-Hunyuan/HunyuanDiT
Dissertation.
https://tencent.github.io/HunyuanDiT/asset/Hunyuan_DiT_Tech_Report_05140553.pdf
Data production process.
https://github.com/Tencent/HunyuanDiT/blob/main/IndexKits/docs/MakeDataset.md