repository·main·Indexed 42 days ago
https://github.com/tyxsspa/anytext2A visual text generation and editing framework for precise control over multilingual text attributes, such as font and color, in natural scene images. It utilizes a WriteNet+AttnX architecture to integrate text rendering capabilities into pre-trained Text-to-Image models.
Training is supported using the updated AnyWord-3M dataset, which includes long captions and text color labels. To start training, execute train.py. Detailed training configurations and parameters are consistent with the AnyText documentation.
python train.pyTo install AnyText2, you need to download the model checkpoints via ModelScope, clone the repository, and set up a Conda environment.
modelscope to download the iic/cv_anytext2 model.AnyText2 repository and move the downloaded checkpoints into a models directory within the project root.environment.yaml file.# 1. Download checkpoint (requires pip install modelscope first)
python -c "from modelscope import snapshot_download; print(snapshot_download('iic/cv_anytext2'))"
# 2. Clone anytext code and move checkpoint
git clone https://github.com/tyxsspa/AnyText2.git
cd AnyText2 && mkdir -p models && mv /path/to/your_downloaded_dir/* models
# 3. Create a new environment and install packages
conda env create -f environment.yaml
conda activate anytext2AnyText2/eval directory. The usage pattern follows the AnyText evaluation workflow. For long caption versions, updated evaluation files are available in the AnyText-benchmark dataset.For users with an advanced GPU (at least 8GB VRAM), you can run the interactive demo which includes a user interface and usage instructions. The demo parameters are consistent with the original AnyText project.
python demo.py