WACV 2026

Pin-Yen Chiu, I-Sheng Fang, Jun-Cheng Chen

Research Center for Information Technology Innovation, Academia Sinica

conda create -n textslider python=3.10 -y
conda activate textslider
pip install -r requirements.txt

We provide ready-to-use notebooks so you can try Text Slider immediately:

• Stable Diffusion XL: SDXL-inference.ipynb
• Stable Diffusion 1: SD1-inference.ipynb

We also include several pre-trained slider checkpoints in the models/ directory. Feel free to try with your own prompts and attributes!

We follow the prompt-generation method from Concept Slider (see their GPT_prompt_helper.ipynb), which requires an OpenAI API key. If you prefer not to use the API, we provide a standalone system prompt in trainscript/prompt_generate.txt. Simply copy–paste it into ChatGPT, provide the attribute you want to train, and it will generate:

• target
• positive
• negative
• preservation

You can then use these to create a new prompt.yaml following our existing format.

Below is an example for training an age slider.

First, move into the training directory:

cd trainscript

Then paste the text after "preservation" (generated from ChatGPT) into the --attributes argument:

python train_text_lora.py \
  --attributes "white race, black race, indian race, asian race, hispanic race ; male, female" \
  --name "ageslider" \
  --rank 4 \
  --alpha 1 \
  --config_file "data/config.yaml" \
  --prompts_file "data/prompts-age.yaml"

If you find our work useful, please consider cite this work as

@inproceedings{chiu2026textslider,
  title={Text Slider: Efficient and Plug-and-Play Continuous Concept Control for Image/Video Synthesis via LoRA Adapters},
  author={Pin-Yen Chiu and I-Sheng Fang and Jun-Cheng Chen},
  booktitle={IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  year={2026}
}