Basic usage of stable diffusion 1.5 including text to image generation and inpainting.
SD 1.5 Text to Image Generation Demo:
sd_1.5_text_to_img_demo.mp4
SD 1.5 Inpainting Demo:
sd_1.5_inpainting_demo.mp4
- Python 3.10 (Might work with other versions as well)
- Nvidia drivers installed along with CUDA
- Nvidia GPU with VRAM atleast 6GB
- CPU can also be used but is awfully slow
- Clone the repo and move to the root dir.
- Create a python virtual environment.
- Install torch separately, then install the requirements (Might take some time).
pip install torch==2.3.0
pip install -r requirements.txt
- Either download sd 1.5 models manually or use hugging face models auto-download option.
- Run either of the 1 or 2 for text to image generation, give a prompt and wait for the image generation.
- Run 3 for inpainting, mask the source image (./input_dir/source_image.jpg), give appropriate prompt and watch the magic happen.
This python app, stable diffusion code and model weights are released under the MIT License. See LICENSE for further details.
By forking or cloning this repository, you agree to adhere to responsible usage guidelines. This ensures that the powerful capabilities of Stable Diffusion are used ethically and do not cause harm. Below are the key principles and practices to follow:
1. Ethical Use: Do not use this technology for generating content that is harmful, offensive, or illegal. This includes, but is not limited to, the creation of:
- Pornographic or sexually explicit material.
- Violent or graphic content intended to shock or disturb.
- Discriminatory or hateful imagery against individuals or groups based on race, ethnicity, religion, gender, sexual orientation, disability, or any other characteristic.
- Misleading information, including deepfakes, that could spread misinformation or false narratives.
2. Respect Privacy: Ensure that the generated images do not infringe on the privacy rights of individuals. Avoid creating images that depict real people without their consent, especially in compromising or damaging contexts.
3. Intellectual Property: Respect intellectual property rights by not using this technology to create and distribute images that infringe on copyrights, trademarks, or other proprietary rights. Ensure that generated content is either original or used in a manner compliant with fair use guidelines.
4. Transparency and Accountability: Be transparent about the use of AI-generated content. Clearly disclose when images are created using Stable Diffusion, especially in contexts where the origin of the content might be significant ( e.g., news, research publications, social media).
5. Mitigate Bias: Be aware of and actively mitigate any biases present in the AI model. Avoid generating content that reinforces harmful stereotypes or biases. Strive for fairness and inclusivity in the images you create.
6. Environmental Considerations: Use computational resources efficiently and be mindful of the environmental impact of running large models. Where possible, optimize your use of hardware and consider the energy consumption associated with training and inference processes.
7. Continuous Learning and Improvement: Stay informed about the latest developments, best practices, and ethical guidelines in the field of AI and machine learning. Be open to updating your practices in line with new insights and recommendations.
By following these guidelines, you contribute to the responsible development and use of AI technologies, ensuring they benefit society while minimizing potential harm.