caricature_lora_sdxl

Caricature LoRA SDXL

Captain: The AI platform that evolves to your needs

• 🚀 Check out Captain
• 👩‍💻 Captain on GitHub

<Gallery />

Model Overview

This model card showcases a LoRA (Low-Rank Adaptation) model trained on our proprietary Caricature Dataset. The model is fine-tuned to specialize in generating exaggerated and distinctive caricature images, drawing from a diverse set of AI-generated portraits.

Training Configuration

• Dataset: Proprietary Caricature Dataset created via Stable Diffusion (SDXL)
• Epochs: 16
• Number of Images: 174
• Repeats per Image: 10 (Each image was utilized 10 times during training to reinforce learning)
• Optimizer: DAdaptAdam (An advanced optimizer for efficient and dynamic AI training)
• Precision: bf16 (Chosen for the optimal balance of performance and computational resource management)
• Main Trigger: Keywords like “caricature” prime the model to generate images within the caricature domain.
• Xformers: Enabled (Enhancing transformer model efficiency)
• Captioning Method: GPT-Vision (Employed to generate relevant captions, crucial for token shuffling in training)
• Base Model: Stable Diffusion XL 1.0 (A robust foundation for image generation tasks)

Model Usage

Employ this model to create a wide variety of caricatures, each with exaggerated features that highlight the subject's distinct characteristics in a stylized manner.

Performance and Limitations

• Performance: Exhibits a strong ability to vary facial features creatively and with high fidelity to the caricature art style.
• Limitations: May exhibit less diversity in scenarios not covered by the 174 training images.

Ethical Considerations

• Intended Use: The model is purposed for creative and educational applications, particularly in the arts and entertainment sectors.
• Bias and Fairness: Attention has been paid to ensure a diverse representation within the dataset to mitigate biases.

Licensing

• Model License: Licensed under Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International (CC BY-NC-ND 4.0) for non-commercial use.

Contributions and Feedback

We welcome feedback and contributions to improve the model further. If you have suggestions or would like to contribute to the model's development, please reach out through the model's Hugging Face page.

Trigger words

You should use "caricature" to trigger the image generation.

Download model

Weights for this model are available in Safetensors format.

Download them in the Files & versions tab.

Related

https://blib.la/blog/crafting-the-future-blibla-s-ethical-approach-to-ai-model-training

Additional Usage Restrictions for Blibla's LoRAs Hosted on Hugging Face

In alignment with our commitment to ensuring the responsible and ethical use of our models, and in addition to the terms set forth in the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0) under which Blibla's LoRAs are licensed, we hereby impose the following specific restrictions:

1. Prohibited Platforms: Reuploading, redistributing, or offering of image generation services using our models on platforms not owned or operated by Blibla or Hugging Face is strictly forbidden. This includes, but is not limited to, any platforms that host, allow, or promote Not Safe For Work (NSFW) content.

2. Explicitly Forbidden Platforms: For clarity, and without limiting the generality of the above, platforms including but not limited to Leonardo AI, Civit AI, and any "Hugging Face spaces" that host or permit NSFW content are explicitly prohibited from hosting, or utilizing Blibla's LoRAs in any form or manner.

3. Responsibility of Users: Users of Blibla's LoRAs are responsible for ensuring that the environments in which they use, share, or promote our models adhere strictly to these restrictions. Violation of these terms may result in revocation of the license granted under CC BY-NC-ND 4.0 and may prompt further legal action to protect the integrity of our models and the safety of the communities we serve.

4. Purpose of Restrictions: These restrictions are put in place to align with Blibla's ethical standards and the intended use of our models. They are designed to prevent associations with content or platforms that do not reflect our values or the intended application of our technology.

By utilizing Blibla's LoRAs, you acknowledge and agree to these additional restrictions, ensuring that the use of our models remains within the bounds of ethical and responsible practice.