Emi

🚀 Emi Model Card

Emi (Ethereal master of illustration) is an AI art-specific image generation AI developed by AI Picasso using the state-of-the-art development equipment H100 and the image generation model Stable Diffusion XL 1.0. A notable feature of this model is that it does not learn from unauthorized reposted images on platforms like Danbooru.

Original(PNG)

English Version

🚀 Quick Start

Download Information

To download this model from this page, you need to provide information registered with Hugging Face. The information provided will be used to guide you on how to utilize the image-generation AI.

Demo and Full Usage

• You can use the demo here.
• Those who want to use it seriously can download the model here. If the normal version doesn't work well for generation, you can try the stable version.

✨ Features

• Training Data Integrity: This model does not learn from unauthorized reposted images on platforms like Danbooru.
• Commercial Use: It is licensed under the CreativeML Open RAIL++-M License, allowing for commercial use.

📦 Installation

ComfyUI or Fooocus

Similar to using Stable Diffusion XL 1.0, use the model file in safetensor format. Refer to this article for detailed installation instructions.

Diffusers

Use 🤗's Diffusers library. First, run the following script to install the library:

pip install invisible_watermark transformers accelerate safetensors　diffusers

Then, run the following script to generate an image:

from diffusers import StableDiffusionXLPipeline, EulerAncestralDiscreteScheduler
import torch

model_id = "aipicasso/emi"

scheduler = EulerAncestralDiscreteScheduler.from_pretrained(model_id, subfolder="scheduler")
pipe = StableDiffusionXLPipeline.from_pretrained(model_id, scheduler=scheduler, torch_dtype=torch.float16)
pipe = pipe.to("cuda")

prompt = "1girl, sunflowers, brown bob hair, brown eyes, sky, transparent"
images = pipe(prompt, num_inference_steps=20).images
images[0].save("girl.png")

For complex operations, refer to the source code of the demo.

💻 Usage Examples

Basic Usage

The following are some simple examples of generated images and their corresponding prompts:

positive prompt: anime artwork, anime style, (1girl), (black bob hair:1.5), brown eyes, red maples, sky, ((transparent))
negative prompt: (embedding:unaestheticXLv31:0.5), photo, deformed, realism, disfigured, low contrast, bad hand

positive prompt: monochrome, black and white, (japanese manga), mount fuji
negative prompt: (embedding:unaestheticXLv31:0.5), photo, deformed, realism, disfigured, low contrast, bad hand

positive prompt: (1man), focus, white wavy short hair, blue eyes, black shirt, white background, simple background
negative prompt: (embedding:unaestheticXLv31:0.5), photo, deformed, realism, disfigured, low contrast, bad hand

Advanced Usage

• Improving Output Quality:
  • To ensure anime-style illustrations, add "anime artwork, anime style" at the beginning of the prompt.
  • Adding the word "transparent" to the prompt can result in a more modern painting style.
  • If drawing a full body doesn't work well, try the stable version.
  • The available prompts are the same as those in Waifu Diffusion, and it can also be used like Stable Diffusion.
  • It is recommended to use Textual Inversion in the negative prompt.
  • Since hands may be unstable, it is recommended to merge with a real - life model like DreamShaper XL1.0.
  • Refining prompts using ChatGPT can lead to works beyond your imagination.
  • Using the FreeU node in the latest ComfyUI or the Web UI extension with the following parameters may further improve the output: b1 = 1.1, b2 = 1.2, s1 = 0.6, s2 = 0.4. report

📚 Documentation

Expected Use Cases

• Art Creation Assistance: For creating illustrations, comics, and anime, both commercially and non - commercially.
• Communication in Commissions: Facilitating communication between clients and creators.
• Commercial Image Generation Services: However, handle the generated products with caution.
• Self - Expression: Expressing one's uniqueness using this AI.
• Research and Development:
  • Using the model on Discord for prompt engineering, fine - tuning (such as DreamBooth), and merging with other models.
  • Evaluating the model's performance using metrics like FID.
  • Verifying the model's independence from other models using checksums or hash functions.
• Education:
  • For graduation projects of art college students and vocational school students.
  • For graduation theses and assignments of university students.
  • For teachers to introduce the current situation of image generation AI.
• Hugging Face Community: Ask questions in Japanese or English.

Unexpected Use Cases

• Representing things as facts.
• Causing trouble for teachers.
• Other actions that may have a negative impact on the creative industry.

Prohibited and Malicious Use Cases

• Do not use it for money laundering.
• Do not publish digital forgeries (Digital Forgery) to avoid potential copyright law violations.
• Do not perform Image - to - Image on others' works without permission to avoid potential copyright law violations.
• Do not distribute pornographic materials to avoid potential violations of Article 175 of the Criminal Code.
• Do not spread false information to avoid potential charges of disrupting business operations.

🔧 Technical Details

Model Information

Model Limitations and Biases

• Limitations: Diffusion models and large - scale language models still have many unknown aspects, and their limitations are not fully understood.
• Biases: Diffusion models and large - scale language models still have many unknown aspects, and their biases are not fully understood.

📄 License

The model is licensed under the CreativeML Open RAIL++-M License, which allows for commercial use.

References

@misc{podell2023sdxl,
      title={SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis}, 
      author={Dustin Podell and Zion English and Kyle Lacey and Andreas Blattmann and Tim Dockhorn and Jonas Müller and Joe Penna and Robin Rombach},
      year={2023},
      eprint={2307.01952},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}