rotoscopee

Introduction
Rotoscopee is a fine-tuned Stable Diffusion model based on version 1.5, designed to generate images in the style of rotoscope animations. It was trained on screenshots from animations such as A Scanner Darkly, Undone, and Tehran Taboo. The model is particularly effective at rendering characters and can be used by including the token "rotoscopee" in prompts.

Architecture
Rotoscopee is developed using the Stable Diffusion framework, utilizing Dreambooth training techniques by TheLastBen. It leverages specific settings such as 20 steps with the Euler sampler and a CFG scale of 7 for optimal results.

Training
The model was trained on 50 images for 5000 steps. It performs better on characters compared to landscapes or scenes.

Guide: Running Locally

1. Download the checkpoint (ckpt) file from the "Files and versions" tab.
2. Place the file in the Stable Diffusion models folder of your preferred web UI.
3. Use prompts with the token "rotoscopee" to generate images in the rotoscope style.

For enhanced performance, consider using cloud GPUs. Some recommended providers include AWS, Google Cloud Platform, and Azure.

License
Rotoscopee is distributed under the CreativeML OpenRAIL-M license. Key points include:

• Prohibition on using the model to generate illegal or harmful content.
• The authors do not claim rights over generated outputs; users are responsible for their use.
• Redistribution and commercial use are allowed, provided the same use restrictions are maintained and the CreativeML OpenRAIL-M license is shared with users. For full details, refer to the license document.