InstantID-SD1.5 Open-Source Model - Supports Training and Inference, Can Be Used in Combination with Multiple Fine-Tuned Models!

Instantid SD1.5

Developed by TheDenk

This is an unofficial InstantID implementation for StableDiffusion 1.5, supporting both training and inference, and can be combined with various fine-tuned models.

Image Generation Open Source License:Apache-2.0 #Facial Keypoint Control #Multi-style Adaptation #Lightweight Training

Downloads 118

Release Time : 7/2/2024

Model Overview

This model is an InstantID implementation for StableDiffusion 1.5, allowing users to utilize InstantID components with various fine-tuned models to generate images with personalized features.

Model Features

Supports StableDiffusion 1.5

Unlike SDXL, this implementation is specifically designed for StableDiffusion 1.5 and is compatible with numerous fine-tuned models.

Includes Training and Inference Code

Not only provides inference capabilities but also includes complete training code, allowing users to customize training as needed.

Facial Keypoint Support

Additional facial keypoint information is included during training, enabling the transfer of different keypoints from other images.

Model Capabilities

Text-to-Image Generation

Image-to-Image Conversion

Facial Feature Retention

Style Transfer

Use Cases

Artistic Creation

Personalized Portrait Generation

Combine InstantID components to generate personalized portraits with specific facial features.

Generate images with highly realistic facial features

Style Transfer

Apply different artistic styles (e.g., Disney Pixar cartoon style) to personalized portraits.

Generate personalized images with specific styles

Entertainment

Role-Playing

Generate images of users portraying specific roles (e.g., rock stars).

Generate personalized images with role-specific features

🚀 InstantID for StableDiffusion 1.5

This is an unofficial implementation of InstantID for StableDiffusion 1.5. SD1.5 has numerous fine - tuned models. You can combine all these models with InstantID components to achieve amazing results.

🚀 Quick Start

This is an unofficial implementation of InstantID for StableDiffusion 1.5. SD15 has a large number of fine - tuned models. So, you can combine all these models with InstantID components to obtain excellent results.

The Official InstantID only works with SDXL and only contains inference code. However, this repository includes both training and inference code. The training process used only 10M images from the LAION - FACE 50M dataset (the original InstantID used 50M Laion - face + 10M custom images).

Feel free to adapt it for your personal use. I'll be glad if someone finds it useful.

✨ Features

Works with StableDiffusion 1.5 and its fine - tuned models.
Contains both training and inference code.
Trained with additional facial keypoints information.
Allows transfer of different keypoints from other images.

💻 Usage Examples

Basic Usage

See Github instruction for detailed usage.

Advanced Usage

Examples with Different Models

epiCPhotoGasm model: Examples with epiCPhotoGasm model + styles from original InstantID.

epicphoto

- **Disney Pixar Cartoon Type A model**: Examples with Disney Pixar Cartoon Type A model + styles from original InstantID.

disneypixart

- **Aniflatmix model**: Examples with Aniflatmix model + styles from original InstantID.

disneypixart

#### Keypoints Visualization and Transfer InstantID SD1.5 components are not compatible with InstantID SDXL. In this work, the model has been trained with additional facial keypoints information. Keypoints visualization:

kp_example