Run DALL-E Mini/Mega on Your PC

Table of Contents

1. Introduction
2. Overview of Dolly Models
   • Dolly One and Dolly Two by Open AI
   • Basic Premise of Dolly Models
3. Examples of Dolly 2 Models
   • Astronaut playing basketball with cats in space
   • Prompt customization: Riding a horse as a Pencil drawing
4. Access to Dolly 2
   • Limited access state
   • Beta access application process
5. Replication of Dolly Models
   • Dolly Mini on the Hugging Face Website
   • Issues with website availability
6. Running Dolly Mini on Local Computer
   • GitHub repository for running the model
   • Docker installation requirement
7. Building Docker Image
   • Steps to build the image
   • Recommended GPU specifications
8. Running Dolly Mini Inference Notebook
   • Setting up the notebook
   • Selecting the model version
   • GPU configurations
9. Generating Images with Dolly Mini
   • Modifying parameters for image generation
   • Looping through Prompts
   • Saving and opening generated images
10. Impressive Results and Future Improvements
    • Reviewing the generated outputs
    • Updates on the Dolly Mega model
11. Conclusion

Introduction

In this article, we will explore the Dolly Mini and Dolly Mega models and discuss how to run these models on your own computer. The Dolly models are Based on Dolly One and Dolly Two by OpenAI. These models have the ability to generate images based on textual prompts. We will provide examples of different prompts and their corresponding outputs. Additionally, we will address the limited access state of Dolly Two and the ongoing efforts to replicate the Dolly models. We will then guide you through the process of running the Dolly Mini model on your local computer, using the Hugging Face website and a Docker image. Finally, we will highlight the impressive results achieved with the Dolly Mini model and discuss potential future improvements.

Overview of Dolly Models

Dolly One and Dolly Two are powerful AI models developed by OpenAI. These models are designed to generate images based on textual prompts. The basic premise behind all the Dolly models is that You can provide a sentence describing a picture, and the AI model will generate the corresponding image.

Examples of Dolly 2 Models

On the OpenAI website, you can find examples of the Dolly Two model in action. These examples showcase the model's ability to generate images based on various prompts. For instance, one example shows an astronaut playing basketball with cats in space, rendered in a watercolor style. You can customize the prompts to generate different images, such as a pencil drawing of someone riding a horse.

Access to Dolly 2

Currently, Dolly Two is in a very limited access state. Only a select few individuals have beta access to the model. If you are interested in accessing Dolly Two, you can Apply for the beta program through OpenAI. The application process enables you to request access and potentially be granted the opportunity to use this advanced AI model.

Replication of Dolly Models

Considering the limited access to Dolly Two, efforts have been made to replicate the Dolly models. One such replication is available on the Hugging Face website. However, due to the recent surge in popularity, this replica platform is often overloaded with traffic. As a result, running the Dolly Mini model on the Hugging Face website can be challenging. Nonetheless, there are alternative methods to run the model on your own local computer.

Running Dolly Mini on Local Computer

To run the Dolly Mini model on your local computer, you can utilize the GitHub repository dedicated to this purpose. The repository contains comprehensive information and guides to help you set up and run the model successfully. Before running the Dolly Mini model, ensure that you have Docker installed on your computer. Ideally, you should also have NVIDIA Docker and a GPU with approximately 24 gigabytes of VRAM. However, it is possible to run the model on CPU as well, albeit with longer processing times.

Building Docker Image

To begin running the Dolly Mini model, navigate to the Docker folder in the repository. Inside this folder, you will find the script to build the Docker image. Execute the script to build the image, ensuring that the required dependencies are properly installed. The process may take some time, depending on your system and internet speed. Once the image is built, you are ready to run the Dolly Mini model.

Running Dolly Mini Inference Notebook

To run the Dolly Mini model, execute the script to run the Dolly Mini inference notebook. This notebook provides the interface for interacting with the model and generating images. The script sets up the Jupyter notebook to run on your local machine and specifies the GPU configuration. If you do not have a GPU or your GPU does not meet the recommended specifications, you can modify the configuration accordingly. Importantly, ensure that the Current directory, which is the entire Dolly Mini project, is correctly passed as an argument.

Generating Images with Dolly Mini

Once you have set up the Dolly Mini inference notebook, you can proceed to generate images. The notebook provides options to customize the parameters for image generation, such as the number of predictions per prompt. You can also specify different prompts and observe how the model interprets them. It is fascinating to witness the model's ability to generate diverse images based on textual cues.

Impressive Results and Future Improvements

The generated outputs from the Dolly Mini model can be truly impressive. Regardless of the prompts provided, the model is capable of producing visually appealing and coherent images. While there are limitations, particularly for prompts that are outside of the model's training data, the overall performance is remarkable. It is worth noting that the Dolly Mega model is also being actively developed and improved. Updates from the maintainer of the GitHub repository indicate ongoing training and advancements.

Conclusion

In conclusion, the Dolly Mini and Dolly Mega models represent significant advancements in AI-generated image synthesis. Although access to Dolly Two is limited, alternatives such as the Dolly Mini model can be run on your own local computer. By following the provided guides and utilizing the Docker image, you can experience the capabilities of the Dolly models firsthand. The generated images are visually impressive, showcasing the potential of AI in creative endeavors. As advancements Continue in this field, we can look forward to even more remarkable results from models like Dolly Mini and Mega.

Highlights

• Dolly models by OpenAI generate images based on textual prompts.
• Dolly Two is currently in a limited access state, with a beta application process.
• Dolly Mini can be run on your own local computer using a Docker image.
• The Hugging Face website offers a replication of the Dolly Mini model.
• Generated outputs from Dolly Mini demonstrate impressive image synthesis capabilities.
• The Dolly Mega model is undergoing continuous training and improvement.

FAQ

Q: How can I access Dolly Two? A: To access Dolly Two, you can apply for the beta program through OpenAI. The application process enables you to request access to this advanced AI model.

Q: Can I run the Dolly Mini model without a GPU? A: Yes, it is possible to run the Dolly Mini model on a CPU. However, be aware that the processing times may be longer compared to using a GPU.

Q: Are the Dolly models capable of generating different styles of images? A: Yes, the Dolly models can generate images in various styles, such as watercolor or pencil drawings. The prompt customization allows for flexibility in image generation.

Q: Can the Dolly models generate images for prompts that are not in their training data? A: While the Dolly models can generate diverse images based on various prompts, there are limitations when it comes to prompts outside of their training data. Results may vary depending on the prompt's relevance to the trained data.

Q: What is the difference between Dolly Mini and Dolly Mega models? A: Dolly Mega is an enhanced version of the original Dolly Mini model, offering improved results and performance. The Dolly Mega model is currently being trained and continues to receive updates for better image generation.