Stable diffusion VS Midjourney: All you need to know

The video compares two leading AI image generators, Stable Diffusion and Midjourney. Stable Diffusion is an open-source, customizable text-to-image generator that's free but requires technical knowledge to use effectively. Midjourney, on the other hand, offers high-quality results with less customization but is subscription-based and more user-friendly, requiring only a Discord account. Both tools use different training methods, with Stable Diffusion learning through a noise addition and reduction process, while Midjourney likely combines Stable Diffusion's approach with a large language model. The video also touches on the legal aspects of AI art, noting that as of August 2023, AI-generated art without human input cannot be copyrighted in the US. However, if a human artist uses AI to create images and then modifies them, the work may be copyrightable. The video concludes by emphasizing the strengths of the open-source approach and inviting viewers to share their preferences in the comments.

Takeaways

• 🎨 **Free and Open Source**: Stable Diffusion is an open-source text-to-image generator available for free, with a customizable model and a supportive community.
• 💰 **Pricing and Accessibility**: Midjourney requires a paid subscription, with its basic plan being expensive and having restrictions on high-speed generation.
• 🛠️ **Customization and Models**: Stable Diffusion offers more customization with thousands of models, while Midjourney has fewer models but provides high-quality results.
• 👶 **User-Friendly**: Midjourney is more beginner-friendly, requiring only a Discord account to use, whereas Stable Diffusion can be challenging for inexperienced users.
• 🌐 **Internet Dependency**: The Midjourney Discord bot needs a constant internet connection, unlike Stable Diffusion, which can run locally or through a cloud server.
• 💻 **Hardware Requirements**: Running Stable Diffusion locally may require a powerful PC to avoid long generation times.
• 🔍 **Training Approach**: Stable Diffusion learns by adding noise to images and attempting to recreate them, while Midjourney's approach is less transparent but likely combines Stable Diffusion with a large language model.
• 📚 **Training Data**: Both tools use extensive datasets for training, with Stable Diffusion relying on fine-tuned models for specific styles and Midjourney using a combination of text and images.
• 🚫 **Content Restrictions**: Midjourney enforces a ban on explicit imagery, while open-source Stable Diffusion has no such restrictions and even includes models for creating adult content.
• ⚖️ **Copyright and Legal Issues**: AI-generated art faces complex copyright issues; in the US, as of August 2023, it cannot be copyrighted if created without human input. However, human-modified AI art may be copyrightable.
• 📈 **Community and Development**: The open-source nature of Stable Diffusion fosters a community that continually expands and improves the technology, potentially outpacing more closed systems like Midjourney.

Q & A

What are the main differences between Stable Diffusion and Midjourney AI in terms of accessibility and cost?

Stable Diffusion is an open-source text-to-image generator available for free to anyone, while Midjourney requires a subscription that is significantly priced, with the basic plan being almost as expensive as the Netflix standard pricing.

How does the customization model of Stable Diffusion compare to Midjourney?

Stable Diffusion offers an extremely flexible customization model with thousands of custom models tailored to specific styles, whereas Midjourney is less customizable with only a couple of models but provides high-quality results.

What is the technical difficulty level associated with using Stable Diffusion?

Stable Diffusion is harder to run for an inexperienced user and requires a significant amount of learning to master.

What is the basic requirement to start using Midjourney?

To use Midjourney, all you need is a Discord account, making it more beginner-friendly.

How does the training approach of Stable Diffusion differ from Midjourney?

Stable Diffusion learns to generate images by adding layers of noise and attempting to recreate the original from the noise, while Midjourney is believed to combine Stable Diffusion's approach with a large language model to understand the relationship between text and images.

What is the source of the images used for training Stable Diffusion and Midjourney?

Stable Diffusion is based on a massive dataset containing various art pieces, while Midjourney has used the LAION-5B dataset, which has more than 6 billion images with text descriptions.

What legal issues have arisen from the use of copyrighted material in AI training?

Midjourney faced a class action copyright infringement lawsuit this year due to the use of copyrighted material in training. Stable Diffusion, being free, is not under the same scrutiny but claims any image created with it can be used commercially.

Can AI-generated art be copyrighted in the US as of August 2023?

AI-generated art cannot be copyrighted in the US because the copyright laws only protect works created by human beings. However, if a human artist uses AI to generate images and then modifies or arranges those images creatively, the resulting work may be subject to copyright.

What are the strengths of the community-built fine-tuned models of Stable Diffusion?

The community-built fine-tuned models of Stable Diffusion are trained on a narrower dataset and can produce the chosen style quite closely, making them more detailed and nuanced for specific styles like anime.

How does Midjourney ensure the quality and relevance of its generated images?

Midjourney has a single, constantly updated model that is trained more meticulously, which results in higher quality images that closely match the prompt.

What are the restrictions on explicit content in Midjourney?

Midjourney has a strictly enforced ban on any explicit imagery, ensuring that the content generated is safe for all users.

What is the final verdict on which AI image generator might be more suitable for users?

The choice between Stable Diffusion and Midjourney depends on the user's needs. Stable Diffusion is free and flexible but requires more technical insight, while Midjourney is easier to use and provides better results on average.

Outlines

00:00

🎨 AI Art Generation: Stable Diffusion vs. Midjourney

The video discusses the current state of AI art, focusing on two leading examples: Stable Diffusion and Midjourney. Stable Diffusion is an open-source text-to-image generator that's free and customizable but can be challenging for beginners to use. It supports a wide range of styles and has a community that continuously expands its capabilities. Midjourney, contrastingly, is a proprietary service with a high subscription cost and limited customization. It offers fewer models but produces high-quality results and is more user-friendly, requiring only a Discord account. The video also touches on the technical training differences between the two, with Stable Diffusion using a noise addition and removal process, while Midjourney's approach remains a mystery due to its closed-source nature. Both tools have faced legal scrutiny over the use of copyrighted material in their training datasets.

05:03

🌟 Community Power vs. Proprietary Precision

The video contrasts the community-driven approach of Stable Diffusion with the more controlled environment of Midjourney. Stable Diffusion benefits from community-built fine-tuned models, which allow for a wide array of styles and creative uses. However, it requires more technical knowledge and can sometimes necessitate the use of negative prompts to refine the generated images. Midjourney, with its single, constantly updated model, provides higher quality images that closely match the prompts without the need for extensive user input. The video also addresses the issue of explicit content, with Midjourney enforcing a strict ban, while Stable Diffusion, being open-source, has no such restrictions. The discussion concludes with the legal considerations surrounding AI art, noting that as of August 2023, AI-generated art without human input cannot be copyrighted in the US. However, if a human artist uses AI to create images and then modifies them, the work may be eligible for copyright. The video ends by emphasizing the strengths of both tools and inviting viewers to share their preferences and experiences.

Keywords

AI art

AI art refers to the creation of artwork using artificial intelligence. It is a rapidly evolving field that raises questions about the role of human creativity in the artistic process. In the video, AI art is the central theme, with a focus on how AI can generate images from textual prompts, and the ethical and legal implications of such technology.

Stable Diffusion

Stable Diffusion is an open-source text-to-image generator that allows users to create images based on textual descriptions. It is characterized by its flexibility and the ability to support a wide range of custom models, each tailored to a specific artistic style. The video discusses how Stable Diffusion's community-driven approach expands its capabilities and contrasts it with Midjourney's more proprietary model.

Midjourney

Midjourney is a proprietary AI image generator that requires a subscription for use. Unlike Stable Diffusion, it is not open source and offers fewer customization options. The video highlights that while Midjourney is more user-friendly, especially for beginners, it is also more expensive and has restrictions on high-speed image generation.

Customization

Customization refers to the ability to tailor a product or service to meet specific needs or preferences. In the context of the video, customization is a key feature of Stable Diffusion, which supports thousands of custom models, each producing a unique style. This contrasts with Midjourney's more limited customization options, which are centered around a single, constantly updated model.

Discord bot

A Discord bot is an application that operates within the Discord platform, providing various services or functionalities. In the video, it is mentioned that using Midjourney requires a Discord account, and the Midjourney Discord bot facilitates the image generation process. This bot requires a constant internet connection, which is a notable difference from Stable Diffusion, which can be run locally or through a cloud server.

Training data

Training data is the information used to teach a machine learning model how to perform a specific task. The video discusses how both Stable Diffusion and Midjourney are trained on large datasets, with Stable Diffusion using a method that involves adding noise to images and then learning to recreate them, while Midjourney is speculated to combine Stable Diffusion's approach with a large language model trained on text and images.

LAION-5B

LAION-5B is a dataset comprising over 6 billion images, each with a text description. It is mentioned in the video as a significant source of training data for AI image generators. The use of such datasets raises legal and ethical questions about copyright and creator attribution, as the creators of the original images are not credited during AI training.

Copyright infringement

Copyright infringement is the unauthorized use of material protected by copyright law. The video touches on a lawsuit faced by Midjourney related to copyright infringement due to its use of copyrighted material in training. It contrasts this with Stable Diffusion's stance that any image created with its tool can be used commercially, subject to local copyright laws.

Fine-tuned models

Fine-tuned models are machine learning models that have been further trained on a specific subset of data to improve their performance on a particular task. In the context of the video, fine-tuned models of Stable Diffusion are popular for generating images in a specific style, such as anime, and can even replicate the work of specific artists with a certain degree of accuracy.

Negative prompt

A negative prompt is a directive used in AI image generation to specify elements that the user does not want to appear in the generated image. The video notes that with Stable Diffusion, using a negative prompt is almost mandatory to avoid generating undesirable imagery, whereas with Midjourney, the need for such prompts is less common due to the higher quality and closer adherence to the prompt.

Commercial use

Commercial use refers to the use of a product or service for monetary gain or profit. The video discusses the commercial potential of images generated by AI, with Stable Diffusion claiming that its images can be used commercially. However, this use may depend on the user's adherence to local copyright laws and the potential need for human input to qualify for copyright protection.

Highlights

AI art is a hot topic with questions about the availability of high-level AI image generation for free versus paid services.

Stable Diffusion AI is an open-source text-to-image generator available for free to anyone.

Stable Diffusion supports thousands of custom models tailored to specific styles.

Stable Diffusion offers a flexible customization model and has a dedicated community.

Stable Diffusion can be difficult for inexperienced users and requires learning to master.

Midjourney AI is a non-open-source image generator requiring a subscription with high pricing and speed limitations.

Midjourney is less customizable with a few models but produces high-quality results.

Midjourney is beginner-friendly and only requires a Discord account to use.

Midjourney requires a constant internet connection, unlike Stable Diffusion which can run locally or on a cloud server.

Stable Diffusion learns to generate images by adding layers of noise and attempting to reverse the process.

Stable Diffusion's original model is based on a massive dataset, but fine-tuned models are more popular.

Using Stable Diffusion with images from a specific artist can replicate their work with a certain accuracy.

Midjourney is a closed-source tool, combining Stable Diffusion's approach with a large language model.

Midjourney uses datasets like Microsoft's Common Objects in Context to learn relationships between text and images.

Most images used for training come from LAION-5B, a dataset with over 6 billion images with text descriptions.

Midjourney faced a copyright infringement lawsuit due to its use of LAION-5B.

Stable Diffusion claims any image created with it can be used commercially, but users are responsible for local copyright laws.

Stable Diffusion's default model is versatile but not as detailed as Midjourney's.

Stable Diffusion's community-built fine-tuned models can transform videos into animations.

Midjourney's single model is constantly updated and produces high-quality images close to the prompt.

Midjourney enforces a ban on explicit imagery, unlike open-source Stable Diffusion.

AI-generated art can't be copyrighted in the US without human input, but works modified by a human artist using AI may be copyrighted.

Stable Diffusion is free and flexible but requires technical knowledge, while Midjourney is easier to use and provides better results on average.

The open-source approach of Stable Diffusion is seen as a fertile ground for technological growth.