Stable Diffusion - how to write the best Prompts… this will surprise you!

Levende Streg
14 Jan 202311:10

TLDRThe video script discusses strategies for crafting effective prompts for Stable Diffusion, an AI image generation tool. It explores alternatives to Google Colab, such as RunDiffusion and, and how they handle prompts. The transcript emphasizes the importance of the first words in a prompt and the use of brackets and parentheses to adjust the weight of different elements. It also touches on the challenges of creating comic book illustrations and dynamic poses with AI, suggesting that while AI can assist artists, it cannot replace the creativity and precision of human artists. The video also covers aspects like inpainting and outpainting prompts, and the use of aspect ratio in achieving desired styles. The speaker shares personal experiences and predictions about the increasing role of AI in creative workflows, while maintaining that AI is a tool for artists, not a replacement.


  • 📝 **Prompt Crafting for AI**: The importance of using the correct structure in prompts for Stable Diffusion, with the subject enclosed in curly brackets and additional details for style and composition.
  • 🔍 **Platform Comparison**: Exploring alternatives to Google Colab and assessing how prompts perform on different platforms, such as RunDiffusion and
  • 🖼️ **Outpainting and Inpainting**: Techniques for using AI to extend canvases and fill in missing parts of images, with careful instructions to guide the AI.
  • 🎨 **Art Style Considerations**: The challenge of creating comic book-style illustrations with clear outlines and colors, and the current preference for photorealistic or 3D styles.
  • ⏱️ **Efficiency vs. AI**: Acknowledging that for certain tasks, like drawing hands and dynamic poses, it's currently faster and more effective to do it manually.
  • 🤖 **AI as a Tool, Not a Replacement**: The assertion that AI will not replace artists but serve as a tool to assist them, especially in iterative creative processes.
  • 🔄 **Model Switching**: The capability of switching between different AI models within platforms like RunDiffusion to suit various creative needs.
  • 📈 **AI Evolution**: The prediction of increased usage of AI in creative workflows as the technology improves and users become more adept at utilizing it.
  • 📐 **Aspect Ratio Impact**: The significance of aspect ratio in determining the outcome of image generation, with different styles favoring different ratios.
  • 🖌️ **Img2Img Prompts**: Utilizing img2img prompts to refine existing artwork, particularly for backgrounds and poses that AI struggles with.
  • 🔗 **Client Feedback Application**: How AI can make it challenging to apply specific client feedback, such as changing angles or colors, compared to manual adjustments by an artist.
  • 🌐 **Community Learning**: The value of community engagement and learning from shared experiences, as well as the encouragement to continue creating despite imperfect conditions.

Q & A

  • What is the main topic of the video script?

    -The main topic of the video script is about creating the best prompts for Stable Diffusion, exploring alternatives to Google Colab, and discussing the use of AI in creative workflows.

  • What is RunDiffusion and what does it claim to offer?

    -RunDiffusion is a site that allows users to set up Stable Diffusion quickly, claiming to do so in as little as 3 minutes. It is designed to integrate smoothly into a user's workflow with full functionality.

  • How does the use of curly brackets and parentheses affect the prompt for Stable Diffusion?

    -The curly brackets '{}' in a prompt are used to indicate the most important part of the image the user wants to be shown. Parentheses are used for upweighting, emphasizing elements that are more important, while square brackets are used for downweighting, indicating less important elements.

  • What are the challenges in creating comic book illustrations with Stable Diffusion?

    -Creating comic book illustrations with clean outlines and colors is more difficult with Stable Diffusion compared to photorealistic or 3D styles. It requires more time and effort to achieve the desired quality and character traits.

  • Why does the speaker prefer to draw comic book characters themselves?

    -The speaker prefers to draw comic book characters themselves because it is more productive and enjoyable. They also find that the quality and special character traits are difficult to achieve with Stable Diffusion.

  • What is the speaker's current usage of AI art generation in their client work?

    -The speaker currently uses AI art generation for about 2% to 5% of their work for clients, but they predict this percentage will increase as AI technology improves.

  • Why does the speaker believe AI will not replace artists?

    -The speaker believes AI will not replace artists because it is difficult to get precisely what you want with AI, and it cannot replicate an artist's ability to make adjustments based on feedback or visualize strategic content on the fly.

  • What is the advantage of signing up for the Creator's Club on RunDiffusion?

    -Signing up for the Creator's Club on RunDiffusion allows users to switch between different models, which can be beneficial for various types of prompts and tasks, such as txt2img, outpainting, and others.

  • How has evolved and what features does it offer for prompt engineering? has evolved to become more helpful with prompt engineering, allowing users to specify the model within the prompt, create the right dimensions, play with aspect ratios, and keep their prompts private.

  • What are some terms that should be avoided when using Stable Diffusion to create drawings?

    -Terms like '4K', 'unreal engine', or 'aperture blur' should be avoided as they can confuse the engine. Instead, terms that refer to drawing styles, such as 'cross hatching', 'outline', 'digital ink', or 'pencil drawing', should be used.

  • How does the aspect ratio affect the output of Stable Diffusion?

    -The aspect ratio significantly influences the output of Stable Diffusion. Some styles look better in certain aspect ratios, and portraits often appear better in tablet sizes. The engine provides different results based on the aspect ratio chosen.

  • What is the speaker's approach to using img2img prompts with Stable Diffusion?

    -The speaker uses img2img prompts to fix up already created artwork and illustrations, focusing on elements like poses and colors. They find that Stable Diffusion is not very good at handling hands and dynamic poses, so they prefer to draw those elements themselves.



🎨 Optimal Prompts for Stable Diffusion & Alternatives to Google Colab

The video begins with an exploration of crafting the best prompts for the AI model Stable Diffusion. It also mentions checking out two alternatives to Google Colab and evaluating how prompts perform in different settings. The host plans to discuss techniques for outpainting and inpainting with prompts and integrating AI into the creative process. The first focus is on RunDiffusion, a tool that promises to set up Stable Diffusion in minutes. The host shares their experience with RunDiffusion, highlighting its ease of use and visual appeal. They delve into the importance of prompt structure, emphasizing the significance of the curly brackets and the use of parentheses and square brackets for weighting elements of the prompt. The video touches on the challenges of creating comic book illustrations and dynamic poses with Stable Diffusion, suggesting that traditional drawing might still be more efficient. However, the host is optimistic about the future of AI in art and its growing role in their workflow. They also discuss the benefits of RunDiffusion's Creator's Club for model switching and the importance of understanding client feedback for customization. Lastly, the host addresses the limitations of AI in replicating the flexibility and creativity of human artists.


🔍 Deep Dive into Prompt Engineering and AI Art Tools

The second paragraph delves into learning more about RunDiffusion, with an upcoming episode promised for further insights. The host shares their experiments with prompt engineering and the discovery that specifying the desired style early in the prompt, just after the curly brackets, works well. They also hint at an upcoming in-depth video on inpainting and outpainting. Transitioning to, the host discusses its utility in prompt engineering and the ability to switch between checkpoint models directly in the prompt, which could be beneficial for businesses and non-technical users. The video highlights the importance of detailed description in prompts for AI to understand the desired output, especially when it comes to drawing styles. It also covers the impact of aspect ratio on the outcome of AI-generated images and how it varies between platforms like Stable Diffusion and Midjourney. The host demonstrates the use of img2img prompts with a personal illustration, emphasizing the tool's utility in refining existing artwork, particularly when it comes to poses and hands that AI struggles with. They also mention the challenges of img2img prompting with non-square images and the need for more computational power, which they find on RunDiffusion.


🖌️ Inpainting and Outpainting Techniques with AI

The final paragraph focuses on the distinct approaches required for inpainting and outpainting prompts compared to img2img or text2img prompts. The host explains that with inpainting and outpainting, only a portion of the image is visible, necessitating careful communication with the AI about the image content. They describe a methodical approach to fixing images part by part, such as hair and eyes, and emphasize the need for detailed explanations in prompts. The host encourages viewers to share their experiences with these techniques in the comments and reminds them that creativity should not wait for the perfect moment, urging them to continue creating.



💡Stable Diffusion

Stable Diffusion refers to a type of generative AI model used to create images from textual descriptions. In the video, it is the main AI tool discussed for generating art, demonstrating its utility in various artistic workflows. The presenter discusses the ease of use and efficiency of Stable Diffusion in different contexts, such as on RunDiffusion and platforms.

💡prompt engineering

Prompt engineering is the process of crafting inputs (prompts) to maximize the effectiveness of AI models like Stable Diffusion. In the video, this concept is crucial as the presenter explains how to compose prompts that result in high-quality images. They emphasize structuring prompts with precise language to guide the AI in generating desired visual outputs, mentioning specific strategies such as using curly brackets and weighting terms differently.


RunDiffusion is presented as an alternative platform for running Stable Diffusion, which allows users to set up and use the AI model quickly. The presenter reviews its ease of use, claiming it took just 5 minutes to set up and start running smoothly, highlighting its functionality and integration into creative workflows.


Inpainting is a technique used in image editing where the AI fills in missing or damaged parts of images. In the context of the video, the presenter discusses how inpainting can be used creatively to complete images or add elements that were not originally captured in the frame, explaining specific strategies for instructing the AI on what to generate.


Outpainting, as discussed in the video, involves extending the borders of an existing image using AI. The presenter describes how outpainting is applied in creating backgrounds for comic books or extending canvases for prints, which requires detailed explanations to the AI on what needs to be generated beyond the original image boundaries.

💡 is another platform mentioned that facilitates the use of Stable Diffusion for creative projects. The video highlights its features like aspect ratio adjustments and model switching within prompts, which are beneficial for artists looking to integrate AI into their creative process.


The img2img technique is a function of AI models like Stable Diffusion where the model transforms an existing image into a new one based on additional input prompts. The video discusses using img2img to tweak the landscape background of a character, showcasing practical applications in digital art creation.

💡comic book illustrations

Comic book illustrations are mentioned as a challenging style to create with AI compared to photorealistic or 3D styles. The presenter shares their personal experience that drawing comic book characters manually often yields better results, due to the detailed and clear outlines required which are difficult for AI to replicate perfectly.

💡Google Colab

Google Colab is briefly mentioned as a common platform for running AI models, with the presenter introducing alternatives to it. This context implies its widespread use in the AI community for projects involving models like Stable Diffusion, before suggesting other platforms that might offer improved performance or features.

💡creative workflow

Creative workflow in the video refers to the integration of AI tools into the artistic processes of the presenter. They discuss how AI art generation currently constitutes a small portion of their client work but predict an increase as AI capabilities improve. This includes feedback mechanisms and adjustments, emphasizing the adaptability needed in creative professions.


The best prompts for Stable Diffusion can be created by focusing on the desired image content within curly brackets {} and adjusting the rest of the prompt for style and composition.

RunDiffusion is a platform that allows for quick setup of Stable Diffusion, boasting a 3-minute setup time and smooth operation.

Prompt templates can be found on Github, and they should be customized in the prompt box for specific needs.

The first words in a prompt are the most important for Stable Diffusion, Midjourney, and DALL-E, with weight decreasing for subsequent words.

Parentheses can be used to upweight important elements in a prompt, while square brackets can downweight less important aspects.

Creating photorealistic or 3D styles is easier with Stable Diffusion than comic book illustrations with clear outlines and colors.

AI art generation is currently used for a small percentage of the work due to the difficulty in achieving certain character traits and dynamic poses.

AI is seen as a tool for artists rather than a replacement, as it cannot replicate the precise adjustments and feedback integration that artists provide.

RunDiffusion offers the ability to switch between models, which is beneficial for different types of prompts such as txt2img, outpainting, and inpainting. is a useful platform for prompt engineering and allows users to specify the model within the prompt for a more tailored output.

Photorealism is more easily achieved with Stable Diffusion compared to drawn styles like anime, which can be challenging.

Using specific terms related to drawing style, such as 'cross hatching' or 'digital ink', can help guide Stable Diffusion to create desired outputs.

Experimenting with aspect ratio can significantly impact the results of image generation, with different styles favoring different ratios.

Img2img prompts can be used to refine existing artwork, leveraging the strengths of artists in areas where Stable Diffusion falls short, like hands and dynamic poses.

Inpainting and outpainting prompts require careful explanation of the visible parts of the image and often need to be addressed in separate parts.

Stable Diffusion can be effectively used for creating backgrounds for comic books and extending canvases, such as old photos.

When working with inpainting and outpainting, it's important to focus on one part of the image at a time for the best results.

Creativity should not wait for the perfect moment; instead, individuals should be encouraged to continuously create and experiment.