HOW TO MAKE BEAUTIFUL STABLE DIFFUSION IMAGES | Negative Prompts

Binks
13 Jan 202305:02

TLDRIn this informative video, Binks discusses the significance of negative prompts in creating stable diffusion images. He uses Protogen version 3.4, a photorealism model from Darkstorm 2150, to demonstrate how to refine image generation. Binks shares his preferred sampling method, DPM++ SDE Keras, and explains how to adjust settings for better results. He emphasizes the importance of crafting negative prompts to avoid unwanted elements in the generated images, such as canvas frames and disfigurements. By tailoring the original prompt and using detailed keywords, Binks achieves a more accurate and desired outcome. The video concludes with an impressive example of a well-trained model, showcasing the artistry and effort involved in mastering the AI system.

Takeaways

  • 🎨 **Importance of Negative Prompts**: Negative prompts in stable diffusion are crucial for refining the generated images to match the desired outcome.
  • 🖼️ **Avoid Unwanted Elements**: Using 'canvas frame' as the first word in a negative prompt can prevent the AI from adding a canvas frame to the generated image.
  • ❌ **Eliminate Unwanted Features**: Negative prompts can instruct the AI to avoid disfigurement, extra limbs, or mutations in the subject of the image.
  • 👩‍🦰 **Customize the Subject**: Tailoring the prompt to specify the subject, such as 'a portrait of a blonde woman', helps the AI generate a more targeted image.
  • 📈 **Sampling Method and Steps**: The choice of sampling method (e.g., DPM++ SDE Keras) and the number of sampling steps can significantly affect the quality of the generated image.
  • 🖥️ **Image Resolution**: Adjusting the image resolution (e.g., from 512 to 768) can give a more portrait-like look to the generated images.
  • 🔍 **Config Scale**: Increasing the config scale (e.g., to 10) can enhance the image quality when using certain models.
  • 🧩 **Restoring Facial Features**: The 'restore faces' option can help in generating more accurate and less distorted facial features.
  • 📝 **Crafting Detailed Prompts**: Detailed and specific prompts, combined with negative prompts, can lead to more refined and desired outcomes.
  • 🎭 **Stylized Imagery**: Adding descriptors like 'medieval model shoot style' can stylize the generated image according to specific themes or aesthetics.
  • 🔗 **Link to Resources**: The video provides links to useful resources such as the protogen 3.4 photorealism model for further exploration.
  • 📈 **Iterative Process**: The process of generating images with AI involves trial and error, encouraging users to repeatedly generate images to achieve the best results.

Q & A

  • What is the main topic of the video?

    -The main topic of the video is the importance of using negative prompts in stable diffusion for image generation and understanding why the generated images might not match what is seen on the internet.

  • Who is the presenter in the video?

    -The presenter in the video is Binks.

  • What version of protogen is Binks using in the video?

    -Binks is using protogen version 3.4, specifically the photorealism release.

  • What is the significance of negative prompts in image generation?

    -Negative prompts are significant because they help the AI avoid generating unwanted elements in the image, such as a canvas frame or disfigured subjects, allowing for more control over the final result.

  • What sampling method does Binks prefer for protogen?

    -Binks prefers the DPM plus plus sde Keras sampling method for protogen.

  • How does Binks adjust the settings for a portrait look in the generated image?

    -Binks increases the sampling steps to around 30, changes the height from 512 to 768, and turns the config scale up to around 10 to achieve a more portrait-like look.

  • What does 'restore faces' do in the settings?

    -The 'restore faces' option is a default setting that helps improve the quality and appearance of faces in the generated images.

  • Why does Binks tailor the original prompt with more keywords?

    -Tailoring the original prompt with more keywords helps to describe the subject in multiple ways, which can lead to a more detailed and accurate image generation.

  • What is the role of artistry in using AI for image generation?

    -Artistry plays a role in understanding how the system works, crafting tailored prompts, and adjusting settings to achieve the desired image, making the process a blend of technical and creative effort.

  • How does Binks encourage viewers to improve their image generation results?

    -Binks encourages viewers to repeatedly click the generate button, experiment with different prompts and settings, and to ask questions if they need further clarification.

  • What is the final outcome of using a well-trained model with a tailored prompt?

    -The final outcome is a high-quality, stunning image that closely matches the desired specifications, showcasing the capabilities of the AI when used with precision and creativity.

  • Why is it not good practice to use a simple prompt, according to Binks?

    -While a simple prompt can sometimes yield good results, it is not good practice because it lacks the detail and specificity that can lead to more accurate and desired image generation outcomes.

Outlines

00:00

🎨 Introduction to Negative Prompts in Stable Diffusion

Binks introduces the video by discussing the significance of negative prompts in the context of image generation using stable diffusion. He explains that these prompts are crucial for refining the output to match the user's expectations, which often differ from what is commonly found on the internet. Binks also mentions using Protogen version 3.4, a photorealism release by Darkstorm 2150, and provides a link to it in the description. He showcases the capabilities of the model by examining examples on the Protogen page and decides to use a negative prompt from an example image. Binks then demonstrates the process of generating an image with a basic prompt and then refining it by incorporating the negative prompt, which leads to a more accurate and desired result.

Mindmap

Keywords

💡Negative Prompts

Negative prompts are phrases or words used in the context of AI image generation to guide the AI away from including certain elements in the generated image. In the video, Binks uses negative prompts to prevent unwanted features such as a canvas frame or disfigured subjects in the generated image of a medieval blonde woman.

💡Stable Diffusion

Stable Diffusion refers to a type of AI model used for generating images from textual descriptions. It is a diffusion model that can create high-quality images. In the video, Binks discusses the importance of negative prompts within the Stable Diffusion model to achieve desired results.

💡Photorealism

Photorealism is a term used to describe the quality of an image where it closely resembles a photograph in terms of detail and realism. Binks mentions using a 'photorealism release' of the Protogen model to achieve highly realistic images.

💡Protogen 3.4

Protogen 3.4 is a specific version of an AI model used for image generation. Binks uses this model in the video to demonstrate how to create images with negative prompts. It is noted for its photorealistic capabilities.

💡Sampling Method

The sampling method is a technique used in AI image generation that determines how the AI interprets the input to create the image. Binks prefers the 'DPM++ SDE Keras' method for Protogen, adjusting the sampling steps for better results.

💡Sampling Steps

Sampling steps refer to the number of iterations the AI goes through to generate an image. Increasing the number of sampling steps can lead to more refined and detailed images, as Binks does by setting it to 30.

💡Config Scale

Config scale is a parameter in AI image generation that adjusts the intensity or focus of the image details. Binks increases the config scale to 10 for the Protogen model to enhance the image quality.

💡Restore Faces

Restore Faces is an option in AI image generation that attempts to ensure that faces in the generated images are clear and recognizable. Binks checks this option to improve the quality of the generated portrait.

💡Medieval

Medieval refers to the Middle Ages period in history. Binks uses 'medieval' as part of the prompt to guide the AI to generate an image with a medieval aesthetic, which influences the style and elements included in the portrait.

💡Model Shoot Style

Model shoot style indicates a professional photography session where a model poses for pictures, often with a specific theme or aesthetic in mind. Binks includes this in the prompt to direct the AI towards a more stylized and thematic image.

💡AI Artistry

AI Artistry refers to the creative process of using AI to generate art, which involves understanding the AI's capabilities and how to guide it with prompts. Binks emphasizes the artistic aspect of working with AI, where there's a bit of work involved in crafting the prompts to achieve the desired outcome.

Highlights

The importance of negative prompts in stable diffusion is discussed to explain why generated images may not match expectations.

Protogen version 3.4, a photorealism release, is used for generating images.

The video provides a link to Protogen 3.4 in the description for those interested.

Examples of images generated with the Protogen model are shown on the Protogen page.

Negative prompts are used to refine the image generation process, avoiding unwanted elements.

The video demonstrates how to use negative prompts to prevent a canvas frame in the generated image.

The DPM plus plus sde Keras sampling method is recommended for Protogen, with sampling steps increased to around 30.

Increasing the image height to 768 and config scale to 10 enhances the portrait look.

Restoring faces is a default option checked to improve the quality of generated faces.

The first image generated serves as a baseline for comparison after applying negative prompts.

Crafting negative prompts can help avoid disfigured subjects and other undesired features.

A second image generation is performed with a tailored negative prompt to refine the result.

The difference between the baseline and refined images is highlighted to show the effectiveness of negative prompts.

The video emphasizes the artistry involved in using AI to generate images, requiring understanding of the system.

The presenter shares tips and encourages viewers to experiment with the generate button.

A detailed negative prompt is provided in the description for viewers to use for portrait images.

The final generated image is presented as an example of a well-trained model with a tailored prompt and sampling settings.

The video concludes with an invitation for viewers to ask questions and engage with the content.