Stable Diffusion 올바로 사용하기 #1 - 프롬프트와 세팅 설정

DigiClau (디지클로) Lab
26 Mar 202321:25

TLDRThe video script introduces the viewer to the functionalities of Stable Diffusion, focusing on its ability to generate images from text prompts. It explains the process of using the web UI, selecting models like 'checkpoint', and adjusting various settings to achieve desired image quality. The script also discusses the use of negative prompts to avoid undesired features and introduces the concept of 'Lora' models for fine-tuning images. The video aims to educate viewers on leveraging Stable Diffusion's features to create high-quality, customized images.


  • 🌟 The primary feature of Stable Diffusion is text-to-image generation, where AI creates images based on text prompts.
  • 🛠️ The video tutorial covers how to use prompts and various settings within the UI of Stable Diffusion.
  • 📌 Stable Diffusion uses 'checkpoint' models, also known as 'checkpoint models', to generate images.
  • 🎨 The default model provided by Stable Diffusion is version 1.5, but other models can be used for diverse image creations.
  • 🔍 is a recommended website to find and download a variety of high-quality checkpoint models.
  • 🖼️ The 'Text to Image' option in Stable Diffusion must be selected to utilize the prompt feature.
  • 🎥 The quality of generated images can be influenced by settings such as sampling method, steps, and face restoration.
  • 📊 The 'CFG Scale' and 'Sampling Steps' affect the quality and detail of the images, with higher values generally yielding better results but taking more time.
  • 🔄 The 'Tiling' option allows for creating images suitable for tiling, while 'File Resolution Fix' enables upscaling the generated images.
  • 🎨 The 'CLIP' and 'LAION' models serve as additional tools to refine the image generation process, with LAION models being smaller in size and offering variations.
  • 🚫 Negative prompts help to exclude undesired elements from the generated images, while 'NG Deep Negative' further refines this exclusion process.
  • 📚 The video encourages viewers to experiment with different prompts and settings to create a wide range of unique images.

Q & A

  • What is the main feature of Stable Diffusion that the video discusses?

    -The main feature discussed in the video is the Text-to-Image functionality of Stable Diffusion, where AI generates images based on text prompts provided by the user.

  • What are the Stable Diffusion Checkpoints and how are they used?

    -Stable Diffusion Checkpoints are models used in the image generation process within Stable Diffusion. They are selected to determine which model the AI will use to create the images based on the text prompts.

  • How can users find additional models for Stable Diffusion?

    -Users can find additional models by visiting websites like, which hosts a variety of models, including Checkpoint models, inversion models, and others.

  • What is the role of the 'Text-to-Image' option in the Stable Diffusion UI?

    -The 'Text-to-Image' option in the Stable Diffusion UI is where users input their text prompts to instruct the AI on what kind of image to generate.

  • What are some of the settings available in the Stable Diffusion UI for image generation?

    -The Stable Diffusion UI offers various settings such as sampling method, sampling steps, restore face option, tiling, file resolution fix, batch count and size, CF scale, and seed value.

  • How does the 'Negative Prompt' work in Stable Diffusion?

    -The 'Negative Prompt' is used to specify what elements should be avoided or excluded in the generated image. It helps guide the AI to prevent unwanted features or content.

  • What is the significance of the 'Seed' value in image generation?

    -The 'Seed' value is unique to each generated image. Using the same seed value will result in very similar images, allowing users to create a series of images with consistent themes or styles.

  • What is the difference between 'Lora' and 'Checkpoint' models in Stable Diffusion?

    -Lora models are smaller in size and provide minor variations to the base Checkpoint models. They are used to introduce slight changes or adjustments to the generated images without the extensive training or large file sizes of Checkpoint models.

  • How can 'Embeddings' like 'Negative Embeddings' and 'Deep Negative Embeddings' be utilized in Stable Diffusion?

    -Embeddings are AI-trained files that can be used within the text prompts to improve the results of image generation. They help the AI to avoid generating certain negative aspects or unwanted features by learning from the embedded data.

  • What is the purpose of the 'CFG Scale' setting in Stable Diffusion?

    -The 'CFG Scale' setting determines how closely the generated image will follow the user's text prompt. A higher value means the AI will adhere more strictly to the prompt, while a lower value allows the AI to incorporate more of its own interpretation.

  • What kind of results can users expect when using different 'Sampling Methods' in Stable Diffusion?

    -Different sampling methods can yield varying results in terms of image quality and style. Methods like DDIM, DPM Plus Plus, and SDE each have their own strengths and are suited to different types of image generation tasks.



🎨 Introduction to Stable Diffusion and Text-to-Image Features

This paragraph introduces the viewer to the Stable Diffusion platform, highlighting its most commonly used feature, Text-to-Image. The narrator explains that the AI generates images based on text prompts and discusses various options available for customization. The video aims to educate viewers on how to use prompts and UI settings effectively to create desired images. It also advises viewers to refer to a previous installation video if they haven't set up Stable Diffusion on their computers.


🖌️ Exploring Model Options and Settings for Image Creation

The paragraph delves into the different models available for image generation in Stable Diffusion, such as the Checkpoint models. The narrator discusses the process of selecting and downloading models, like the 'Curaian Mix' and 'Delivery' models, and emphasizes the importance of using the right model for desired image outcomes. It also introduces the website as a resource for finding various high-quality models and explains how to download and integrate these models into the Stable Diffusion UI for use.


🌟 Creating Images with Specific Characteristics

This section focuses on the process of creating images with specific features, such as a high-quality, realistic photo of a Korean girl with detailed facial features. The narrator explains how to use the 'Text-to-Image' option, incorporating negative prompts to avoid undesired outcomes. It also discusses the role of the 'Sampling Method' and 'Steps' in determining the quality of the generated images, and how to use the 'Restore Face' option to correct facial anomalies.


📸 Customizing Images with Additional Models and Prompts

The paragraph demonstrates how to further customize image creation by incorporating additional models like 'Lora' to introduce subtle variations into the generated images. The narrator explains the concept of 'Lora' models and how they can be used to modify existing Checkpoint models without significantly impacting performance. It also covers the use of 'Negative Prompts' and 'NG Deep Negative' to exclude certain elements from the images and improve the quality of the final output.


🌈 Experimenting with Different Prompts and Settings

In this final paragraph, the narrator encourages viewers to experiment with various prompts and settings to create a diverse range of images. It provides examples of different prompts that can be used to generate images, such as a college student on a Korean street or a person in a gym. The video concludes by inviting viewers to subscribe and set notifications for future content, emphasizing the creative potential of Stable Diffusion for generating a wide variety of images.



💡Stable Diffusion

Stable Diffusion is a type of AI model used for generating images from text prompts. It is the main focus of the video, where the user is introduced to its functionalities and how to utilize it effectively. The script mentions the process of creating high-quality images by inputting detailed text prompts into the Stable Diffusion model.


Text-to-Image refers to the AI functionality that converts textual descriptions into visual images. In the context of the video, it is the primary feature of Stable Diffusion, where users input descriptive prompts to generate corresponding images.


Checkpoints in the context of the video are specific versions of AI models used in the image generation process. They are crucial for the Stable Diffusion process, as they determine the style and quality of the generated images.

💡UI (User Interface)

UI refers to the interactive interface of the Stable Diffusion web platform that allows users to input text prompts and adjust settings for image generation. It is the medium through which users interact with the AI model.

💡Sampling Method

Sampling Method is a technique used in the image generation process to determine how the AI selects and combines elements from the input prompt to create the final image. Different methods can affect the quality and style of the output.

💡Negative Prompts

Negative Prompts are instructions given to the AI to avoid certain elements or characteristics in the generated images. They help guide the AI to produce images that align more closely with the user's preferences by excluding undesirable features.


Embeddings are AI-trained files that assist in refining the output of the AI model by learning to avoid negative aspects such as incorrect body structures or mismatched colors. They are used in conjunction with the main AI model to improve the quality of generated images.

💡CFG Scale

CFG Scale is a setting within the AI model that determines how closely the generated image adheres to the text prompt provided by the user. Higher values mean the AI will try to follow the prompt more closely, while lower values allow for more creative freedom from the AI.


Seed is a unique value assigned to each generated image that allows for the creation of similar images by using the same seed value. This provides consistency in image generation when a user wants to produce multiple images with a similar theme or style.


LoRa is a type of model used in AI image generation that is smaller in size and provides minor variations to the base model. It is used to introduce slight changes to the generated images without the need for extensive training data or large file sizes.

💡Image Resolution

Image Resolution refers to the dimensions and quality of the generated images. Higher resolutions result in more detailed and larger images, while lower resolutions produce smaller and less detailed images.


Stable Diffuser is a popular feature for generating images from text prompts.

The video explains how to use the Stable Diffuser's web UI and various settings for image generation.

The Checkpoint models, also known as 'Stable Diffuser Checkpoints', are crucial for image generation in Stable Diffuser.

The default model provided by Stable Diffuser is version 1.5, but other models can be used for diverse image creation. is a representative site where you can find and download various Checkpoint models.

The video introduces the process of downloading and using the 'Turboon Mix' model for higher quality image generation.

The settings in Stable Diffuser UI, such as sampling method, steps, restore face, tiling, file resolution fix, and batch count and size, play a significant role in the outcome of the generated images.

The 'cfg scale' determines how closely the generated image follows the prompt, with higher values leading to more accurate adherence.

The 'seed' value ensures that identical seed values produce very similar images, allowing for continuity in image generation.

The 'Lora' model is introduced as a smaller file that can add variations to existing Checkpoint models without extensive training.

The video demonstrates the impact of using different embeddings like 'Lora' and 'Elden Neg' on the final image generation.

The 'Elden Neg' and 'Deep Neg' embeddings are AI-trained files that help avoid undesirable features in the generated images.

The practical application of Stable Diffuser is showcased by generating images of a Korean girl with various styles and settings.

The importance of negative prompts in guiding the AI to avoid certain undesirable outcomes is emphasized.

The video provides a comprehensive guide on utilizing the Stable Diffuser platform for creating diverse and high-quality images.

The process of downloading and integrating additional models and embeddings is detailed, enhancing the user's understanding of the platform's capabilities.

The video concludes with an invitation for viewers to experiment with the platform and create their own unique images, showcasing the creative potential of Stable Diffuser.