AI로 실제사진 만들어 내는 방법.너무 쉽고 간단해서 바로 할 수 있습니다. AI 사진, AI 그림

4 Mar 202314:50

TLDRThe video script outlines a step-by-step guide on creating realistic images using AI through Stable Diffusion's web UI. It emphasizes the importance of downloading necessary files, setting up the environment using Google Colab, and fine-tuning the AI with specific prompts. The guide also touches on avoiding commercial use without proper attribution to the model and its creators, ensuring ethical and responsible use of the technology.


  • 🎨 The video provides a tutorial on creating AI-generated images that resemble real-life photographs.
  • 🔗 The process involves downloading four specific files: a checkpoint file (7 Outmix), a lora file for facial details, a vae file for image quality enhancement, and a negative prompt file to avoid unwanted elements.
  • 🌐 The tutorial introduces the use of Stable Diffusion web UI for generating the images, which can be accessed via Google Colab for users without high-end graphics cards.
  • 🔄 The video emphasizes the importance of following the steps carefully, without giving up, even if the process seems complex or the instructions are in English.
  • 📂 The files downloaded need to be uploaded to Google Drive and then used within the Stable Diffusion web UI for seamless integration and operation.
  • 🖼️ The quality of the generated images can be fine-tuned by adjusting various settings such as sampling method, steps, and other generation options.
  • 📝 Both positive (desired features) and negative (unwanted features) prompts can be input to guide the AI in creating the desired image.
  • 🔢 The 'cfg scale' option determines how closely the AI adheres to the user's prompt, with a balance needed to avoid either too much or too little adherence.
  • 🛠️ The tutorial mentions the possibility of using AIprm from ChatGPT or Google Translate to assist with inputting prompts if the user finds it challenging.
  • ⚠️ There are restrictions on using the checkpoint file for commercial purposes, and users must credit the model and include a link to the model card if they plan to use or host the model or its derivatives.
  • 📅 The video concludes with a promise of a follow-up video that will delve deeper into optimizing and refining the image generation process.

Q & A

  • What is the main topic of the video?

    -The main topic of the video is about creating realistic images using AI through the Stable Diffusion web UI.

  • What are the four files that need to be downloaded before starting?

    -The four files that need to be downloaded are the Checkpoint file (7 Outmix), the Lola file for facial details, the VAE file for image post-processing, and the Negative Prompt file to prevent unwanted elements in the generated images.

  • How can one access the Stable Diffusion web UI?

    -To access the Stable Diffusion web UI, one can either search for it on Google or click the link provided in the video description.

  • What is the purpose of the Checkpoint file?

    -The Checkpoint file, named 7 Outmix, serves as the overall model that helps generate the entire image.

  • What is the role of the Lola file?

    -The Lola file is a model that focuses on learning and generating specific parts of the image, typically the face.

  • Why is the VAE file important?

    -The VAE file is crucial for post-processing the generated images, ensuring they have a more photo-like quality.

  • What does the Negative Prompt file do?

    -The Negative Prompt file helps to prevent unnecessary elements, such as extra fingers or limbs, from appearing in the generated images.

  • How can one use Google Colab for this process?

    -Google Colab allows users to access and use Google's network and computers for the process, enabling the execution of tasks that would require high-performance computers, even if the user's own computer has lower specifications.

  • What are the key components of the prompt input in the Stable Diffusion web UI?

    -The key components of the prompt input include the Positive Prompt for specifying the desired image features, the Negative Prompt for specifying elements to exclude, the sampling method, the number of steps for AI to take, and additional generation options like width, height, and seed for specific image results.

  • How can users ensure they are using the models responsibly?

    -Users should ensure they are not using the models for commercial or semi-commercial purposes unless specified. They should also credit the model names and include links to the model cards when hosting or using the models or their derivatives.

  • What is the advice given for users who find the prompt input challenging?

    -For users who find the prompt input challenging, the video suggests using AIprm from the Chat GPT or translating the provided script into a format that can be inputted into the Stable Diffusion web UI.



📝 Introduction to AI Image Creation Process

The speaker, Titan, introduces the complex process of creating AI-generated images. They mention that the process is intricate and has been delayed due to the challenge of simplifying it for the audience. The speaker promises to provide a method for creating photo-like images without going into the complex details in this session, and plans to cover those in-depth in future videos. They emphasize the importance of patience and persistence when following the tutorial, assuring that it is not as difficult or complicated as it may seem at first glance.


📂 Preparing and Installing Necessary Files

The speaker guides the audience through the preliminary steps of downloading and installing four specific files required for the image generation process. These files include a checkpoint file named '7 Outmix', a 'Lora' file for facial details, a 'VAE' file for image refinement, and a 'Negative Prompt' file to prevent unwanted features. The speaker provides links for downloading these files and explains the installation process of Stable Diffusion Web UI, offering both local and Google Colab cloud options. They detail the steps for setting up the files in Google Drive and accessing the Stable Diffusion Web UI for further instructions.


🖌️ Configuring the Stable Diffusion Web UI and Prompts

The speaker delves into the configuration of the Stable Diffusion Web UI, explaining how to upload and set up the previously downloaded files. They provide a step-by-step guide on how to use the interface, including selecting the checkpoint, Lora, and VAE files. The speaker also explains the importance of the Positive and Negative Prompts for guiding the AI in creating the desired image while avoiding undesired elements. They discuss the various settings such as sampling method, steps, and other generation options that influence the quality and characteristics of the final image. The speaker emphasizes the balance between giving the AI enough freedom and providing clear instructions to achieve the best results.




Artificial Intelligence (AI) refers to the simulation of human intelligence in machines that are programmed to think and learn like humans. In the context of the video, AI is the core technology behind the creation of realistic images and is used to generate human-like photographs through complex processes.


Photorealistic refers to the creation of images or visuals that closely resemble photographs, achieving a high level of detail and realism. In the video, the goal is to produce photorealistic images using AI, which involves various models and techniques to ensure the output looks like a real photograph.


In the context of AI and machine learning, a model refers to a system that is trained to process input data and produce accurate output predictions or decisions. The video mentions several types of models, such as '체크 포인트' and '로라', which are used to generate specific parts of the AI-generated images.

💡Stable Diffusion

Stable Diffusion is a type of AI model that utilizes diffusion processes to generate images or other types of media. It is known for its ability to produce high-quality, photorealistic images from textual descriptions. In the video, Stable Diffusion is likely one of the technologies used to create the images.

💡Web UI

Web UI stands for Web User Interface, which is the platform or medium through which users interact with a particular service or application over the internet. In the video, the speaker discusses the installation and use of Stable Diffusion through its Web UI, which allows users to generate images without the need for extensive technical knowledge.

💡Google Colab

Google Colab is a cloud-based platform offered by Google that allows users to run Python programs in a collaborative environment. It is often used for machine learning and data analysis tasks, as it provides free access to computing resources and GPU support.


In the context of AI and machine learning, a checkpoint refers to a point during the training process where the model's state is saved. This saved state can be used to resume training or to infer new data without starting from scratch. The video mentions a '체크 포인트' file, which is a model checkpoint used for generating the overall image.


Variational Autoencoder (VAE) is a type of generative model that learns to encode and decode data, often used for image generation and compression. VAEs are capable of creating new data points that are similar to the training data, which makes them useful for post-processing images generated by AI.

💡Negative Prompt

A negative prompt, in the context of AI image generation, refers to a set of instructions or keywords that tell the AI what elements to avoid or exclude from the generated image. This helps guide the AI to create images that meet specific criteria by preventing unwanted features.

💡Sampling Method

Sampling method in AI image generation refers to the algorithmic approach used to select data points from the model's latent space to create an image. Different sampling methods can affect the quality and uniqueness of the generated images, with some methods producing more varied or higher-quality results.

💡CFG Scale

CFG Scale, or Control Flow Graph Scale, is a parameter in AI image generation that adjusts the level of control the user has over the AI's creativity. A lower CFG Scale gives the AI more freedom to deviate from the prompt, while a higher value constrains the AI to follow the prompt more closely.


The speaker introduces themselves as Titan and explains that they are an AI, not a real person.

The process of creating AI-generated images is complex, but the speaker aims to simplify it for the audience.

Four files need to be downloaded before starting: Checkpoint, Lola, VAE, and Negative Prompt files.

The Checkpoint file, named '7 Outmix', provides the overall image model.

The Lola file focuses on specific parts, typically the face, and is a model that concentrates on learning these areas.

The VAE file is responsible for post-image generation adjustments to achieve a more photo-like quality.

The Negative Prompt file prevents unnecessary elements, such as extra fingers or limbs, from appearing in the generated images.

The speaker guides the audience through installing Stable Diffusion Web UI, which can be done locally or using Google Colab.

Google Colab allows users to access Google's network and utilize Google's computers, enabling high-performance capabilities regardless of the user's computer specifications.

The speaker provides a step-by-step guide on how to access and use Google Colab for the image generation process.

Once Stable Diffusion Web UI is installed, users need to upload the four downloaded files to Google Drive and set them up within the UI.

The speaker explains the purpose of each section in the UI, including the Positive Prompt, Negative Prompt, Sampling Method, and other generation options.

The speaker emphasizes the importance of balancing the cfg scale to allow AI autonomy while ensuring the desired output.

The speaker provides a default prompt template for users to begin generating images, which can be adjusted as needed.

Users are instructed to input the names of the downloaded files into the prompt to ensure proper functionality.

The speaker discusses the limitations and guidelines for using the Checkpoint file, emphasizing the need for proper attribution and non-commercial use.

The speaker concludes by promising a future video that will delve deeper into optimizing and improving the image generation process.