【Stable Diffusion】背景を上手く生成させるためのコツと呪文集をまとめて紹介

AIジェネ【AIイラスト生成の情報発信】
18 Aug 202309:11

TLDRThe video provides tips and 'spells' for generating high-quality backgrounds using AI models like 'anylora' and 'anythingV5' from the Civitai distribution site. It suggests using 'background only:2' in prompts and negative prompts like 'character, person, human, people' to avoid generating images with people. For better quality, use phrases like 'masterpiece, best quality, detailed background' and increase image size to 1024 or use 'hires.fix'. Different models may require different sampling methods, so experimentation is key. The video also covers specific background spells for various settings like 'morning', 'sea', 'day', 'evening', and 'sun', along with indoor and outdoor scenarios. It emphasizes that not all backgrounds can be generated with a single model and recommends using multiple models or a specialized model for backgrounds. Adjusting image size can also help achieve the desired outcome. The video concludes by encouraging viewers to subscribe for more informative content.

Takeaways

  • 🎨 Use specific models like 'anylora' and 'anythingV5' from the Civitai model distribution site for generating backgrounds.
  • 🚫 Add 'background only:2' in brackets to the prompt to avoid generating people in the background.
  • 🙅‍♂️ Utilize negative prompts such as 'no character', 'no person', to focus solely on the background.
  • 🔍 Experiment with different models if you consistently get unwanted elements like people in your background.
  • 📈 Enhance the quality of the background by using prompts like 'masterpiece', 'best quality', 'detailed background'.
  • 🖼️ Increase image size to 1024 or more on one side, or use 'hires.fix' to improve the quality of the generated background.
  • 🔄 Change the sampling method if the background style is not as desired to find the best match for the model.
  • 🏞️ Recognize that not all backgrounds can be generated with a single model; consider using multiple models or lora models.
  • 🌅 Add specific time or weather-related spells like 'morning', 'sea', or 'evening' to generate corresponding backgrounds.
  • 🏡 For indoor settings, specify the type of room, such as 'japanese-style room', to get more accurate results.
  • 🌟 If generating the sun or a specific object is challenging, combine it with a background spell like 'sun, sea' for better results.
  • 🔄 Adjusting the image size can sometimes solve issues with background generation if the desired image is not coming out as expected.

Q & A

  • What is the main focus of the video?

    -The video focuses on providing a collection of tricks and 'spells' for successfully generating backgrounds using specific models from the Civitai model distribution site.

  • How can one ensure that the generated images contain only the background and no people?

    -To generate only the background, one can add 'background only:2' to the prompt and use negative prompts like 'character, person, human, people' to omit people from the generated images.

  • What is recommended to increase the quality of the generated background?

    -To increase the quality, one should use the spell 'masterpiece, best quality, detailed background' and consider setting the image size to 1024 or more on one side or using 'hires.fix' to double the image size.

  • Why might changing the sampling method help in achieving the desired style of the background?

    -Different models have different sampling methods that can affect the style of the generated images. Trying all available sampling methods can help find the one that best matches the desired style.

  • What is a common challenge when generating indoor backgrounds?

    -Generating certain buildings or indoor scenes can be difficult because the model may not have been trained on those specific types of backgrounds.

  • How can changing the image size affect the generation of the background?

    -Changing the image size can sometimes help in generating the desired background if the initial attempts are not successful. Reducing or increasing the size can be a trick to successfully generate the image.

  • What should be added to the prompt for generating an early morning background?

    -To generate an early morning background, one should add the spell 'morning' to the prompt.

  • What is the significance of specifying the weather when generating a background?

    -Specifying the weather, such as 'morning, sea', can make it easier to create a more detailed and accurate background that matches the desired atmosphere.

  • null

    -null

  • How can one generate a daytime image?

    -To generate a daytime image, the spell 'day' should be added to the prompt.

  • What is the advice for generating a sunset image?

    -For sunset images, use the spell 'evening'. If a particular model cannot generate the sunset, it is suggested to switch to a different model and try again.

  • What is the approach for generating an image of a Japanese-style room?

    -To generate an image of a Japanese-style room, one should enter 'japanese-style room' as part of the prompt. Using a Japanese-style human model or background model can help create a more authentic image.

  • What is the summary advice for generating only the background?

    -While it is possible to create only the background using a model, it is important to recognize that not all backgrounds can be generated with a single model. Using multiple models or introducing a model that specializes in background generation can be beneficial. Adjusting the image size can also help in successfully generating the desired background.

Outlines

00:00

🎨 Techniques for Background-Only Image Generation

This paragraph discusses the desire to create only the background for various purposes like illustration collections or game development. It introduces 'anylora' and 'anythingV5' from Civitai as models that can be used to generate backgrounds. The speaker provides tips on using prompts and negative prompts to exclude characters and focus on the scenery. They suggest using 'background only:2' in prompts and negative prompts like 'character, person, human, people' to enhance the generation of background-only images. The paragraph also emphasizes the importance of image quality, recommending larger image sizes or using 'hires.fix' to improve it. Different sampling methods are mentioned as a way to achieve the desired style, and the use of various models is encouraged due to the limitations of a single model. Lastly, the paragraph touches on the potential need to adjust image size to successfully generate the desired background.

05:08

🌅 Specific Background Spell Examples and Model Limitations

The second paragraph provides specific examples of spells to generate different types of backgrounds, such as 'morning' for early morning scenes or 'japanese-style room' for indoor settings. It acknowledges that the quality of generated images might vary and suggests that using a Japanese-style model could improve the outcome. The paragraph also covers outdoor spells and summarizes the key points: the possibility of generating only backgrounds with a model, the limitation of a single model to create all types of backgrounds, and the potential need to change the model or image size to achieve the desired result. The speaker concludes by encouraging viewers to try different models and spells, and to subscribe to the channel for more informative content.

Mindmap

Keywords

💡Background Generation

Background generation refers to the process of creating a visual backdrop for an image or scene. In the context of the video, it is about using specific models and techniques to generate only the background without any characters or people, which can be used for illustration collections or game development materials.

💡Model

A model, in this video, refers to a specific algorithm or neural network architecture used in AI to generate images. Different models have different capabilities and are trained on different datasets, which affects their ability to generate certain types of backgrounds.

💡Prompts

Prompts are the textual instructions or descriptions given to an AI model to guide the generation of an image. In the video, the use of 'background only:2' and negative prompts such as 'no character, no person' are mentioned to help the model focus on generating only the background.

💡Negative Prompts

Negative prompts are used to exclude certain elements from the generated image. The video suggests using negative prompts like 'character, person, human, people' to prevent the AI from including people in the generated background.

💡Quality

Quality, in the context of image generation, refers to the visual clarity and detail of the generated image. The video recommends using terms like 'masterpiece', 'best quality', and 'detailed background' to prompt the model to produce high-quality backgrounds.

💡Image Size

Image size is the resolution or dimensions of the generated image. The video suggests setting the image size to 1024 pixels or more on one side, or using 'hires.fix' to double the size, to increase the quality of the background.

💡Sampling Method

The sampling method is a technique used by the AI model to select data points during the image generation process. Different models may perform better with different sampling methods, and the video suggests experimenting with all available options to find the best one for a particular background.

💡Checkpoint Models

Checkpoint models are saved states of a neural network at different stages of training. The video mentions using different checkpoint models or lora models to increase the chances of generating the desired background, as no single model can generate all types of backgrounds.

💡Indoor and Outdoor Spells

Indoor and outdoor spells are specific types of prompts used to generate indoor or outdoor backgrounds. For example, 'japanese-style room' for an indoor setting or 'morning, sea' for an outdoor morning scene. These spells help the AI model to generate backgrounds that match the desired environment.

💡Aicchi

Aicchi seems to be a term used in the video as a form of address or a call for attention, possibly a name or a nickname for the AI assistant or the viewer. It is used to engage the audience and draw attention to the topic of background spells.

💡Ai Gene

Ai Gene is mentioned as a source of information about generated AI. It might be a reference to the channel or the community that shares knowledge and techniques related to AI-generated content, which is relevant to the video's theme of background generation.

Highlights

A collection of tricks and background spells are introduced for successfully generating backgrounds using 'anylora' and 'anythingV5' from the Civitai model distribution site.

Background-only spells may include a lot of people, so adding specific spells to prompts and negative prompts can help generate backgrounds more easily.

Adding 'background only:2' in brackets to the prompt can help generate images with only the background.

Using negative prompts such as 'character, person, human, people' can increase the number of images generated with only the background.

If a person always appears in the generated image, it may be necessary to change to a different model.

To make the background high quality, use the spell 'masterpiece, best quality, detailed background'.

Different models learn differently, so some backgrounds may be possible with one model but not another.

Increasing the image size to 1024 or more on one side, or using 'hires.fix', can improve the quality of the generated background.

The sampling method can be changed if the background does not look like the ideal style.

Natural backgrounds are easier to generate than certain indoor buildings.

Including as many models as possible when creating the background is recommended to increase the chances of successful generation.

If the background cannot be generated well, changing the image size might help create the desired image.

Various background spells will be introduced, such as 'morning' for early morning backgrounds.

Combining spells like 'sun, sea' can make it easier to generate certain elements like the sun.

For generating an image of a Japanese-style room, entering 'japanese-style room' as a spell can help, despite potential blurriness.

Outdoor spells can be used to generate specific environments, such as adding 'day' for daytime images or 'evening' for sunsets.

It is acknowledged that it is impossible to create all backgrounds with a single model and that multiple models or specialized models may be necessary.

The video provides a summary of the tricks and spells for creating a good background and encourages viewers to subscribe for more information.