Img2img Tutorial for Stable Diffusion.

Sebastian Kamph
30 Jun 202323:09

TLDRThis tutorial dives into the image-to-image capabilities of Stable Diffusion, a powerful tool for generating and transforming images. The presenter shares tips and tricks for using the software effectively, emphasizing the importance of the denoising strength setting, which controls the extent to which the original image's features are retained or altered in the transformation process. The video covers various features such as sketching, inpainting, and resizing, demonstrating how to introduce elements like glasses or change facial features while maintaining the composition's integrity. The presenter also discusses the iterative process of refining images, suggesting that while AI can be a one-click solution, achieving desired results often requires multiple adjustments and a good understanding of the tool's capabilities. The tutorial concludes with a discussion on upscaling low-resolution images to higher resolutions, providing viewers with a comprehensive guide to leveraging Stable Diffusion for creative image manipulation.


  • 🎨 Use Stable Fusion for image-to-image transformations with various input types like generated images, photos, or paintings.
  • πŸ“· The denoising strength is crucial in image-to-image settings, controlling how much of the first image is transferred to the second.
  • πŸ” Start with a denoising strength between 0.4 and 0.7 for a balance between retaining the original image and introducing changes.
  • πŸ‘‰ Adjust the sampling method and steps for better results; DPM++ 2M Keras is a good general-purpose sampler.
  • πŸ”„ Iteratively refine the image by adjusting the denoising strength and using features like 'Inpaint' and 'Sketch' for detailed edits.
  • πŸ–ŒοΈ 'Inpaint' can be used to generate specific parts of the image in higher detail, while 'Sketch' can introduce colors into the image.
  • πŸ” The 'Resize' feature can upscale low-resolution images and introduce more detail, with options to control the resizing behavior.
  • πŸ“ˆ Increase the resolution incrementally for better results, but be mindful of the increased GPU power required for higher resolutions.
  • πŸ”§ For more control and finer edits, use external tools like Photoshop, or iterate within Stable Fusion using 'Image to Image'.
  • πŸ”„ It's important to experiment with different settings as there's no one-size-fits-all approach in image generation with Stable Fusion.
  • βš™οΈ Styles and models can be customized and downloaded from external sources to enhance the capabilities of Stable Fusion.

Q & A

  • What is the main topic of the tutorial?

    -The main topic of the tutorial is how to use the image-to-image feature in Stable Diffusion, including tips and tricks for generating images.

  • What types of images can be used with Stable Fusion?

    -Stable Fusion can use a variety of image types, including Stable Fusion generated images, mid-turn generated images, photos, and paintings.

  • What is the significance of the denoising strength setting in image-to-image?

    -The denoising strength setting determines how much of the first image is transferred into the second image, controlling the degree of change between the original and the generated image.

  • What is the recommended denoising strength value for most uses?

    -The recommended denoising strength value for most uses is between 0.4 and 0.7, depending on the desired level of detail and change.

  • How does the resize feature work in Stable Diffusion?

    -The resize feature in Stable Diffusion allows users to increase the resolution of an image while introducing more detail, without simply upscaling and losing quality.

  • What are the different modes available for inpaint in Stable Diffusion?

    -In Stable Diffusion, there are two inpaint modes: regular inpaint, which focuses on a specific part of the image, and inpaint sketch, which allows for the introduction of color into the masked area.

  • How can one introduce more changes into an image during the generation process?

    -To introduce more changes into an image, one can increase the denoising strength slider, which adds more noise to the image and gives the AI more freedom to make alterations.

  • What is the role of the sampling method in the image generation process?

    -The sampling method, such as DPM++ 2m Keras mentioned in the script, is used to determine the algorithm's approach to generating the image, affecting the overall quality and characteristics of the output.

  • What is the purpose of the 'in paint sketch' feature in Stable Diffusion?

    -The 'in paint sketch' feature allows users to add color to specific areas of an image that they want to modify, giving them more control over the final result.

  • How can one improve the quality of generated images through iteration?

    -By continuously generating new images with adjusted settings, such as denoising strength, and selecting the best results to use as the input for the next generation, one can incrementally improve the quality of the generated images.

  • What is the recommended approach when not getting the desired results from image generation?

    -When not getting the desired results, one should adapt the denoising strength value, use different styles, or make adjustments in post-processing tools like Photoshop to refine the image.

  • How does the tutorial suggest using the 'image to image' feature for upscaling low-resolution images?

    -The tutorial suggests using the 'resize' feature within 'image to image' to increase the resolution of the image while maintaining or enhancing the detail, rather than simply upscaling the image.



πŸ˜€ Introduction to Stable Fusion Image-to-Image Tutorial

The video begins with a greeting and an introduction to the tutorial on Stable Fusion, focusing on image-to-image processes. The speaker discusses the new camera setup and engages the audience with a question. The primary subject of discussion is an image of a woman with blue hair, generated from a journey, which will be manipulated using various tools within Stable Fusion. The importance of the denoising strength setting is emphasized, which determines the extent of changes made to the original image. The tutorial covers changing the subject of the image, using different styles, and the impact of the sampling method on the output. It concludes with a demonstration of how varying the denoising strength can lead to different results, from minimal to dramatic changes.


🎨 Adjusting Image Composition with Denoising Strength

This paragraph delves into the nuances of altering an image's composition using the denoising strength slider in Stable Fusion. The speaker illustrates how different values can yield distinct outcomes, from barely altered images to completely transformed subjects. It is highlighted that a denoising strength of around 0.6 is often used for a balance between retaining the original composition and introducing new details. The paragraph also covers the process of adding elements to an image, such as painting on features like glasses or altering lip color, and the iterative process of refining the image through multiple generations with adjusted denoising strength.


πŸ‘“ Iterative Image Refinement and Inpainting Techniques

The focus shifts to iterative image refinement and the use of inpaint sketch features in Stable Fusion. The speaker demonstrates how to improve specific parts of an image, such as enhancing the appearance of glasses or eyes, through repeated generations and adjustments. The concept of 'in painting' is introduced, where parts of the image are masked and regenerated for higher detail. The importance of adjusting the denoising strength for each iteration to achieve the desired level of change is emphasized. The paragraph concludes with a live demonstration of generating images with yellow glasses and red eyes, showcasing the process's dynamic nature.


πŸ–ΌοΈ Image Upscaling and Resizing in Stable Fusion

The final paragraph discusses the process of upscaling and resizing images within Stable Fusion. The speaker explains how to increase the resolution of a low-resolution image while retaining or enhancing its details. Techniques such as 'resize by' are covered, which allow users to scale images up without losing essential content. Different resize modes are introduced, including crop and resize, resize, and resize and fill, each serving a specific purpose when altering the aspect ratio of an image. The paragraph concludes with a live demonstration of upscaling an image of a character standing in front of a castle, showing the improved detail and resolution achieved through the process.


πŸ“ Conclusion and Next Steps

The video concludes with a summary of the key points covered in the tutorial and an invitation for viewers to share their thoughts and ask questions in the comments section. The speaker emphasizes the importance of understanding the image-to-image process in Stable Fusion and encourages viewers to provide feedback. There is also a prompt for viewers to suggest topics for future videos. The speaker signs off with well wishes and a reminder to check out additional guides for more detailed information on specific features of Stable Fusion.



πŸ’‘Image to Image

Image to Image is a process within the Stable Diffusion software that allows users to transform one image into another by adjusting various settings. It is a core concept in the video, as it is the main feature being explained. The process involves using a denoising strength slider to control how much of the original image is retained or altered in the output. For example, setting the denoising strength to 0.6 results in a visually different image but retains the composition of the original.

πŸ’‘Denoising Strength

Denoising strength is a crucial setting in the Image to Image feature of Stable Diffusion that determines the degree of change from the original image to the generated one. A lower value means less change, while a higher value leads to more dramatic alterations. It is used to fine-tune the output, ensuring that the desired level of detail and composition is achieved. In the script, it is mentioned that values between 0.4 and 0.7 are typically used, depending on the desired outcome.

πŸ’‘Sampling Method

The sampling method in the context of Stable Diffusion refers to the algorithm used to generate the new image based on the input. DPM++ 2m Keras is mentioned as a good sampler to use, although it's not the default and may require additional setup. The choice of sampling method can affect the quality and characteristics of the generated images.

πŸ’‘Resize Mode

Resize mode is a setting that determines how an image is scaled up or down in resolution. The script discusses three options: crop and resize, which crops the image to fit the new dimensions; resize, which simply scales the image; and resize and fill, which scales the image and fills in the extra space. This feature is important when upscaling low-resolution images to introduce more detail.


Inpainting is a feature that allows users to 'paint' over certain areas of an image, instructing the software to regenerate those parts. It can be used to fix or enhance specific elements of an image, such as adding glasses or changing the color of the eyes. The script demonstrates how inpainting can be used iteratively to improve the appearance of details in an image.


Sketch is another feature within the Stable Diffusion software that enables users to create a sketch-like version of an image. It can be used to add or modify elements in a more stylized way. The script shows how sketching can be combined with inpainting to achieve different visual effects, such as painting red over the mouth to create red lips.

πŸ’‘Random Seed

A random seed is a starting point for the image generation process that ensures a certain level of unpredictability in the output. In the context of the video, the random seed ensures that each generated image is unique, even when using the same settings. It is an important aspect of the creative process when working with generative AI.


Styles in Stable Diffusion refer to specific sets of parameters that influence the visual characteristics of the generated images. The video mentions that certain styles, such as 'realistic photo portraits,' are not default and need to be downloaded separately. These styles can greatly affect the final look of the images produced by the software.

πŸ’‘Tile Upscaling

Tile Upscaling is a technique used to increase the resolution of an image, especially when working with very high resolutions like 8K. It involves breaking the image into smaller tiles and processing each tile separately to maintain quality. The script suggests using this method when upscaling to very high resolutions to optimize resource usage.

πŸ’‘Iterative Process

The iterative process is emphasized throughout the video as a key approach to working with Stable Diffusion. It involves generating an image, evaluating the results, making adjustments, and repeating the process until the desired outcome is achieved. This method is crucial for achieving more control and fine-tuning the final output.

πŸ’‘Generative AI

Generative AI refers to the use of artificial intelligence to create new content, such as images, based on existing data or input. In the context of the video, Stable Diffusion is a form of generative AI that allows users to generate images through features like Image to Image, inpainting, and sketching. The video demonstrates how generative AI can be a powerful tool for image manipulation and creation.


The tutorial introduces the concept of image-to-image transformation using Stable Diffusion, a powerful tool for generating and modifying images.

The presenter shares tips and tricks for using Stable Diffusion effectively.

Different types of images can be used as input, including those generated by Stable Fusion, mid-turn, photos, or paintings.

The importance of the denoising strength setting is emphasized, which controls how much of the first image is transferred to the second.

The tutorial demonstrates how to change the subject of an image, such as turning a woman with blue hair into a man.

Custom styles can be loaded into Stable Diffusion for more control over the image generation process.

The sampling method and steps are discussed, with a recommendation to use DPM++ 2m Keras for general use.

The concept of image composition is explained, showing how to retain the original composition while making significant changes to the subject.

In-painting and sketching features are introduced as ways to modify specific parts of an image.

The presenter shows how to use the in-painting feature to fix or enhance details within an image.

Iterative image generation is highlighted as a method to achieve the desired outcome by making incremental adjustments.

The tutorial covers how to upscale low-resolution images while introducing more detail.

Different resize modes are explained, including their impact on the aspect ratio and content of the image.

The process of combining image-to-image transformation with other tools like Photoshop for further refinement is discussed.

The presenter emphasizes that generative AI is not a one-click solution and requires iteration for best results.

The tutorial concludes with a reminder to engage with the community through comments and questions for further learning.