Inpainting Tutorial - Stable Diffusion

Sebastian Kamph
6 Apr 202312:31

TLDRThis tutorial delves into the art of inpainting within Stable Diffusion, a technique used to enhance and fix parts of a generated image. The video explains the process of refining details like facial features and adding elements such as a coffee cup to an image. It emphasizes the importance of understanding mask modes, denoising levels, and the use of latent noise for inpainting. Tips on using extensions like canvas zoom and the iterative process of refining the image are also provided, showcasing how to achieve better quality and detail in the final render.


  • 🎨 Inpainting is a valuable technique in Stable Diffusion for improving the quality of generated images, especially for larger fixes.
  • 🖌️ The inpainting model is not necessary but can be helpful; regular models can also be used for inpainting tasks.
  • 🏠 A painter's joke about 'paint on the house' serves as a light-hearted introduction to the tutorial.
  • 🔍 In Stable Diffusion, inpainting is accessed by selecting 'image to image' and then the 'inpainting' tab.
  • 👤 The tutorial focuses on fixing facial features, such as a distorted nose or ear, which are common issues in generated images.
  • 🔍 The 'canvas zoom' extension is recommended for better detail viewing during the inpainting process.
  • 🎭 Mask mode is set to 'inpainting mask' to specify the area that needs to be changed, and 'original' is chosen to keep the content under the mask.
  • 🖼️ The 'in paint area' setting determines the part of the image that will be rendered in full resolution.
  • 🔧 Euler A is a preferred sampling method, and different steps are recommended for various sampling methods like DPM 2M caris and SDE caris.
  • 🔄 Adjusting the denoising strength滑块 allows control over how much the image will be changed, with higher values leading to more significant alterations.
  • 🛠️ Additional elements like a coffee cup can be added to the image by changing settings and using the 'latent noise' option or sketching the item in the 'inpainting sketch' mode.

Q & A

  • What is inpainting in the context of Stable Diffusion?

    -Inpainting in the context of Stable Diffusion is a technique used to improve or modify parts of a generated image, particularly when there are imperfections or details that need enhancement.

  • Is the inpainting model necessary for making improvements to a generated image?

    -The inpainting model is not necessary, but it can be helpful for making larger fixes to the generated images.

  • How does the mask mode work in inpainting?

    -The mask mode in inpainting is set to 'inpaint mask' when there is an area of the image that has been altered or painted over, which indicates what part of the image should be changed. If the rest of the image needs to be changed, 'inpaint not masked' would be the appropriate choice.

  • What is the significance of the 'original' and 'latent noise' options in mask content?

    -The 'original' option is used to keep the content under the mask and use it to create the next iteration of the image, while 'latent noise' is used when there is no content under the mask, and the system generates new content based on the noise.

  • Why is the 'canvas zoom' extension useful in Stable Diffusion?

    -The 'canvas zoom' extension is useful for getting a closer look at the details of the image, which can be particularly helpful when working on intricate parts like faces or other fine details.

  • How does changing the 'in paint area' setting affect the resolution of the image?

    -Altering the 'in paint area' setting allows you to specify which part of the image should be rendered in full resolution. If the entire image is selected, it will maintain the same resolution as the rest of the image, but focusing on a specific area, like a face, will render that part in higher detail and resolution.

  • What are some of the sampling methods mentioned in the script and how are they used?

    -Euler A, DPM 2M caris, and SDE caris are mentioned as sampling methods. Euler A is often set at 25 steps, while DPM 2M caris and SDE caris are used at 30 to 35 steps, although they are slower. These methods are used to refine the image generation process.

  • How does the denoising strength setting impact the inpainting process?

    -The denoising strength setting determines how much the image will be changed. A setting of one will change the image completely, while a setting of zero will not change it at all. Adjusting this setting is crucial for maintaining the desired level of detail and originality in the inpainted area.

  • What is the process for adding a new object to an image using inpainting?

    -To add a new object, you can switch the mask content to 'latent noise' and increase the denoising strength. Alternatively, you can use the 'inpaint sketch' feature to manually draw the object and then adjust the denoising and mask settings to integrate it into the scene.

  • How can you adjust the blurriness of an added object in the image?

    -The blurriness of an added object can be adjusted by adding a blur-related term to the prompt, such as 'blurred' or 'out of focus'. Additionally, the mask blur setting can be tweaked to control the extent and intensity of the blur around the object.

  • What are some tips for achieving better results with inpainting in Stable Diffusion?

    -To achieve better results, it's important to carefully select the mask mode, adjust the denoising strength, and choose the appropriate sampling methods. Additionally, manually sketching elements in 'inpaint sketch' and iteratively refining the image can lead to more satisfactory outcomes.



🎨 Art of Stable Diffusion and Image Refinement

This paragraph introduces the concept of inpainting within the realm of stable diffusion, a technique used to enhance the quality of generated images. It explains that while inpainting models can be helpful, they are not strictly necessary. The speaker shares a personal anecdote about a painter to lighten the mood. The main focus is on using the inpainting feature in stable diffusion to fix imperfections in images, particularly facial features. The speaker provides a step-by-step guide on how to use the inpainting tab, including setting up the canvas zoom extension for better detail. The importance of selecting the correct mask mode and understanding the difference between 'original' and 'latent noise' for mask content is emphasized. The paragraph also discusses the significance of denoising levels and the impact it has on the final image. A practical example is given where the speaker attempts to fix a distorted face and improve image quality by adjusting various settings.


🖌️ Enhancing and Adding Elements to Images

This paragraph delves into the process of adding new elements to an image and the challenges that may arise when using inpainting techniques. The speaker demonstrates how altering denoising levels can lead to vastly different outcomes, from completely changing the subject of the image to leaving it unaltered. The focus then shifts to adding a coffee cup to the scene, highlighting the importance of switching to 'latent noise' mode when there's nothing to base the addition on. The speaker also explains the need to increase denoising strength when working with latent noise. Furthermore, the paragraph explores alternative methods such as sketching the desired element in the 'paint sketch' mode and adjusting denoising levels accordingly. The speaker provides a practical example of adding a coffee cup and improving its integration into the scene through iterative adjustments and blending it with the surroundings.


👁️‍🗨️ Iterative Refinement of Facial Features

In this paragraph, the focus is on the iterative process of refining specific facial features within an image using stable fusion. The speaker guides the audience through enhancing the details of the eyes by adjusting denoising levels and rendering multiple images for better results. The concept of 'mask blur' and 'only masked padding pixels' is introduced to manage the blur around the subject, emulating a Gaussian blur effect. The speaker provides a detailed walkthrough of changing an earring in the image, showcasing the capability of the tool to add intricate details. The paragraph concludes with a reminder that with practice and familiarity with the settings, inpainting in stable fusion becomes an accessible technique for creating advanced scenes with multiple characters and elements. The speaker encourages the audience to like and subscribe if they found the content useful and promises to continue sharing knowledge in future videos.




Inpainting is a technique used in image editing to fill in missing or unwanted parts of an image with new content that matches the surrounding areas. In the context of this video, inpainting is crucial for improving the quality of generated images, particularly in fixing facial features that may not have rendered correctly. The process involves masking the area to be fixed and using the inpainting mode to generate a more detailed and accurate representation of the intended subject matter.

💡Stable Diffusion

Stable Diffusion is a term that likely refers to a stable and reliable diffusion model used in AI-generated imagery. In the video, it is the platform or method through which the inpainting process is applied to enhance images. The main theme revolves around using this system to achieve better-looking images by correcting errors or adding new elements without compromising the overall composition or quality.

💡Mask Mode

Mask mode is a setting in image editing that allows users to isolate specific areas of an image for manipulation while protecting other areas from changes. In the video, the mask mode is set to 'inpaint mask' to focus the inpainting process on the designated masked area, such as the face, ensuring that only the targeted part is altered while the rest of the image remains unchanged.

💡Canvas Zoom

Canvas Zoom is a feature that enables users to magnify areas of an image for detailed work. In the context of the video, it is mentioned as an optional extension that can be installed for better precision when working on inpainting tasks. This tool is particularly useful for achieving higher resolution details in the inpainted areas.


Resolution refers to the clarity or sharpness of an image, measured by the number of pixels. In the video, the goal is to improve the resolution of specific parts of the image, such as the face, by using the inpainting technique to render those areas in higher detail, resulting in a more realistic and visually appealing final image.

💡Sampling Method

A sampling method in the context of AI image generation is an algorithm used to select or generate new pixel values based on the input data. Euler A, mentioned in the video, is a sampling method that can be used during the inpainting process to refine the quality of the generated image. The choice of sampling method can affect the final appearance and detail level of the inpainted content.


Denoising is the process of reducing noise or unwanted visual artifacts in an image. In the video, denoising strength is an adjustable parameter that determines how much the AI will alter the image during the inpainting process. A higher denoising value results in more significant changes, potentially improving the quality of details in the inpainted area.

💡Latent Noise

Latent noise refers to the underlying random variation in the AI model's generated images. In the context of the video, switching to latent noise allows the model to create new content for the inpainting process, especially when there is no suitable content in the original image to work with, such as when adding a new object like a coffee cup.


Upscaling is the process of increasing the resolution of an image, often to enhance its quality or to prepare it for larger displays. In the video, the presenter discusses using inpainting to upscale images and improve details, such as increasing the resolution of the face to achieve a clearer and more detailed appearance.


In image editing, a mask is a tool that hides or reveals certain parts of an image. In the video, the presenter uses a mask to select the area of the image that needs inpainting. The mask is essential for localized editing, ensuring that the changes made are confined to the specified area and do not affect the rest of the image.


迭代, in English 'iteration', refers to the process of repeating a procedure with each successive step building on the results of the previous ones. In the context of the video, iteration is used to gradually refine the inpainting process, making adjustments and generating new images until the desired outcome is achieved. This could involve tweaking settings, adding new elements, or improving the quality of specific details.


Inpainting is a key technique for enhancing images generated by Stable Diffusion.

The inpainting model is not necessary, but it can be helpful for significant corrections in images.

The tutorial begins with a humorous anecdote about a painter to establish a friendly tone.

The process of inpainting in Stable Diffusion starts with selecting the 'image to image' and then the 'inpainting' tab.

When inpainting, it's crucial to select the correct mask mode and options to target the desired changes.

The 'canvas zoom' extension is recommended for better detail viewing during the inpainting process.

The 'original' mask content setting is used to preserve certain parts of the image while altering others.

For most users, 'latent noise' and 'original' are the two primary options for inpainting.

Adjusting the 'in paint area' setting can enhance the resolution of specific parts of the image.

Euler A sampling method is a preferred choice for its effectiveness in the inpainting process.

Denoising strength determines how much an image will be altered during inpainting.

Negative prompts like 'nfixer' can be used to refine the inpainting process, though not necessary.

When adding new elements to an image, changing the mask content to 'latent noise' and adjusting denoising strength can yield results.

Sketching the desired element in 'inpainting sketch' can help guide the AI in creating a more accurate addition.

Mask blur and padding pixels settings can adjust the blur effect around the object being inpainted.

Iterative adjustments and multiple renderings can lead to progressively better results in inpainting.

Inpainting can be applied to various elements such as faces, accessories, and even entire objects like a coffee cup.

The video concludes with an encouragement to like and subscribe, emphasizing a casual and approachable learning environment.