ULTIMATE Upscale for SLOW GPUs - Fast Workflow, High Quality, A1111

Olivio Sarikas
6 May 202313:48

TLDRThis video tutorial offers a solution for enhancing the quality of upscaled images on slow GPUs using the 'Ultimate Upscale' extension for the Automatic1111 software. The process involves dividing images into 512x512 tiles, applying settings, and then recombining them for higher quality results. The video provides a step-by-step guide on installing the extension, using the '4X Ultra Sharp Upscaler' model for improved detail in skin and hair, and adjusting settings such as denoise strength and sampling methods for optimal results. The tutorial also demonstrates how to address issues like pixelation and blurry details through techniques like inpainting and using ControlNet for additional detail. The presenter shares examples comparing the original, upsampled, and Ultimate Upscale results, highlighting the significant improvement in detail and quality. The video concludes with a reminder to join a live stream for more tips and concludes with a call to action for likes and engagement.


  • 📚 Use the 'Ultimate Upscale' extension for Automatic1111 to improve image quality when upscaling from a low resolution to a higher one.
  • 🔗 Install the extension by copying the provided web address into the 'Extensions' tab in Automatic1111 and using the 'Install from URL' option.
  • 📈 After installation, check for updates and restart the UI to ensure the extension is loaded.
  • 📱 For even better quality, consider using the 'Ultra Sharp Upscaler', which is another method detailed in a separate video.
  • 📂 Download the '4X Ultra Sharp Upscaler' model and place it in the correct folder within your Automatic1111 directory.
  • 🧩 The 'Ultimate Upscale' works by dividing the image into 512x512 tiles, processing each tile, and then recombining them for a higher quality final image.
  • 🖼️ Choose the right model for the upscaling process based on the original image used, such as 'Realistic Vision version 2.'
  • ⚙️ Adjust settings such as removing specific prompts, turning off 'Restore Face', and selecting an appropriate sampling method like Euler a with 20-30 steps.
  • 🔍 Set the denoise strength between 0.2 and 0.3 to balance detail and original image integrity.
  • 🖌️ For images with faces, use 'Inpaint' to manually enhance areas like eyes after the initial upscaling.
  • ⏱️ Be aware that using ControlNet with tiling and the Ultimate Upscale script will increase render time but can produce more detailed results.

Q & A

  • What is the main issue with using the 'extras' tab in Automatic 1111 for upscaling?

    -The main issue is that it provides an acceptable quality when upscaling from a small resolution image to a higher resolution, but the quality is not considered great.

  • What is the name of the extension for Automatic 1111 that is used to improve image upscaling quality?

    -The extension is called 'Ultimate Upscale'.

  • How does the Ultimate Upscale extension work?

    -It works by splitting the image into tiles of 512 by 512, applying settings to each tile, and then combining them to create a higher quality image.

  • What is the recommended denoise strength for most cases when using the Ultimate Upscale extension?

    -The recommended denoise strength for most cases is between 0.2 and 0.3.

  • How does the Ultimate Upscale extension affect the face in an image?

    -Since the image is tiled, the full face is not visible at any point during the process, which may cause the 'restore face' feature to fail. It is suggested to turn off 'restore face' to avoid issues.

  • What is the recommended method to fix the details in the face after using the Ultimate Upscale extension?

    -After the upscale render has finished, the image can be sent to 'Inpaint' to mask out the face and then repaint the eyes using a detailed prompt.

  • What are some alternative upscaling methods mentioned in the script?

    -Some alternative methods include using the 4X Ultra Sharp Upscaler and ControlNet 1.1 with tile resampling.

  • How does the ControlNet 1.1 method differ from the Ultimate Upscale extension?

    -ControlNet 1.1 uses the original image tiles and applies the upscaling process to each tile individually, which can result in more detail but might also be more time-consuming.

  • What is the recommended model to use for upscaling anime-style images?

    -The recommended model for upscaling anime-style images is the 'ESRGAN 4X anime' model.

  • What is the significance of the 'seems fix' option in the Ultimate Upscale extension?

    -The 'seems fix' option helps to correct potential issues with seams that may appear in the final upscaled image due to the tiling process.

  • How can one join the live stream mentioned in the script?

    -The script does not provide specific details on how to join the live stream, but it suggests that viewers should not forget to join the live stream for more content.



🖼️ Enhancing Image Quality with Ultimate Upscale

This paragraph introduces the viewer to a method for improving the quality of upscaled images, particularly for those with slower GPUs. The speaker proposes the use of an extension for 'automatic 1111' called 'ultimate upscale' that divides images into 512x512 tiles, applies settings to each, and then recombines them for higher quality. The process involves installing the extension from a provided URL, updating it, and using it in conjunction with the '4X Ultra sharp upscaler' for even better results. The paragraph also discusses the importance of using the correct model for upscaling and adjusting settings such as denoise strength and sampling methods for optimal results.


🎨 Upscaling Techniques for Different Image Styles

The second paragraph delves into the specifics of upscaling different types of images, including photos and anime-style drawings. It explains the use of the 'ultra sharp' upscaler for photos to enhance skin and hair details, and 'r ESR gen 4X anime' for anime-style images. The importance of denoise strength is highlighted, with recommendations on how to balance it for maintaining the original image's integrity while adding details. The paragraph also covers the use of 'ControlNet 1.1' for additional detail enhancement and provides a comparison of the results from different upscaling methods, including the effects of using a high denoise setting.


👁️‍🗨️ Addressing Facial Details in Upscaled Images

The final paragraph addresses the common issue of facial details, particularly the eyes, becoming blurred or distorted after upscaling. The speaker suggests a solution involving the use of 'in paint' to manually enhance the eyes after the initial upscale. It also mentions an alternative method using 'ControlNet with tile' for higher denoise strength and more intricate details, but notes that this approach requires more time and may not be suitable for those with slow GPUs. The paragraph concludes with a reminder of the speaker's live stream and an invitation for viewer engagement.




Upscaling refers to the process of increasing the resolution or quality of an image or video. In the context of the video, upscaling is used to enhance the quality of images, particularly for slow GPUs, by using specialized algorithms to interpolate additional pixels.


An extension is a software component that adds extra features or functionality to an existing application. In the video, the speaker mentions using extensions for Automatic 1111, such as Ultimate Upscale, to improve image quality.


Resolution refers to the amount of detail that an image or video contains, typically measured in pixels. Increasing resolution results in sharper and more detailed images. In the video, the speaker discusses upscaling images from a lower resolution to a higher resolution for better quality.


In image processing, tiling involves dividing an image into smaller, manageable sections called tiles. Each tile is processed individually before being combined back into the final image. In the video, tiling is used as part of the upscaling process to improve image quality.


In the context of image processing, a model refers to a set of parameters or algorithms used to manipulate and enhance images. Different models may produce different results based on their settings and specifications. In the video, various models are mentioned for upscaling images, such as Realistic Vision version 2 and Ultra Sharp Upscaler.


Denoising is the process of removing noise or unwanted artifacts from an image to improve its clarity and quality. It involves applying filters or algorithms to smooth out imperfections without significantly affecting important details. In the video, the speaker discusses adjusting denoise strength to balance between preserving image details and reducing noise.

💡Sampling Method

Sampling method refers to the technique used to select data points or pixels from an image for processing. Different sampling methods can impact the final quality and appearance of the image. In the video, the speaker suggests experimenting with sampling methods to achieve the best results when upscaling images.


In the context of image processing software like Automatic 1111, a script is a set of instructions or commands that automate certain tasks or processes. Scripts can be used to apply specific settings and algorithms to images for desired outcomes. In the video, the speaker discusses using scripts for upscaling images with Ultimate Upscale.


GPU stands for Graphics Processing Unit, which is a specialized electronic circuit designed to accelerate the creation and rendering of images, videos, and 3D graphics. GPUs are commonly used in computers and gaming consoles to improve graphics performance. In the video, the speaker addresses techniques for improving image quality on slow GPUs.

💡Live Stream

A live stream is a real-time broadcast of audio and video content over the internet. It allows viewers to watch and interact with the content as it is being produced. In the video, the speaker mentions hosting a live stream where viewers can learn more about photo editing techniques and image upscaling.


The video discusses an advanced upscaling method for slow GPUs using the 'Ultimate Upscale' extension for Automatic 1111.

Ultimate Upscale splits images into 512x512 tiles, applying settings to each tile for higher quality.

Users can install the Ultimate Upscale extension from a provided URL within Automatic 1111's extensions tab.

The video recommends the 'Ultra Sharp Upscaler' for another high-quality upscaling method.

The 4X Ultra Sharp Upscaler is suggested for use in the models database for high-quality results.

After installing the model, restarting Automatic 1111 is necessary to load the new model.

The Ultimate SD Upscale option is found in the 'Image to Image' tab under 'Script'.

Different upscaling routes are available, including creating a new image or loading one from the PNG info tab.

The model used to create the original image should be the same as the one used for upscaling.

Adjustments for tiling include removing specific image information and turning off the 'Restore Face' feature.

Euler a with 20 to 30 steps is suggested for the sampling method, with 25 steps being optimal.

Denoise strength between 0.2 and 0.3 is recommended for most cases.

ControlNet 1.1 can be used for more detailed upscaling, but it requires additional setup and longer render times.

Examples are provided to illustrate the differences between various upscaling methods and their results.

The video demonstrates how to fix blurry faces post-upscaling by using the 'In Paint' feature.

ControlNet with tile can be used for higher denoise strength and more intricate details.

The final result of using ControlNet and Ultimate Upscale is shown, emphasizing the increased detail and quality.

The video concludes with a reminder to join the live stream for more content and a call to action for likes.