ComfyUI : Ultimate Upscaler - Upscale any image from Stable Diffusion, MidJourney, or photo!

Scott Detweiler
23 Aug 202309:02

TLDRScott Weller, the QA guy for stability.ai, presents his image upscaling technique using the Ultimate SD Upscale tool. He demonstrates how to upscale images to large sizes by leveraging a calculator to determine the optimal size for SD Excel, which has been trained in specific resolutions. The process involves using custom nodes like Control Net, Comfy Manager, and a recommended resolution calculator to ensure the upscaled image matches the required resolutions. The video also covers creating positive and negative prompts for additional guidance and setting the denoising level to add detail without losing the original image's integrity. The final result is an upscaled image broken down into 1024x1024 tiles, providing a detailed and high-quality output.

Takeaways

  • πŸ–ΌοΈ Scott Weller, the QA guy for stability.ai, is sharing his image upscaling technique using SD Excel.
  • πŸ” He wants to upscale images without manually inputting numbers or figuring out resolutions.
  • πŸ“ Scott introduces a calculator to determine the optimal size for upscaling, ensuring it matches SD Excel's training resolutions.
  • πŸ› οΈ Custom nodes are used, including Control Net, Comfy Manager, Ultimate SD Upscaler, and a resolution calculator.
  • 🧩 The Ultimate SD Upscaler breaks down the image into tiles, which is a key part of the upscaling process.
  • πŸ“ The recommended resolution calculator is crucial for matching SD Excel's recommended resolutions.
  • πŸ“ˆ Scott demonstrates how to use the calculator to upscale images by a factor of four, without manually typing in numbers.
  • πŸ” The process involves multiplying the image's dimensions by the scale factor and using the nearest SD Excel resolution.
  • πŸ”§ There's an option for reverse upscaling, which can downscale an image after upscaling to achieve a specific size.
  • ⏱️ The upscaling process is time-consuming but results in high-quality images.
  • πŸ“ Scott provides instructions on how to save the upscaled image and offers the graph to his supporters for download.

Q & A

  • What is the main topic of the video?

    -The video is about an upscale technique for getting large images using SD Excel, which is trained in certain resolutions.

  • Who is the presenter of the video?

    -Scott Weller, the QA guy for stability.ai, is the presenter.

  • What is the purpose of using a calculator in the upscaling process?

    -The calculator is used to determine the optimal size for the image to match the resolutions that SD Excel is trained in.

  • What are the custom nodes mentioned in the video?

    -The custom nodes include the control net, comfy manager, ultimate SD upscaler, and a recommended resolution calculator.

  • How does the Ultimate SD Upscaler work?

    -It breaks the image down into tiles and upscales each tile, adding details to the image.

  • What is the recommended tile size for SD Excel?

    -The recommended tile size for SD Excel is 1024 by 1024.

  • How does the upscaler handle the addition of details?

    -The upscaler allows for a certain percentage of the upscaled image to change, with a setting of 0.25 meaning only 25% of the upscaled image is allowed to change to add detail.

  • What is the purpose of the recommended resolution calculator?

    -The calculator ensures that the upscaled image matches the recommended resolutions for SD Excel, aiding in achieving the desired upscaled size.

  • What is the scale factor used in the example?

    -In the example provided, the scale factor used is 4, meaning the image is multiplied by 4 to upscale it.

  • How does the process handle different image sizes?

    -The process automatically calculates the necessary scale factor based on the image's dimensions and the desired upscaled size.

  • What is the advantage of using the nearest SD XL resolution from Company Math?

    -Using the nearest SD XL resolution ensures the initial image resolution is suitable for SD Excel before it is passed on to the recommended resolution calculator for further processing.

  • How can viewers access the graph used in the video?

    -The graph will be available in the private area for anyone with a sponsor level or higher on the YouTube channel. Otherwise, viewers can build the graph themselves following the instructions in the video.

Outlines

00:00

πŸ“ˆ Upscaling Images with SD Excel: Introduction and Technique Overview

Scott Weller, the QA guy for stability.ai, introduces his upscaling technique for generating large images. He mentions a previous video where he shared a valid method, which he often combines with the new approach. The goal is to create a process that can upscale any image without manual input. Scott emphasizes the importance of matching the resolutions that SD Excel was trained on. He guides viewers through setting up custom nodes, including Control Net, Comfy Manager, Ultimate SD Upscaler, and a Recommended Resolution Calculator, which is crucial for ensuring the upscaled images meet the required resolutions. The video demonstrates how to load an image and a checkpoint, and how to use the Ultimate SD Upscaler to break the image into tiles for processing. Scott also discusses the importance of the tile size, which should align with what SD Excel expects, and how to adjust the denoising parameter to add detail without creating random squares.

05:01

πŸ” Scaling Up Images with a Resolution Calculator and Model Guided by Prompts

The second paragraph delves into the specifics of using a resolution calculator to upscale images. Scott explains the process of using the calculator to determine the optimal size for the upscaled image, ensuring it matches the resolutions that SD Excel is designed for. He demonstrates how to input the image's width and height into the calculator, which then calculates the necessary scale factor to achieve a four times upscale. Scott also discusses the use of a reverse upscale feature, which can adjust the scale factor if a different level of upscaling is desired. The video shows how to connect the calculated desired X and Y values to the upscaler, which then processes the image in 1024 by 1024 tiles. He emphasizes the flexibility of the system to handle different images and resolutions without the need for manual calculations. The paragraph concludes with Scott saving the upscaled image and inviting viewers to share their thoughts in the comments.

Mindmap

Keywords

πŸ’‘Upscale

Upscaling refers to the process of increasing the resolution of an image or video, typically to improve its quality or to make it suitable for larger displays. In the video, Scott Weller discusses an 'upscale technique' for enlarging images without losing detail, which is central to the video's theme of image enhancement.

πŸ’‘Stable Diffusion (SD)

Stable Diffusion is a term used to describe a type of artificial intelligence model that can generate images from textual descriptions. In the context of the video, Scott mentions 'SD Excel', which seems to be a specific model or tool used for image upscaling, indicating the importance of AI in modern image processing.

πŸ’‘Tiles

In the context of image processing, 'tiles' refer to the smaller segments that a larger image is divided into for processing purposes. Scott talks about breaking down the image into 'tiles of 1024 by 1024', which is a technique used to manage the computational load and improve the efficiency of the upscaling process.

πŸ’‘Custom Nodes

Custom nodes are user-defined functions or modules within a software application that can be used to perform specific tasks. Scott mentions using custom nodes such as 'control net', 'comfy manager', and 'ultimate SD upscaler' to facilitate the upscaling process, highlighting the role of customization in achieving desired results.

πŸ’‘Resolution Calculator

A resolution calculator is a tool used to determine the optimal size for images to match specific display or processing requirements. Scott uses a 'recommended resolution calculator' to ensure that the upscaled images match the resolutions that the SD Excel model is trained to handle, emphasizing the importance of correct resolution for image quality.

πŸ’‘Denoise

Denoise refers to the process of reducing noise or random variations in an image that can degrade its quality. Scott discusses setting a denoising level of '0.25', which means that only 25% of the upscaled image is allowed to change, helping to retain detail while removing noise.

πŸ’‘Checkpoint

In the context of machine learning, a checkpoint is a saved state of a model's training progress. Scott talks about loading in a 'checkpoint', which is a step in preparing the AI model for the upscaling task, indicating the iterative nature of working with AI models.

πŸ’‘Upscale Factor

The upscale factor is the multiplier used to increase the size of an image. Scott mentions using a scale factor of '4', which means the original image dimensions are multiplied by four to achieve the desired upscaled size. This term is crucial as it determines the final dimensions of the upscaled image.

πŸ’‘Image Quality

Image quality refers to the clarity, sharpness, and overall visual appeal of an image. The entire video is focused on improving image quality through upscaling, with Scott demonstrating techniques to achieve 'really big images' without compromising on quality.

πŸ’‘Graph

In the context of the video, a 'graph' likely refers to a visual representation or a workflow diagram that outlines the steps of the upscaling process. Scott mentions setting up a graph where any image can be dropped in and automatically upscaled, indicating the goal of creating a streamlined, user-friendly process.

πŸ’‘Mid-Journey

Mid-Journey is not explicitly defined in the transcript, but it could refer to a stage in the image creation or upscaling process. Since it's mentioned in conjunction with 'Stable Diffusion' and 'photo', it might imply a midpoint in the transformation of an image, either during generation or enhancement.

Highlights

Scott Weller, the QA guy for stability.ai, introduces an upscale technique for creating large images.

The technique involves using a calculator to determine the optimal size for SD Excel, which is trained in specific resolutions.

Custom nodes are used to manage the process, including the control net, comfy manager, and ultimate SD upscaler.

The ultimate SD upscaler breaks the image down into tiles, which is crucial for the upscaling process.

The recommended resolution calculator is key to ensuring the image matches the recommended resolutions for SD Excel.

An image is loaded and a checkpoint is introduced to start the upscaling process.

The upscaler requires an image, a model, and a VAE, with options to add detail and guide the upscale.

The process involves setting the tile size to 1024x1024, which is expected by SD Excel.

Denoising is adjusted to ensure only a portion of the upscaled image is allowed to change, preserving detail.

A calculator is used to automate the scaling process, multiplying the image size by a factor of four.

The upscale model is loaded to apply texture and detail to the upscaled image.

The process is automated to upscale any image by a factor of four without manual input of numbers.

The reverse upscale feature allows for adjusting the scale of the image after initial upscaling.

The width and height of the image are automatically calculated and upscaled by the desired factor.

The process is designed to work with any image from various sources, including Stable Diffusion, MidJourney, or photographs.

The upscaled image is created in 1024x1024 tiles, with the number of tiles depending on the input image.

The process is time-consuming but results in a high-quality upscaled image.

The upscaled image is unguided by prompts, relying on the SD Excel model and the amount of denoise applied.

The graph for the upscaling process will be made available to sponsors on the private area of the YouTube channel.

Viewers are encouraged to build their own graph and save it for reuse by reloading the image.