ComfyUI : Ultimate Upscaler - Upscale any image from Stable Diffusion, MidJourney, or photo!
TLDRIn this tutorial, Scott Weller from stability.ai introduces an advanced technique for upscaling images using a custom graph in SD Excel. He explains how to automate the upscaling process without manually entering numbers, utilizing a calculator to determine optimal resolutions. The video details the setup of custom nodes, including the 'Ultimate SD Upscaler', and how they work together to efficiently upscale images while adding detailed textures. Scott emphasizes the utility of the resolution calculator for seamless integration with SD Excel, ensuring precise adjustments for any image, regardless of its initial size.
Takeaways
- 📈 Scott Weller, the QA guy for stability.ai, introduces an upscale technique for enlarging images without manual input of numbers.
- 🧩 He utilizes a combination of custom nodes, including Control Net, Comfy Manager, Ultimate SD Upscaler, and a recommended resolution calculator.
- 🔍 The upscale process involves breaking down the image into tiles, specifically 1024x1024 tiles, which is the expected size for SD Excel.
- 📐 A resolution calculator is key to determining the optimal size for upscaling, ensuring it matches the trained resolutions of the model.
- 🖼️ The upscaler requires an image, a checkpoint, and can use a positive or negative model for additional guidance.
- 🔄 The upscaler can also perform a reverse upscale, adjusting the scale if the desired upscale factor is different from the model's default.
- 🔍 The process includes a denoising step, which can be adjusted based on the desired level of detail preservation.
- ⚙️ Scott demonstrates how to use the nodes and calculator to automate the upscale process, avoiding manual entry of scale factors.
- 📉 The script mentions the use of Comfy Math for mathematical operations, which is not built into the system by default.
- ⏱️ Upscaling is a time-consuming process, but Scott emphasizes the value of the wait for high-quality results.
- 🔗 The final upscaled image can be saved and reloaded as needed, offering flexibility in the workflow.
- 📌 The video concludes with an invitation for viewers to share their thoughts and support the channel, with a mention of a private area for sponsors to access the graph.
Q & A
What is the main purpose of the video?
-The main purpose of the video is to demonstrate an upscale technique for getting large images using SD Excel, which has been trained in certain resolutions.
Who is the presenter of the video?
-The presenter of the video is Scott Weller, the QA guy for stability.ai.
What is the significance 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 has been trained in, ensuring the best upscaling results.
What are the custom nodes required for the upscaling process?
-The custom nodes required include the control net, comfy manager, ultimate SD upscaler, comfy math, and a recommended resolution calculator.
How does the Ultimate SD Upscaler work?
-The Ultimate SD Upscaler 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 pixels.
How does the upscaler handle the addition of detail to the image?
-The upscaler allows for a certain percentage of the upscaled image to change, which is set to 25% in the example, to add detail without creating random squares of content.
What is the role of the 'nearest sdxl resolution' in the upscaling process?
-The 'nearest sdxl resolution' ensures that the image starts at the correct resolution that SD Excel can work with most effectively before the upscaling process begins.
What is the purpose of the upscale factor in the calculator?
-The upscale factor is used to determine the scale at which the image should be upscaled. It can be adjusted based on the desired outcome, such as a four times upscale.
How does the process handle images of different sizes?
-The process automatically calculates the appropriate scale factor for any given image size, ensuring that the image is upscaled correctly regardless of its initial dimensions.
What is the expected outcome of the upscaling process?
-The expected outcome is an upscaled image that is four times larger and has added detail, without the need for manual input of numbers or resolution adjustments.
How can viewers access the graph used in the video?
-Viewers who are at the sponsor level or higher on the YouTube channel can access the graph in the private area. Others can build the graph themselves following the instructions in the video.
Outlines
😀 Introduction to Upscaling Images with SD Excel
Scott Weller, the QA guy for stability.ai, introduces his technique for upscaling images to large sizes using a graph-based approach. He mentions a previous video where a valid technique was shown but prefers a more automated process. The focus is on using a calculator to determine the optimal size for images to match the training resolutions of SD Excel. Scott outlines the need for custom nodes, including ControlNet, Comfy Manager, and the Ultimate SD Upscaler, which breaks images into tiles for processing. He emphasizes the importance of using a recommended resolution calculator to ensure compatibility with SD Excel's expected resolutions.
🔍 Detailed Walkthrough of the Image Upscaling Process
The video script continues with Scott demonstrating the image upscaling process step by step. He guides viewers on how to load an image and a checkpoint into the system. Scott explains the requirements for the Ultimate SD Upscaler, including the need for an image, a positive or negative value, and a VAE. He discusses creating blank values for the positive and negative inputs and the option to add additional detail if desired. The script also covers setting the denoising level and tile size, which is crucial for SD Excel's expected output. Scott then introduces the use of a resolution calculator to determine the target size for upscaling, emphasizing the need to multiply the image size by a scale factor. He concludes by showing the final upscaled image, which is achieved without manual input of numbers, and invites viewers to share their thoughts in the comments.
Mindmap
Keywords
💡Upscale
💡Stable Diffusion (SD)
💡Tiles
💡ComfyUI
💡Denoising
💡Resolution Calculator
💡Checkpoint
💡VAE
💡Control Net
💡Ultimate SD Upscaler
💡Mathematical Operations
Highlights
Scott Weller, QA guy for stability.ai, demonstrates an upscale technique for creating large images.
A calculator is used to determine the optimal size for SD Excel, which is trained in certain resolutions.
Custom nodes are introduced for the upscaling process, including the control net, comfy manager, and ultimate SD upscaler.
The ultimate SD upscaler breaks the image down into tiles, utilizing comfy math for mathematical operations.
Recommended resolution calculator ensures images match the SD Excel recommended resolutions.
Images are upscaled without the need to manually type in numbers or figure out settings.
The process involves loading a checkpoint and an upscaler to see its requirements.
Positive and negative values are created to guide the upscale, depending on the amount of denoise.
Images are broken down into 1024x1024 tiles, which is the expected tile size for SD Excel.
A denoising level of 0.25 is set, allowing only 25% of the upscaled content to change and add detail.
A desired X and Y target is calculated using the recommended resolution calculator for upscaling by a factor of four.
The upscale model adds skin texture and is guided by user prompts and the amount of denoise.
The process is automated to upscale any image dropped into the system by a scale factor of four.
The upscale factor can be adjusted for different upscaling needs, such as 1.5 or 1.7 times.
The system handles the resolution calculations, making it easy to upscale images without manual input.
The upscale process may take significant time due to the image size and the tiling process.
The final upscaled image is unguided by a prompt, relying on the SD Excel model and denoise levels.
The process is applicable to any image type, from Stable Diffusion to photographs.
The upscaled image is created in 1024x1024 tiles, with the system automatically determining the number of tiles needed.
The graph used in the process will be available for download to patrons at a higher赞助 level.