ULTRA SHARP Upscale! - Don't miss this Method!!! / A1111 - NEW Model

Olivio Sarikas
23 Mar 202310:22

TLDRThe video script outlines a method for achieving ultra-sharp upscales in image rendering. It instructs viewers to download a specific model, 'four times Ultra sharp', and use it in conjunction with the high-res fix feature in an application, setting the denoise strength to 0.5. The process involves a two-step upscaling approach, first using the high-res fix to render a high-resolution version, followed by further upscaling using the 'Ultra sharp' model. The script emphasizes the importance of detail preservation and enhancement, comparing the effectiveness of different upscaling methods and sampling techniques to achieve the best results.

Takeaways

  • 📂 Download the 'four times Ultra sharp' model and place it in the 'automatic 1111' folder within the 'models' folder in the ESR folder.
  • 🖼️ When rendering an image, use the 'hi-res fix' option and set the denoise strength to 0.5 for a high-resolution upscale.
  • 🔄 After obtaining the high-res version, use the 'send to extras' feature and select 'Ultra sharp' with a 2x upscale to achieve a final 4x upscaled image.
  • 🔍 The 'latent image' concept allows for more detailed upscales by working with the underlying data before the image is fully rendered.
  • 🚀 High-res fix is more resource-intensive but ensures the image is rendered in high detail before upscaling.
  • 💡 An alternative method is to turn off high-res fix, render a low-resolution image until satisfied, then upscale using 'image to image' for efficiency.
  • 🔢 Calculate the double size of the image dimensions for the 'image to image' upscale process to ensure correct scaling.
  • 🎨 Experiment with denoise strength to find the balance between image quality and added details.
  • 🌟 The 'Ultra sharp' model adds texture and finer details compared to the 'ESR again' model, resulting in more coherent and sharper images.
  • 🔧 Changing the sampling method during 'image to image' upscaling can affect the sharpness and detail of the final output, with Euler, DPM++ 2m Keras, and DPM-22sde Keras offering varying results.
  • 💻 For facial features, use a lower denoise strength (e.g., 0.25) to maintain similarity to the original while still enhancing sharpness and detail.

Q & A

  • What is the primary method described in the transcript for achieving ultra-sharp upscales?

    -The primary method involves downloading a model called 'four times Ultra sharp' and placing it into the models folder. Then, instead of rendering an image at normal resolution, the image is rendered with a 'hi-res fix' at two times upscale and denoise strength set to 0.5. The high-resolution version is then sent to extras and upscaled using the 'Ultra sharp' model set to two times.

  • Why is the 'hi-res fix' used before upscaling?

    -The 'hi-res fix' is used to first render the image and then upscale it before it becomes an actual image. This process allows for more detail to be added to the upscaled image, resulting in higher quality and sharper results.

  • What is the benefit of using the 'image to image' upscaling method over the 'high-res fix' during the initial search for a good result?

    -The 'image to image' upscaling method is more economical and saves on GPU time during the initial search for a good result. It allows for faster rendering without the need to upscale every single image with the 'high-res fix' beforehand.

  • How does the 'latent image' concept contribute to the quality of upscaled images?

    -The 'latent image' concept refers to the potential detail within an image before it is fully rendered. By upscaling the image based on this latent data, the AI can add more detail to the image, resulting in a higher resolution and more detailed final product.

  • What are the differences in quality between the 'high-res upscaled' and 'image to image upscaled' models as demonstrated in the transcript?

    -Both the 'high-res upscaled' and 'image to image upscaled' models result in identical quality. They both produce highly detailed, sharp images with improved quality and expression compared to a standard low-resolution render followed by a four-time upscale using the ESR model.

  • Why is the 'Ultra sharp' model preferred over the 'ESR again' model in the context of upscaling?

    -The 'Ultra sharp' model is preferred because it adds more texture and finer details to the image, especially at the ends of hair and in areas like the skin texture and clothing. It provides a more coherent and sharper result, making the final image appear more real and detailed.

  • What is the significance of the sampling method in 'image to image' upscaling?

    -The sampling method in 'image to image' upscaling affects the quality and sharpness of the final image. Different methods like Euler, DPM plus plus 2m Keras, and dpm-22sde Keras can produce varying levels of detail and sharpness, allowing users to experiment and find the best match for their desired outcome.

  • How does adjusting the denoise strength value affect the image during upscaling?

    -Adjusting the denoise strength value allows users to balance the sharpness and detail of the image with the original appearance. A lower value like 0.25 can be used to stick closely to the original image while still allowing the AI to add new details for a sharper and more textured result.

  • What is the main advantage of using the 'Ultra sharp' model in conjunction with different sampling methods?

    -Using the 'Ultra sharp' model with different sampling methods allows for fine-tuning of the image's sharpness and detail. It can enhance the quality of specific features like facial hair, skin texture, and clothing details, making the final upscaled image more realistic and visually appealing.

  • How does the quality of an image upscaled with the 'ESR again four times upscaler' compare to one upscaled using the described method?

    -An image upscaled with the 'ESR again four times upscaler' tends to have lower quality, with less detail and sharpness. It often appears more like a digital drawing, with blurry and untextured areas, especially in comparison to an image upscaled using the described method, which retains more detail and sharpness.

  • What is the final result of using the 'Ultra sharp' model and the 'image to image' upscaling method with a low denoise strength and a specific sampling method?

    -The final result is an image with significantly improved quality, sharpness, and detail. It retains the facial features closely to the original while adding new details and textures, resulting in a more realistic and visually appealing upscaled image.

Outlines

00:00

🎨 Ultra Sharp Upscaling Method

This paragraph introduces a method for achieving ultra sharp upscales in images. It begins by instructing the viewer to download a specific model called 'four times Ultra sharp' and place it in the appropriate folder. The process involves rendering an image with a high-resolution fix and a denoise strength of 0.5, then using the 'hi-res fix' option and selecting 'ultra sharp' from the app scalers. The result is a four times upscaled image that is high resolution and very detailed. The paragraph also delves into the concept of a latent image and explains why upscaling before the image is fully rendered can yield better results. It contrasts this method with a more economical approach that involves rendering an image normally, then upscaling it using the 'image to image' option with a denoise strength of 0.5. The summary highlights the importance of experimenting with denoise strength to achieve the best results and the benefits of using the ultra sharp model over the ESR model.

05:01

🔍 Comparison of Upscaling Techniques

This paragraph compares the quality of images upscaled using different methods. It points out the deficiencies of a simple upscale, such as the lack of detail and the blurry appearance of features like eyelashes, pupils, irises, hair, and clothing. The speaker then demonstrates the superior quality of images upscaled using their method, which retains more detail and sharpness. A direct comparison is made between the 'ESR again' and 'ultra sharp' models, showing that the latter provides more coherent and detailed results, especially in areas like the beard, skin texture, and clothing. The paragraph concludes with a discussion on the importance of choosing the right sampling method for image to image upscaling, and how tweaking the denoise strength can lead to sharper images with more textures and details. The speaker suggests experimenting with different sampling methods like Euler, DPM plus plus 2m Keras, and dpm-22sde Keras to find the best result.

10:03

👋 Conclusion and Engagement

In the final paragraph, the speaker thanks the viewers for watching and encourages them to like the video if they enjoyed it. They also invite the viewers to explore other content and express hope to see them again in future videos. The paragraph serves as a closing remark, aiming to engage the audience and promote further interaction.

Mindmap

Keywords

💡Ultra sharp upscales

The term 'Ultra sharp upscales' refers to a method of significantly increasing the resolution of an image while maintaining or enhancing its sharpness and detail. In the context of the video, this process involves using a specific model to upscale an image four times, resulting in a high-resolution output with improved clarity and detail. The method is showcased as a way to achieve better visual quality compared to standard upscaling techniques.

💡High-res fix

The 'High-res fix' is a feature in the upscaling process that allows for the rendering of an image at a higher resolution before it is actually created. This method is intended to improve the quality of the final upscaled image by working with the latent data within the image to add more detail. It is more resource-intensive on the rendering side but yields higher quality results.

💡Latent image

A 'latent image' refers to the underlying data or potential within an image that is not yet fully realized. In the context of the video, it is the undeveloped or unseen details that can be enhanced during the upscaling process. The term is used to describe the information that the AI can tap into when upscaling an image, allowing for the addition of more detail and a higher quality final product.

💡Denoising

Denoising is the process of reducing or removing noise from an image, which can be visual artifacts or distortions that degrade the quality of the image. In the context of the video, denoising strength is set to 0.5 during the upscaling process to achieve a balance between preserving details and reducing noise, resulting in a clearer and sharper image.

💡Image to image upscaling

Image to image upscaling is a method where a low-resolution image is directly upscaled to a higher resolution without first rendering it at a higher resolution. This approach is more resource-efficient, as it saves on GPU time, and allows for faster rendering while searching for a desirable result. It involves determining the double size of the original image and then applying the upscaling and denoising settings to achieve a higher quality output.

💡ESR (Enhanced Super Resolution)

ESR, or Enhanced Super Resolution, is a term used to describe a suite of models or algorithms designed to improve the resolution of images. In the context of the video, ESR is used as part of the upscaling process, where the 'ESR again' folder contains models that are used to further upscale the image after an initial upscaling has been performed.

💡Sampling method

The sampling method refers to the technique used to select data points from a larger dataset or, in the context of the video, the approach used to process and generate an image. Different sampling methods can result in varying levels of detail and sharpness in the final image. The video discusses experimenting with different sampling methods, such as Euler and DPM, to find the one that best preserves facial features and provides the desired level of detail.

💡Texture

Texture in the context of the video refers to the visual detail that gives an image a sense of depth and material quality. It is the detail that makes the image appear more realistic, especially in elements like clothing, skin, and hair. The video emphasizes the importance of maintaining and enhancing texture through the upscaling process to achieve a more realistic and high-quality final image.

💡Sharpness

Sharpness is a term used to describe the clarity and definition of details in an image. A sharp image has well-defined edges and fine details, whereas a less sharp image may appear blurry or indistinct. In the video, the goal is to achieve ultra sharp upscales, which means significantly increasing the image resolution while maintaining or improving sharpness to create a detailed and clear final image.

💡Upscaling

Upscaling is the process of increasing the resolution of an image or video. This is often done to improve the quality of the content when displayed on larger screens or for higher resolution outputs. In the context of the video, upscaling is central to the content, with the focus on achieving ultra sharp results through specific models and techniques.

Highlights

The method for achieving ultra-sharp upscales involves downloading a specific model called 'four times Ultra sharp' and placing it in the appropriate folder.

To render an image with high resolution, use the 'hi-res fix' option and set the denoise strength to 0.5.

After rendering a high-res version, send the image to 'extras' and select 'Ultra sharp' from the app scalers, setting the upscale to two times.

The process results in a four times upscaled image that is high resolution, super sharp, and detailed.

Latent image upscale allows for more detail by working with the latent data before the image is fully rendered.

An alternative, more economical method involves turning off 'high-res fix', rendering an image, and then upscaling it using the 'image to image' option.

Experiment with the denoise strength value to achieve a balance between image quality and detail.

The 'Ultra sharp' model is preferred over the 'ESR again' model for its coherence, finer details, and more realistic textures.

When upscaling images, the 'Ultra sharp' model adds texture and sharpness, enhancing the quality of the final output.

Sampling methods can be adjusted for different results, with Euler providing a softer result and DPM plus plus 2m Keras offering a more detailed and sharper outcome.

The 'sde' method within the DPM plus plus 2m Keras sampling provides slightly sharper details, especially at the edges of hair.

Upscaling with the 'Ultra sharp' model maintains texture and detail across the entire image, including clothing.

The 'image to image' upscaling approach is faster for initial searches, saving on GPU time before the desired result is found.

A direct comparison between low-resolution upscaled images and the method described reveals a significant difference in quality and detail.

The 'Ultra sharp' model consistently brings out sharper and finer details in upscaled images.

The video provides a comprehensive guide on achieving high-quality, sharp upscales through a step-by-step process.

The presenter shares tips on how to optimize the settings for different types of images, such as facial features and textures.

The method can be applied to various types of images, including portraits and detailed artwork.