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

Olivio Sarikas
23 Mar 202310:22

TLDRThis video provides a method for achieving ultra-sharp upscales using the A1111 model. First, download the 'four times Ultra sharp' model and place it in your ESRGAN folder. When rendering an image, use the 'hi-res fix' and a 2x upscale with a denoise strength of 0.5. After rendering the high-res version, send it to 'extras' and select 'Ultra sharp' for a 2x upscale, resulting in a 4x upscaled, high-resolution image with enhanced details. The video explains the importance of latent image data and the benefits of using the high-res fix for better rendering quality. It also compares the results of using the Ultra sharp model versus the ESRGAN model, highlighting the superior coherence, texture, and detail of the Ultra sharp model. The video concludes with tips on adjusting the denoise strength and sampling method for even better results, showcasing the significant improvement in image quality.

Takeaways

  • 📚 Download the 'Four Times Ultra Sharp' model and place it in the 'Automatic 1111' folder within the 'ESR' folder.
  • 🖼️ Use the 'Hi-Res Fix' feature for rendering images with a two times upscale and set the denoise strength to 0.5.
  • 🔄 After rendering a high-resolution image, send it to 'Extras' and select 'Ultra Sharp' from the app scalers, setting it to two times for the final four times upscale.
  • 🔍 The 'latent image' concept allows for more detailed upscaling by examining the latent data before the image is fully rendered.
  • ⏱️ Turn off 'Hi-Res Fix' to save on GPU time and render images until you find a satisfactory one, then upscale it using 'Image to Image'.
  • 🔢 Determine the double size for your image dimensions for the upscale process, ensuring the height and width are doubled appropriately.
  • 🎛️ Experiment with the denoise strength value; a value of 0.5 can provide a balance between image alteration and detail enhancement.
  • 📈 Upscaling using 'Image to Image' allows for faster rendering when searching for a good result, as it saves time compared to upscaling every image with 'Hi-Res Fix'.
  • 📉 Direct comparison between 'Hi-Res upscaled' and 'Image to Image upscaled' models shows identical quality, validating the effectiveness of both methods.
  • 🚫 Avoid using the standard ESRGAN four times upscaler for low-resolution images as it results in poor quality with lack of detail.
  • ✂️ The 'Ultra Sharp' model provides more coherence, texture, and finer details compared to the ESRGAN model, especially in areas like hair, skin, and clothing.

Q & A

  • What is the name of the model that is recommended for ultra sharp upscales?

    -The recommended model for ultra sharp upscales is called 'four times Ultra sharp'.

  • Where should the 'four times Ultra sharp' model be placed?

    -The model should be placed in the 'automatic 1111' folder, within the 'models' folder in the 'ESR again' folder.

  • What is the denoise strength setting recommended for rendering a high-resolution image?

    -The denoise strength should be set to 0.5 when rendering a high-resolution image.

  • How many times should you upscale the image in the Ultra Sharp app scaler?

    -In the Ultra Sharp app scaler, you should set the upscale to two times because it has already been upscaled two times previously.

  • What is the result of using the described method for upscaling?

    -The result is a four times upscaled image that is high resolution, super sharp, and has a lot of details.

  • What is a latent image and how does it contribute to better upscaling?

    -A latent image refers to the data that is upscaled without the image already being created. It allows for more detail to be added to the upscaled image compared to upscaling a low-resolution image directly.

  • Why is turning off the high-res fix and rendering an image first more economical?

    -Turning off the high-res fix and rendering an image first saves on GPU time because it avoids the need to upscale every single image with high-res fix before finding a satisfactory result.

  • What is the recommended denoise strength when using the image to image upscale method?

    -The recommended denoise strength when using the image to image upscale method is also 0.5, which can be adjusted for better details in the result.

  • How does the Ultra Sharp model differ from the ESR again model in terms of quality?

    -The Ultra Sharp model adds more texture and finer details to the upscaled image, making it more coherent and realistic compared to the ESR again model, which tends to look more like a digital drawing.

  • What is the benefit of using the image to image upscaling method?

    -The benefit of using the image to image upscaling method is that you can render faster while searching for a good result, as it saves time and resources by not upscaling every image with high-res fix initially.

  • How does changing the sampling method affect the final upscaled image?

    -Changing the sampling method can result in a softer or sharper image, depending on the method used. For example, DPM plus plus 2m Keras provides a more detailed and sharper result, which can be beneficial for photography results.

  • What is the final advice given by the speaker regarding the use of the Ultra Sharp model?

    -The speaker advises experimenting with different sampling methods and denoise strength values to achieve the best results when using the Ultra Sharp model for upscaling images.

Outlines

00:00

🖼️ Ultra Sharp Upscaling Technique

The video introduces a method to achieve ultra sharp upscales by downloading a specific model called 'four times Ultra sharp' and placing it in the appropriate folder. The process involves rendering an image at high resolution with a two times upscale and denoise strength of 0.5, then sending it to an app scaler for another two times upscale using the Ultra sharp model. The result is a four times upscaled image with high resolution and sharp details. The explanation delves into the concept of a latent image and why the high-res fix renders the image before upscaling, which is more resource-intensive. An alternative, more efficient method is also discussed, which involves rendering a low-resolution image, finding a satisfactory result, and then upscaling it using image-to-image scaling with a denoise strength of 0.5. The benefits of using the Ultra sharp model over the ESR again model are highlighted through a comparison of upscaled images, emphasizing the superior detail and quality of the former.

05:01

🔍 Comparison of Upscaling Models

The video compares the quality of upscaled images using different models, specifically the Ultra sharp and ESR again four times upscaler. The ESR again model is shown to produce lower quality images with less detail, especially in areas like eyelashes, hair, and clothing. The Ultra sharp model, on the other hand, provides more coherence, better fitting details, and sharper elements. The video demonstrates the effectiveness of the Ultra sharp model through close-up comparisons, showing finer details and textures, especially in areas like hair ends, skin texture, and clothing. The importance of consistency and finer details in upscaling is emphasized. Additionally, the video discusses the use of different sampling methods in image-to-image upscaling, such as Euler and DPM plus plus 2m Keras, which can affect the sharpness and detail of the final image. The presenter suggests experimenting with these settings to achieve the desired level of detail and sharpness.

10:03

📺 Viewer Engagement and Closing Remarks

The video concludes with a call to action for viewers to like the video if they enjoyed it and to explore other content suggested on the end screen. The presenter expresses hope to see the viewers again, reiterating the request for a like before signing off.

Mindmap

Keywords

💡Ultra sharp upscale

Ultra sharp upscale refers to a method of significantly enhancing the resolution and clarity of an image. In the video, this is achieved through a two-step process involving initial rendering at a normal resolution followed by a high-resolution fix and subsequent upscaling using a specific model called 'Ultra sharp'. The result is an image that is four times the original resolution, with greatly improved detail and sharpness.

💡Model (A1111)

In the context of the video, a 'model' refers to a specific algorithm or set of instructions used by an AI or image processing software to perform a task, such as upscaling images. The 'A1111' model mentioned is a tool within the software that aids in achieving ultra sharp upscales by following a particular set of parameters and processes.

💡High-res fix

High-res fix is a feature within the image processing software that allows for the rendering of an image at a higher resolution than the original. This feature is crucial in the video's method as it provides a base for further upscaling to achieve ultra sharp results. It is used after the initial rendering to create a more detailed image before upscaling.

💡Denoise strength

Denoise strength is a parameter that controls the amount of noise reduction applied to an image during the upscaling process. A value of 0.5, as mentioned in the video, is used to balance between reducing noise and preserving image details. It's a critical setting when using the high-res fix feature to ensure the upscaled image retains its sharpness and clarity.

💡Latent image

A latent image refers to the underlying data or representation of an image before it is fully rendered or materialized. In the video, the concept is used to explain why upscaling works better when the high-res fix is applied first; the software can utilize the latent data to add more detail to the upscaled image, resulting in higher quality.

💡Image to image upscale

Image to image upscale is an alternative method mentioned in the video for increasing the resolution of an image. Unlike the high-res fix, which upscales during rendering, image to image upscale takes a rendered image and increases its size, often resulting in a more detailed and sharper final product. This method is more resource-efficient and allows for faster iteration during the creative process.

💡Sampling method

The sampling method is a technique used in the upscaling process to determine how the new pixels are generated to increase the image size. Different sampling methods can produce different results in terms of sharpness and detail. In the video, the presenter experiments with different sampling methods like Euler and DPM++ 2m Keras to achieve varying levels of sharpness and detail in the upscaled images.

💡ESRGAN (Enhanced Super-Resolution Generative Adversarial Network)

ESRGAN is a type of generative adversarial network (GAN) specifically designed for super-resolution tasks, which is the process of increasing the resolution of images. In the video, ESRGAN is compared to the 'Ultra sharp' model, with the latter providing more consistent and detailed results. ESRGAN is used as a reference point to demonstrate the superior performance of the 'Ultra sharp' model.

💡Texture

Texture in the context of the video refers to the fine details and surface qualities of the elements within an image, such as skin, hair, and clothing. The goal of the upscaling process is to preserve or enhance these textures, ensuring that the upscaled image looks natural and detailed. The 'Ultra sharp' model is praised for adding texture and making elements appear more real.

💡Coherence

Coherence in the video refers to the consistency and logical continuity of the details within an upscaled image. When an image is upscaled using the 'Ultra sharp' model, the elements within the image fit together more coherently, with details aligning and making the overall image appear more natural and less like a digital drawing.

💡Sharpness

Sharpness is the clarity and definition of the edges and details within an image. The video emphasizes the importance of maintaining or enhancing sharpness during the upscaling process. The 'Ultra sharp' model is highlighted for its ability to add finer details and improve the sharpness of upscaled images, particularly in areas like hair, skin texture, and clothing details.

Highlights

Download the 'four times Ultra sharp' model and place it in your ESR again folder.

Render images at high resolution with a 2x upscale and denoise strength set to 0.5.

After rendering a high-res image, use the 'send to extras' feature and select 'Ultra sharp' for further upscaling.

The final result is a 4x upscaled image that is high resolution, super sharp, and detailed.

Latent image upscaling allows for more detail by using latent data before the image is fully rendered.

An alternative method is to render an image normally, then upscale using 'image to image' for a more economic approach.

Experiment with denoise strength for better image details and quality.

The 'Ultra sharp' model provides more coherent and detailed upscales compared to the ESR again model.

Upscaling with the 'Ultra sharp' model adds texture and finer details, especially in areas like hair and skin.

The 'image to image' upscaling method allows for faster rendering during the search for a good result.

Direct comparison shows the 'Ultra sharp' model produces higher quality and sharper images than standard upscaling methods.

Different sampling methods like Euler and DPM++ 2m Keras can be experimented with for varying levels of detail and sharpness.

Using a lower denoise strength value with certain sampling methods can help maintain facial features while adding new details.

The 'sde' method provides a slightly sharper result, especially noticeable in details like hair edges.

The 'Ultra sharp' model is more consistent in bringing out finer details across the entire upscaled image.

Clothing textures are more consistently sharp and detailed with the 'Ultra sharp' model compared to other upscaling methods.

The video provides a comprehensive guide on achieving ultra sharp upscales with practical applications and comparisons.