Negative Embeddings - ULTRA QUALITY Trick for A1111

Olivio Sarikas
14 Apr 202306:33

TLDRThis video script offers valuable tips for enhancing AI-generated images. It introduces the concept of negative embeddings, which are trained on undesirable image features to guide the AI towards better results. The script also details a multi-step upscaling technique that involves rendering at a higher resolution, using image-to-image upscaling, and applying photo editing tools like Affinity Photo or Photoshop to sharpen and refine the image details. By combining these methods, users can significantly improve the quality and detail of AI renders, resulting in more lifelike and intricate artwork.


  • 🎨 Utilize negative embeddings to improve AI renders by training on undesirable image features.
  • πŸ“ Add negative prompts to your AI rendering process to avoid undesired outcomes, such as 'bad artist' or 'bad hands'.
  • πŸ” Download and integrate negative embeddings into your AI software's embedding folder for direct application.
  • πŸ–ŒοΈ Use negative embeddings with a weight of 0.8 for effective results in AI rendering.
  • πŸ“± Render an image at a higher resolution (double the original) for better quality using image to image upscaling.
  • 🌐 Apply 'Restore Face' feature to maintain facial details during upscaling, using tools like DPM++ SDE Keras.
  • πŸ”§ Adjust denoise strength to a lower value (e.g., 0.25) to allow AI to introduce new details without losing original texture.
  • πŸ–ΌοΈ Use photo editing software like Affinity Photo or Photoshop to sharpen and enhance the upscaled image.
  • πŸ‘€ Turn off face restore after the initial upscaling to prevent blurring of already optimized facial features.
  • ⚠️ Be cautious of over-sharpening, especially on edges, and adjust sharpening settings or manually correct any issues.
  • πŸ‘ Save the final image in PNG format to preserve quality without JPEG artifacts.

Q & A

  • What is the main topic of the video?

    -The main topic of the video is about improving the quality of AI renders using two specific techniques: negative embeddings and upscaling.

  • What are negative embeddings in the context of AI renders?

    -Negative embeddings are a type of embedding trained on undesirable elements in AI renders, which are then used in negative prompts to guide the AI towards better results.

  • How can negative embeddings be applied to an AI render?

    -Negative embeddings can be applied by downloading the specific embeddings related to undesired features, such as bad artist styles or poor hand renderings, and then including their names in the negative prompt of the AI render.

  • What is the weight applied to negative embeddings in the video?

    -In the video, a weight of 0.8 is often applied to negative embeddings for effective use in AI renders.

  • What is the second trick for enhancing AI render quality discussed in the video?

    -The second trick is related to upscaling the image. It involves rendering a high-resolution image, using image-to-image upscaling with specific settings, and then sharpening the image in photo editing software to bring out more details and textures.

  • What software is recommended for sharpening the upscaled image?

    -The video recommends using photo editing software like Affinity Photo or Photoshop for sharpening the upscaled image.

  • What is the recommended setting for the unsharp mask filter in photo editing software?

    -The recommended setting for the unsharp mask filter is one pixel radius with 0% strength to sharpen the image without introducing JPEG artifacts.

  • Why is face restore turned off in the second part of the upscaling process?

    -Face restore is turned off because the face has already been optimized, and face restore might blur the details that have been sharpened during the photo editing process.

  • What is a potential issue that can arise from sharpening an image too much?

    -Over-sharpening can lead to highlighted edges or artifacts appearing on the image, which can be problematic and require manual adjustment or reduction in sharpening intensity.

  • What format is suggested for saving the final sharpened image?

    -It is suggested to save the final sharpened image in PNG format to avoid any JPEG artifacts that might degrade the image quality.

  • How can viewers support the creator of the video?

    -Viewers can support the creator by liking the video and leaving positive feedback, which helps the content reach a wider audience.



🎨 Enhancing AI Renders with Negative Embeddings

This paragraph introduces viewers to the technique of using negative embeddings to enhance the quality of AI-generated images. The speaker explains that negative embeddings are trained on undesirable image features and can be incorporated into the negative prompt of an AI render. This method can be applied to various aspects of the image, such as the art style and the depiction of hands. The speaker demonstrates how to download and use these embeddings by naming them in the negative prompt without the '.PT' file extension and enclosing them in angle brackets. The effectiveness of this approach is shown by comparing different renders of the same image with various negative embeddings, highlighting the improvement in image quality.


πŸ” Upscaling Images for Higher Quality

The second paragraph discusses the process of upscaling images to achieve better quality. The speaker shares a trick involving the use of an image-to-image upscaling method, which involves rendering an image at a higher resolution and then applying a denoising step with a specific setting (DPM++ SDE Keras with 20 steps and a denoise strength of 0.25). After the initial rendering, the image is copied into photo editing software like Affinity Photo for further enhancement. The speaker emphasizes the importance of sharpening the image using the unsharp mask filter set to one pixel and zero percent to bring out more texture and detail. Finally, the speaker advises viewers to save the image in PNG format to avoid JPEG artifacts and to be cautious of over-sharpening, which can create unwanted bright edges on the image.



πŸ’‘Negative Embeddings

Negative embeddings refer to a machine learning technique where an embedding is trained on undesirable outcomes to guide the AI towards generating better results. In the context of the video, this method is used to improve the quality of AI renders by instructing the AI on what the image should not look like. For instance, the speaker mentions using 'bad minus artist' and 'bad hands' as examples of negative embeddings to avoid undesired art styles and hand renderings.

πŸ’‘AI Renders

AI renders are the outputs generated by artificial intelligence algorithms that create visual content, such as images, based on given inputs or prompts. The video's main theme revolves around enhancing the quality of these renders through specific techniques. AI renders are the central focus, and the video provides tips on how to achieve higher fidelity and more accurate visual representations.


In the context of AI generation, prompts are the input text or instructions given to the AI system to guide the creation of specific outputs. Prompts are crucial in determining the subject, style, and other aspects of the AI renders. The video discusses the use of negative prompts to exclude certain unwanted elements in the AI-generated images.


Upscaling refers to the process of increasing the resolution of an image, typically to enhance its quality and detail. In the video, upscaling is presented as a technique to improve AI renders by doubling the image's resolution and using specific software settings to refine the details and textures.

πŸ’‘Image to Image

Image to image is a method used in AI where one image is used as a reference or input to generate or modify another image. In the context of the video, this technique is employed to upscale the AI renders, aiming to achieve better quality and more intricate details in the final output.

πŸ’‘DPM Plus Plus

DPM Plus Plus is a deep learning model used for image upscaling and enhancement. It is mentioned in the video as the tool used for rendering during the upscaling process. DPM Plus Plus helps in restoring facial features and increasing the image quality by adding details where necessary.

πŸ’‘Denoise Strength

Denoise strength is a parameter used in AI rendering and image processing to control the level of noise reduction or detail preservation. A lower denoise strength allows for more variations and details in the AI-generated image, while a higher value results in a smoother image with less noise. In the video, the speaker sets the denoise strength to 0.25 to maintain a balance between similarity and introducing new details.

πŸ’‘Affinity Photo

Affinity Photo is a photo editing software used for various image manipulation tasks, including sharpening and enhancing image quality. In the video, it is recommended as a tool to sharpen the upscaled AI renders, which helps to bring out more texture and detail in the image.


Sharpening is a photo editing technique that enhances the contrast along the edges and details of an image, making it appear more defined and clear. In the context of the video, sharpening is used to improve the quality of the upscaled AI renders, focusing on bringing out the initial texture and details in the image.


Oversharpening occurs when an image is sharpened to the point where it introduces unnatural or exaggerated edges and details, often resulting in a visually unappealing or distorted image. The video warns against this effect and suggests adjusting the sharpening settings or manually correcting any oversharpened areas to maintain a natural look.

πŸ’‘Face Restore

Face restore is a feature in image processing software that specifically enhances or corrects facial features in an image. In the video, it is initially used during upscaling but is later turned off to prevent blurring and maintain the optimized facial details achieved through photo editing.


Introduction to improving AI-rendered image quality with two tricks.

Explanation of negative embeddings to improve AI renders.

Utilizing negative prompts to correct art styles, prompts, and details like hands.

Demonstration of using negative embeddings for specific issues like bad artists or prompts.

How to download and use negative embeddings in your projects.

Guidance on embedding usage within negative prompts, including weight adjustments.

Comparison of AI-rendered images with different negative embeddings.

Introduction to upscaling trick for enhanced image quality.

Steps for using image to image upscaling for better quality.

Detailing the process of doubling the resolution for upscaling.

Using photo editing software for final quality improvements.

Sharpening the upscaled image using unsharpen mask filter.

Comparison between original and sharpened upscaled images.

Tips on avoiding over-sharpening and preserving image quality.

Conclusion and encouragement to explore these techniques.