BeautifAI Image Upscaler! ft. MidJourneyMan & // Civitai Guest Stream

8 Apr 202467:31

TLDRIn this engaging Twitch stream, Tyler hosts a guest creator session featuring Julian (MidJourneyMan) and Brad (, where they discuss and demonstrate an open-source AI image upscaling workflow. The conversation covers the process of enhancing and upscaling images using various tools and models, emphasizing the importance of good prompts and clean input images for optimal results. They showcase the workflow's ability to compete with paid upscaling tools, offering users the freedom to experiment with different settings and nodes to achieve desired results. The stream also provides practical tips for refining images, including the use of custom nodes and the significance of random seeds in the upscaling process. The presenters encourage the audience to try the workflow, share their creations, and engage with the community for further improvements.


  • ๐ŸŽจ The stream features a collaboration between Tyler, MidJourneyMan, and to introduce an open-source image upscaling tool that can compete with paid alternatives.
  • ๐Ÿš€ MidJourneyMan and have been working on this project for over a month, iterating through 12 to 13 versions to refine the workflow.
  • ๐Ÿค– The upscaling tool is designed to enhance image details and control the level of changes through a 'freedom' parameter, allowing users to decide how much new detail to introduce.
  • ๐Ÿงฉ The workflow incorporates various nodes and is built to be modular, enabling users to experiment with different settings and combinations for their specific needs.
  • ๐Ÿ“ˆ The upscaling process involves two passes, with the first pass typically yielding a 2K image and the second pass further enhancing to a 3K to 4K image.
  • ๐Ÿ’พ The tool is resource-intensive, with VRM usage peaking at around 20GB during the stream, suggesting the need for a powerful GPU for optimal performance.
  • ๐Ÿ” Users can adjust the level of detail preservation versus enhancement through settings like the 'freedom' value, with higher values leading to more significant changes and potential artifacts.
  • ๐Ÿ–ผ๏ธ The workflow includes options for post-upscaling adjustments such as color matching, HDR effects, sharpening, and grain addition to fine-tune the final image.
  • ๐Ÿ”ง The presenters encourage users to experiment with different SDXL models and Control Nets to achieve desired styles and avoid artifacts like extra faces or eyes in the upscaled images.
  • ๐ŸŒ The workflow is shared openly on Civitai for anyone to download and use, promoting a collaborative approach to image upscaling within the community.
  • โ“ The stream ends with an invitation for viewers to ask questions and a reminder that experimenting with the tool will yield varying results based on the input image and settings chosen.

Q & A

  • What is the main topic of the stream?

    -The main topic of the stream is about an AI image upscaler tool developed by MidJourneyMan and, which they discuss and demonstrate during the Civitai Guest Stream.

  • Who are the special guests featured in the stream?

    -The special guests featured in the stream are Julian, also known as MidJourneyMan, and Brad, also known as SuperBeast.

  • What is the purpose of the 'freedom' parameter in the upscaling tool?

    -The 'freedom' parameter allows users to control the level of change and enhancement applied to the image during the upscaling process, enabling more or less detail to be added based on the user's preference.

  • How long have Julian and Brad been working on the upscaling tool?

    -Julian and Brad have been working on the upscaling tool for over a month, with around 12 or 13 versions and iterations developed during that time.

  • What is the significance of the 'seam fix mode' in the upscaling process?

    -The 'seam fix mode' helps to denoise the edges of the tile when it is put back together, which can result in a cleaner and more polished image. It can be switched to 'none' for faster testing of prompts.

  • Why might an artist choose to use a lower 'freedom' value for upscaling?

    -An artist might choose to use a lower 'freedom' value to minimize changes to the original image, retain more of its original details, and avoid introducing artifacts that they may not want in the upscaled result.

  • How does the upscaling tool handle multiple passes?

    -The upscaling tool performs two passes. The first pass upscales the image and introduces some enhancements, while the second pass further increases the resolution and refines the image based on the set parameters and prompts.

  • What is the role of the 'control nets' in the upscaling workflow?

    -The 'control nets', including canny and depth edges, are used to maintain the structure and details of the image during the upscaling process, helping to prevent the introduction of unwanted artifacts.

  • How can users customize the upscaling process to their own style?

    -Users can customize the upscaling process by adjusting the 'freedom' parameter, using different SDXL models, experimenting with various luras, and tweaking the prompts to guide the style and outcome of the upscaled image.

  • What are the system requirements for running the upscaling tool?

    -The upscaling tool can run on systems with at least 8GB of VRAM, but for better performance and to handle larger image sizes, a system with more VRAM, such as 20GB or higher, is recommended.

  • How can users share their results and get feedback on the upscaling tool?

    -Users are encouraged to post their upscaled images on social media, tagging MidJourneyMan and SuperBeast to share their results and receive feedback or reshares from the creators of the tool.



๐Ÿ˜€ Introduction and Upscaling Tool Presentation

The host, Tyler, welcomes viewers to the Friday guest Creator stream, expressing excitement about the special guests, Julian (mid journeyman) and Brad (super beast). They discuss their work on an open-source image upscaling tool that rivals paid versions, sharing their experiences and the process they've developed. Julian and Brad introduce themselves, their backgrounds, and their focus on AI art and upscaling. They provide information on where to find the workflow on civii and guide users through downloading and using it.


๐Ÿ” Workflow Overview and Upscaling Process

The paragraph provides a detailed walkthrough of the upscaling workflow. It covers the process of extracting files from a zip, using Confy UI, and the significance of the Json and PNG files. The host explains the use of the workflow, the importance of the 'freedom' parameter, and the tiling upscaling technique. The paragraph also touches on the system's VRAM usage and the practical aspects of running the workflow, including the time it takes for different passes and the recommended system specifications.


๐Ÿ–ผ๏ธ Before and After Image Comparison and Editing Tips

The host demonstrates the upscaling process by showing before and after comparisons using a slider tool. They highlight the removal of noise and pixelization, and the addition of details to the upscaled image. The paragraph also discusses the use of color matching, filters, and sharpening within the workflow. It advises on potential issues like artifacting and the need for editing certain elements of the image post-upscaling. The importance of experimenting with different SDXL models and the role of prompts in achieving desired styles are also covered.


๐Ÿ”ง Customization and Control Over Upscaling

The paragraph delves into the customization options available in the upscaling tool. It discusses the heart of the workflow, the control nets, and how they affect the upscaling process. The host explains the impact of the 'freedom' setting on the image outcome, ranging from minimal changes to significant detail enhancements. The paragraph also covers the upscaling factors for the first and second passes and the importance of allowing new details to emerge through scaling down and up techniques.


๐Ÿ“ˆ Freedom Values and Their Impact on Image Upscaling

The host illustrates the impact of different freedom values on the upscaling process. They showcase examples with varying levels of freedom, from minimal to maximal changes, and discuss the corresponding effects on the image. The paragraph emphasizes the balance between upscaling and maintaining the original image's integrity. It also mentions the potential for artifacts at higher freedom values and the need for post-processing to clean up images.


๐ŸŽจ Artistic Preferences and Model Selection

The paragraph focuses on artistic preferences in upscaling and the selection of models to achieve desired outcomes. It discusses the use of different upscale models and the importance of testing to find the best fit for various image styles. The host shares their personal preferences for certain models and the benefits of using pass one for most upscaling needs. The paragraph also touches on the use of seeds in upscaling and the advantages of randomization.


๐Ÿ› ๏ธ Control Net Settings and Image Enhancement

The paragraph discusses the control net settings within the workflow and their role in image enhancement. It covers the use of canny and depth nodes to maintain image structure during the upscaling process. The host explains how to add preview images for better control over these settings. The paragraph also explores the HDR effect for boosting image vibrancy and the importance of adjusting highlight and shadow intensities for optimal results.


๐Ÿ“š Final Tips and Stream Conclusion

The host provides final tips on using the upscaling workflow effectively. They emphasize the importance of good input (clean images, proper prompts, and suitable models) for achieving quality results. The paragraph concludes with a Q&A session, addressing questions about VRM usage for 4K upscaling and the potential conversion of the workflow for different versions of SD. The host thanks the guests, Julian and Brad, for their contributions and guides viewers on where to find more information and resources.


๐Ÿ“… Upcoming Streams and Social Media Engagement

The host outlines the schedule for upcoming streams, including a session with Sir Spence, a technical researcher and artist. They encourage viewers to follow them on social media and to tag the guests when posting creations made using the shared workflow. The host thanks the audience for joining the stream and provides information on where to find the video and previous streams on YouTube.



๐Ÿ’กImage Upscaler

An image upscaler is a software tool or process that increases the resolution of an image while attempting to maintain or improve its quality. In the context of the video, the upscaler is a significant part of the discussion as the creators have developed a workflow that can enhance images to compete with paid upscaling tools, which is a major theme of the video.


MidJourneyMan, whose real name is Julian, is one of the special guests in the video and a good friend of the host. He is known for his work with AI art and has transitioned from using MidJourney to focusing on stable diffusion, which is a type of AI model for generating images. His expertise and contribution to the development of the upscaling tool is central to the video's content.

๐Ÿ’ก, run by Brad, is another special guest featured in the video. Brad is known for his work with anthro beastly art and his exploration with various AI tools. His collaboration with Julian on the upscaling workflow is a key element of the video, showcasing the synergy between different artists in the AI art community.


In the video, the term 'workflow' refers to the step-by-step process that the guests have developed for upscaling images. It involves using various nodes and parameters within a software called Comfy UI to achieve the desired results. The workflow is a core concept in the video as it is what the creators are sharing with the audience.

๐Ÿ’กComfy UI

Comfy UI is a user interface for interacting with AI models like Stable Diffusion. It is mentioned multiple times in the video as the platform where the upscaling workflow is built and executed. Comfy UI allows for customization and control over the image upscaling process, which is essential for the creators' demonstration.

๐Ÿ’กFreedom Value

The 'Freedom Value' is a parameter within the upscaling workflow that controls the degree to which the AI can alter the original image to add details or enhance the image. A higher freedom value allows for more changes and enhancements, while a lower value results in a more conservative upscale that stays closer to the original image. It is a key concept as it determines the balance between image quality and fidelity to the original.

๐Ÿ’กUpscaling Tools

Upscaling tools are software applications used to increase the size of digital images without a loss in quality, or to improve quality. In the video, the discussion around upscaling tools is focused on the high costs associated with many of them and the open-source nature of the workflow being presented, which is intended to offer a more accessible alternative.

๐Ÿ’กControl Nets

Control Nets are AI models used within the upscaling process to help maintain the structure and details of the original image as it is enlarged. In the video, they are an important part of the workflow as they help prevent the introduction of artifacts and unwanted changes to the image, ensuring the upscaled image retains its integrity.

๐Ÿ’กStable Diffusion

Stable Diffusion is an AI model used for generating images from textual descriptions. It is referenced in the video as the type of AI that Julian has shifted his focus to, indicating a move towards more detail-oriented and customizable image creation. The model's capabilities are central to the upscaling workflow being discussed.


Civitai is a platform for artists and creators to share resources, tools, and artwork related to AI-generated art. In the video, it is the host's platform where the upscaling workflow is being shared with the community. Civitai serves as a hub for collaboration and the dissemination of knowledge within the AI art community.

๐Ÿ’กVRAM Usage

VRAM, or Video RAM, is the memory used by a graphics processing unit (GPU) to store image data for rendering. In the context of the video, VRAM usage is discussed in relation to the demands of running the upscaling workflow, particularly for higher-resolution images. Efficient VRAM usage is crucial for the smooth operation of the workflow on various hardware configurations.


Tyler introduces a special guest stream featuring Julian (MidJourneyMan) and Brad ( to discuss AI image upscaling tools.

Julian and Brad have been working on an open-source workflow to compete with paid image upscaling tools.

The workflow has been in development for over a month with multiple iterations to refine the process.

The upscaling process involves using Stable Diffusion models and custom nodes created by Brad.

The tool allows users to upscale images with different levels of freedom, controlling how much the original image can change.

The workflow is designed to be user-friendly, enabling artists to download, install, and start upscaling with ease.

Julian demonstrates the upscaling process, showing the before and after results with a slider for comparison.

The upscaling tool includes options for color matching, HDR effect, sharpening, and grain addition to enhance image quality.

The workflow is highly customizable, allowing users to adjust settings like the upscaling factor and control net parameters.

Julian and Brad discuss the importance of using different seeds for each pass to prevent detail burn-in and overfitting.

The stream showcases how the upscaling tool can handle various image styles, from anime to photorealistic.

The presenters emphasize the need for experimentation with the tool's settings to achieve the desired results.

The workflow is designed to work with RTX 40 series GPUs, with discussions on VRM usage during the upscaling process.

Julian and Brad share tips and tricks for optimizing the upscaling process, including editing the original image for better results.

The stream includes a Q&A session where the presenters answer questions about the workflow and its application.

The workflow is available for download, and the presenters encourage users to share their results and tag them on social media.

The stream concludes with a teaser for the next week's guest, Spence, a technical researcher and AI artist known for creating the ZeroScope model.