How to improve 3D people in your renders using AI (in Stable Diffusion) - Tutorial

The Digital Bunch
7 Feb 202407:26

TLDRIn this tutorial, the digital bunch demonstrates how to enhance 3D characters in renders using Stable Diffusion, an open-source AI tool. They discuss the importance of staying updated with AI advancements in the creative industry and share their mixed experiences with the technology. The video guides viewers through installing Stable Diffusion, using the web interface, and cropping images for processing. It explains selecting a model specialized in faces and people, crafting prompts, and adjusting settings like resolution and denoising strength for optimal results. The tutorial also covers the process of generating images, selecting the best output, and integrating the improved 3D characters back into the visualization. The creators encourage viewers to share their experiences and suggest further tools and tests for future content.


  • 🎨 **Stable Diffusion Introduction**: Stable Diffusion is an open-source software project for text-to-image models, which can significantly enhance 3D people in renders using AI.
  • 💻 **Installation & Interface**: Before starting, install Stable Diffusion and use the web interface, which, despite initial confusion, offers many features and options.
  • 🖼️ **Image Processing Limitation**: Stable Diffusion doesn't process large images, so users need to crop the area of interest and save it separately before using the tool.
  • 🖌️ **Editing with Brush Tool**: Select the elements to edit within the cropped image using a brush tool within the Stable Diffusion interface.
  • 🧑 **Model Selection**: Choose a model specialized in faces and people, such as 'Realistic Vision', for editing human elements in the image.
  • 🌿 **Alternative Models**: Other models like 'Photon' work well with realistic vegetation and environments, showcasing the versatility of different models.
  • ✍️ **Prompt Crafting**: Use both positive and negative prompts to guide the AI, being as clear and simple as possible to achieve the desired outcome.
  • 📏 **Resolution Settings**: Set the resolution to 768 pixels, which is optimal for the model, to maintain quality and detail.
  • 🔁 **Batch Processing**: Generate a batch of four images to choose from, balancing processing time with the option for selection.
  • 🔍 **Noising Strength**: Adjust the denoising strength between 25 to 45 to control the difference between the original and the new image, aiming for realism.
  • ⏱️ **Processing Time**: Be patient, as generating images can take about a minute on a 4070 TI card, computed locally.
  • 📝 **Post-Processing**: After generation, paste the improved image back into your visualization for a final, more realistic render.

Q & A

  • What is the main focus of the tutorial in the transcript?

    -The tutorial focuses on how to use Stable Diffusion, an open-source software project, to improve 3D people in renders using AI.

  • What type of feedback did the digital bunch receive after sharing their initial tests with Stable Diffusion?

    -They received amazing feedback, with many people asking for a tutorial on how to use the tool.

  • Why is it important to keep an eye on AI developments in creative industries?

    -It's important because AI was not previously thought to impact creative industries, but it's now proving to be a valuable tool for artists, so staying on top of its evolution is crucial.

  • What is the first step in using Stable Diffusion for a project?

    -The first step is to install Stable Diffusion, following the clear instructions provided on their website.

  • Why is it necessary to crop the image before using Stable Diffusion?

    -Stable Diffusion does not process large images yet, so it's necessary to crop the part of the image you're most interested in and save it as a separate file.

  • What is the recommended resolution for Stable Diffusion to work optimally?

    -The optimal resolution for the model to work with is 768 pixels.

  • How many different images does Stable Diffusion generate by default?

    -By default, Stable Diffusion generates four different images to choose from.

  • What is the purpose of the noising strength setting in Stable Diffusion?

    -The noising strength setting determines how different the newly generated image will be from the original image, with higher values resulting in more differences.

  • Why might someone choose to select the face and body separately in Stable Diffusion?

    -Selecting the face and body separately can lead to better results, as it allows for more precise control over the AI's adjustments.

  • What are some of the potential issues with using AI tools like Stable Diffusion?

    -AI tools can sometimes hallucinate or produce weird, creepy, or funny results, especially if the noising value is set too high.

  • How can users share their outcomes and experiences with the digital bunch?

    -Users can share their outcomes and experiences by leaving comments and suggesting what else they'd like the digital bunch to show or test.

  • What is the digital bunch's attitude towards AI in creative projects?

    -The digital bunch is excited about the potential of AI in creative projects and is actively engaged in research and development to explore its capabilities.



🎨 Introduction to Stable Diffusion in Art Projects

The speaker introduces themselves as part of the Digital Bunch and outlines the purpose of the video: to demonstrate the use of stable diffusion in their projects. They mention previous tests with stable diffusion and AI that received positive feedback and express the importance of staying updated with the evolving technology. The video is set to provide a tutorial on stable diffusion, an open-source text-to-image model, and the speaker invites viewers to share their thoughts in the comments section. They also note the need to install stable diffusion and use a web interface for the demonstration, with instructions available on the stable diffusion website.


🖼️ Using Stable Diffusion for Image Editing

The video script explains the process of using stable diffusion for image editing, particularly when dealing with large images that the software cannot process. It details the steps to crop the desired part of an image and save it separately before using stable diffusion. The speaker guides viewers through the web interface, emphasizing the selection of the appropriate model for the task, such as 'Realistic Vision' for faces and people or 'Photon' for realistic vegetation and environments. They also discuss the importance of crafting effective prompts, including both positive and negative prompts, to guide the AI in generating desired outcomes. The script covers various settings and features within the software, such as masked options, resolution input, and batch size, before concluding with the generation of new images and the potential for further refinement and experimentation.



💡Stable Diffusion

Stable Diffusion is an open-source software project that utilizes deep learning to generate images from text descriptions. It is a fast-growing tool in the AI field, particularly useful for artists and designers looking to enhance their work with AI-generated content. In the video, it is used to improve the quality of 3D people in renders by allowing users to input prompts and generate more realistic images.

💡Digital Bunch

Digital Bunch is the group or company that the speaker is a part of. They are likely involved in digital art, design, or technology. In the context of the video, they are sharing their experiences and tutorials on using AI tools like Stable Diffusion in their projects.

💡Artificial Intelligence (AI)

Artificial Intelligence refers to the simulation of human intelligence in machines that are programmed to think and learn. In the video, AI is central to the process of enhancing 3D renders through Stable Diffusion, which uses AI algorithms to interpret text prompts and generate images.

💡Deep Learning

Deep Learning is a subset of machine learning that involves the use of artificial neural networks to analyze various factors of data. In the context of the video, Stable Diffusion employs deep learning to create images that match the text prompts provided by the user.

💡Text to Image Model

A text to image model is a type of AI system that converts textual descriptions into visual images. Stable Diffusion is an example of such a model, which is used in the video to take textual prompts and generate corresponding images, enhancing the realism of 3D people in renders.


Photoshop is a widely used software for image editing and manipulation. In the video, it is mentioned as the tool where the user opens and prepares their files before using Stable Diffusion to process and enhance specific parts of the image.


Cropping is the process of cutting out a part of an image to focus on the area of interest. The video mentions that Stable Diffusion does not yet process large images, so users need to crop the part they are most interested in and save it as a separate file before using the AI tool.

💡Model Selection

Model selection in the context of AI refers to choosing the appropriate pre-trained model for a specific task. The video discusses selecting a model like 'Realistic Vision' that is specialized in faces and people to enhance the appearance of 3D people in renders.


In the context of AI image generation, a prompt is a text description that guides the AI in creating an image. The video explains the use of both positive and negative prompts to direct the AI to generate desired images while avoiding undesired elements.

💡Noising Strength

Noising strength is a parameter in AI image generation that determines how different the generated image will be from the original. The video suggests setting it to a value between 25 to 45 for Stable Diffusion to make the model more realistic without drastically altering the original image.


Resolution refers to the detail an image will hold, which is usually measured in pixels. The video mentions that an optimal resolution for the Stable Diffusion model is 768 pixels, as it provides a balance between quality and detail.

💡Batch Size

Batch size in AI image generation is the number of images generated in one go. The video explains setting the batch size to four, which means the AI generates four different images at a time, providing users with options to choose from.


Stable Diffusion is an open-source software project for deep learning text to image models.

The tutorial demonstrates how to use Stable Diffusion to improve 3D people in renders.

Stable Diffusion has received positive feedback for its ability to assist artists.

The presenter emphasizes the importance of staying updated with AI tools in the creative industry.

The tutorial covers the installation process and usage of Stable Diffusion through its web interface.

Cropping and saving specific parts of an image is necessary as Stable Diffusion does not process large images.

Selecting the right model is crucial, with 'Realistic Vision' being recommended for faces and people.

The 'Photon' model is effective for realistic vegetation and environments.

Positive and negative prompts are used to guide the AI in generating desired and undesired results.

Defining the element to change and describing it with adjectives is key in crafting prompts.

Negative prompts help to avoid unwanted attributes such as anime or cartoon styles.

Optimal resolution for the model is 768 pixels for quality and detail.

Batch size determines the number of different images generated, with four being a good balance.

Denoiising strength is set to 35 for a balance between realism and change from the original image.

Stable Diffusion can generate images locally and is computed on the user's hardware.

The tool is adept at tweaking clothes and sometimes produces more realistic results than 3D models.

AI can sometimes hallucinate, leading to weird, creepy, or funny results when noise values are high.

The presenter invites viewers to share their experiences and outcomes achieved with Stable Diffusion.

The tutorial concludes with an encouragement for further exploration and testing of AI tools.