AMAZING AI Art Trick - Better Faces, Similar Variations, Improve old Renders / Stable Diffusion

Introduction

Today, I want to share an incredible technique to enhance your favorite AI art and create variations that maintain the exact composition as the original work. We will be utilizing the D Forum Stability Fusion on Google Colab. If you found the previous tutorial confusing, don't worry! I'll guide you through the setup with all essential steps.

Setting Up Google Colab

1. Access the Link: First, navigate to the link provided in the video description.
2. Connect to GPU: Click the connect button to access the Google server's Nvidia GPU (this does not relate to your own GPU), and grant permission when prompted.
3. Link Google Drive: Allow the Colab to connect to your Google Drive. This step is crucial for saving your work.
4. Configure Initial Settings: Click the play button for the subsequent steps, which may take some time to load.

Uploading the Model

You will need a model file stored in Google Drive:

• Check that there’s a folder named “Ai” in your Google Drive with a subfolder named “models.” If it doesn’t exist, create these folders.
• Download the required model from a suggested link (available in the video description).
• Upload this file to the "models" folder in your Google Drive.

Creating Variations of Your Artwork

1. Change Animation Mode: Set the animation mode to "none" via the dropdown menu, then click the play button to save this setting.

2. Input Prompt: Use the original prompt for your artwork in the designated section. Make sure to surround it with quotation marks, and add a comma at the end.

3. Configure Rendering Settings:

   • Set the same aspect ratio as the original, keeping in mind the width and height in pixels.
   • Set the seed to -1 for random results.
   • Choose a sampler and adjust the steps as desired, typically between 50 to 200 for quality.
   • Set the scale to 7 and configure any additional parameters.
   • Specify the number of variations to create (for example, 10).
   • Name your batch folder for the resultant images.

4. Image Initialization:

   • To create variations, upload the original image (e.g., a cyborg Geisha) to the temporary Google Colab drive.
   • Copy the file path where the image is uploaded and paste it into the init image section.
   • Adjust the strength setting (0.5 for a mix of 50% original image and 50% prompt).

5. Render the Variations: Click the play button to start rendering variations of your original artwork.

Improving Old Renders

To enhance and improve your artworks:

1. Download and Upscale: Use an image upscaling tool, such as Gigapixel AI, to upscale the rendered image (consider 4x or 6x larger).
2. Edit in Affinity Photo: Remove any imperfections (like extra hands or unwanted details) using editing software like Affinity Photo.
3. Re-Upload and Adjust: Follow a similar prompt setup by pasting the new file path and hitting render again, modifying the pixel dimensions as necessary.

Enhancing Faces

For headshot improvements:

1. Cut Out Faces: Extract the face from the upscaled render and sharpen it.
2. Upload and Render: Upload the enhanced face to the Colab temporary drive and copy the path into the init image section.
3. Adjust Settings: Maintain the same prompt but resize as needed (e.g., 800x800).
4. Download and Composite: After rendering, download and overlay the new face onto the original image in Affinity Photo. Adjust opacity and use a mask to reveal the improved face delicately.

By following these detailed steps, you can create visually stunning variations of your AI art, improve the quality of old renders, and particularly enhance the quality of facial details.

Keyword

• AI Art
• Variations
• Composition
• Google Colab
• Stable Diffusion
• Upscaling
• Image Editing
• Face Enhancement

FAQ

Q1: What is the main goal of this tutorial?
A1: The tutorial focuses on improving AI art quality and creating variations with the same composition using Google Colab and Stable Diffusion techniques.

Q2: Do I need an Nvidia GPU to use Google Colab?
A2: No, you are using the Google server's GPU, so you do not need personal hardware.

Q3: How do I upload my AI model?
A3: Create a folder named "Ai" in your Google Drive with a subfolder called "models" and upload your model file there.

Q4: Can I adjust the size of my artwork?
A4: Yes, you can change the resolution within allowable dimensions, as long as the aspect ratio is consistent.

Q5: What tools can be used for upscaling and editing?
A5: Tools like Gigapixel AI for upscaling and Affinity Photo for image editing are recommended.