Coconote
AI notes
AI voice & video notes
Try for free
🎨
Character Consistency Tutorial
Jun 29, 2024
Character Consistency Tutorial
Introduction
Expansion on character creation using face grid.
Focus on in-depth creation process.
Tools:
Microsoft Paint
,
Focus
(with various models).
Face Grid
Original grid: 9 faces (good for cartoons/anime).
Improved grid: 4 faces (better for realistic images).
Method: Image search on Google (e.g., "different angle face reference sheet").
Creating a Face Grid
Using Microsoft Paint:
Copy and paste the image into Paint.
Use the selection rectangle tool to outline and copy faces.
Resize the canvas to 1024x1024 pixels.
Paste and resize faces to fit within the grid (90%).
Separate boxes with the line tool.
Save the grid (background removal optional).
Using Focus
Load the face grid into Focus.
Set the size to 1024x1024.
Avoid Focus V2 for realism (use other models like Juggernaut, Realviz, Realistic Stock Photo).
Simple prompt start: e.g., "Italian female".
Add details gradually (age, hair, facial features).
Aim for realism: add phrases like "no makeup, imperfections, blemishes".
Generating Emotions
Using Array Support Function:
Uncheck random seed.
Enable disable seed increment in debug settings.
Generate multiple images with a consistent seed and different prompts (e.g., happy, laughing, angry, crying).
Enhance emotions with specific phrases (e.g., "hysterically crying").
Inpainting & Detail Fixes
Use the same model for inpainting as for generating the images.
Fix one set of eyes at a time.
Save images after editing.
Split the grid into separate images using image software (e.g., Microsoft Paint).
Using Generated Faces
Face Swap
:
Load images into face swap.
Expect likeness, not exact match.
Keep settings high (weight at 0.9-1).
Use different angles for better consistency.
Outpainting
:
Use a close-up image and expand using inpaint/outpaint settings.
Drag down the image, select directions, and generate.
Night-time Shoots
:
Use the same process with darker images for evening/night-time pictures.
Poses
:
Find high-resolution reference images on Google.
Use inpaint to correct issues (mask face area).
Advanced settings: enable mixing image prompt within the inpaint box.
Photoshop Integration
Blending Specific Faces:
Create a transparent image of the face and overlay it on a photo.
Correct color and lighting as needed.
Use detail settings in Focus to blend edges.
Low denoise for minor corrections.
Reverse Process (Blank Canvas Outpaint)
Use a face on a blank canvas.
Mask the face and invert the mask (debug settings).
Select directions and generate to expand around the face.
Adjust prompts as necessary.
Watch for mask issues (activate/unactivate invert mask as needed).
General Tips
Use appropriate weights and stop at settings for desired angles and likeness.
Works with all image types (realistic, comic book, anime, etc.).
For non-human faces, gradually build new form using grid and prompt.
Conclusion
Practice and experimentation are crucial.
Use these methods to develop consistent character images with realism or creative stylization.
Continuous learning and adaption of techniques.
📄
Full transcript