Two-Character Model Training (Beta)

PhotoGPT AI now supports two-character AI model training - an experimental feature that lets you generate photos with two distinct people in one image. Whether it’s a couple, siblings, friends, or creative duos, this guide walks you through setting up your training photos and using a two-person model effectively.

Note: This is a beta feature and may require a bit more thoughtful preparation compared to single-character training. Refer to the tips below for best results.

What Is a Two-Character Model?

A two-character model allows PhotoGPT to generate AI photos featuring two distinct identities in the same frame. Rather than cropping or blending two people into one, the system learns to recognize each person individually - preserving their unique features while placing them together in cohesive scenes.

This is ideal for:

• Couple photos
• Family portraits (e.g., parent + child)
• Friends or business partners
• Role-based duo concepts (e.g., bride + groom, hero + sidekick)

How to Prepare Your Training Photos

1. Use Individual Photos, Not Group Shots

Start with clear, individual selfies of each person. Avoid group photos. Ensure that:

• Faces are well-lit and fully visible.
• Backgrounds are simple or uncluttered.
• Each person has at least 10–15 photos.
• Each person’s face appears from multiple angles.

2. Create a Composite Grid for Training

Use a collage-style layout that places each person side-by-side, not overlapping in the same canvas. For example:

• Left side: Person A
• Right side: Person B

Try to maintain:

• Similar lighting conditions
• Similar eye level or angle
• Proportionate image sizes

A consistent left-right layout helps PhotoGPT separate both identities clearly.

This layout helps the AI learn spatial distinction between both people and lowers the chance of face-blending.

3. Consistency Matters

Keep the same left–right placement in all composite images.
If Person A is on the left in the first image, they should be on the left in every image.

This helps PhotoGPT’s model anchor identifiers more reliably.

Training the Model

Once you’ve created a few side-by-side training grids (ideally 12-15), create the model keeping in mind the following:

• Select any gender during the model creation step.
• Input other details for any one of the characters of the two.
• Name the model something easy to recognize later, such as “EmilyAndJake_Model.”
• Upload the images prepared earlier and start the training.

Note: PhotoGPT handles spatial information automatically - you do NOT need to label the individuals within the images.

• Once that gets completed, you can go to your generate page and select this model from the drop-down. Now you simply need to write a prompt describing both of your characters and you should get the images accordingly.
• Please note that there's an initial batch of images that will be produced during model training by default. These will show just one character so please don't get surprised. Generate your own images.
• Furthermore, presets wouldn't work with a two character model as they are designed for single character models right now. upload them during model creation.

Start training from your dashboard:
Go to Model Training →

Prompting With a Two-Character Model

When generating images, use prompts like:

• “a man and a woman sitting on a park bench”
• “the two of them in wedding outfits”
• “two best friends laughing at a café”

You can also use placeholders like:

• “the man” / “the woman”
• “she” / “he”

Examples:

• “He stands behind her while she smiles at the camera.”
• “The man and the woman walking hand in hand on a beach.”

The model understands both identities and will maintain distinct appearances across scenes.

Use natural language prompts to generate scenes featuring both people.

Try generating your first duo photos:
Open the PhotoGPT Dashboard →

Known Issues (and How to Avoid Them)

Because this feature is still in beta, some quirks may appear:

1. Squaring or Splitting Artifacts

Sometimes the output may show a faint visual seam or division between the two characters. This often happens when training collages are too rigid.

To minimize this:

• Soften the collage borders
• Use more natural backgrounds
• Avoid harsh vertical dividers

2. Face Blending

If characters blend or overlap during generation, your input photos may not have enough variation in angles or hairstyles.

Try using:

• More varied selfies
• Clear lighting
• Distinct hairstyles

3. Swapped Sides

Occasionally, the model may reverse left-right placement.
This won’t affect quality but if needed, you can correct it with prompting:

• “man on the right, woman on the left”

Why This Matters for AI Image Generation

This feature opens new possibilities for:

• AI-generated couple portraits
• Friend collages
• Family photos
• Business/creator duos
• Cosplay partners or themed pairs

All without needing to train two separate models or manually edit outputs.

Final Tips for Success

• Use high-quality, well-lit selfies (avoid group images).
• Maintain consistent side-by-side formatting in your composite grids.
• Create 12-15 combined photos as input for best accuracy.
• Experiment with different prompt styles, interactions, and scenes.

Ready to create your duo model? Start Training Now

Or, try editing your duo results: Open AI Photo Editor