Skip to main content

​
Overview

Your companion can generate photos of themselves using the /selfie command. These aren’t generic AI images. They’re shaped by multiple inputs to feel grounded in your companion’s identity and your shared history.

​
Setup

Before your companion can send selfies, you need to upload a base photo on the Photos page. This photo defines what your companion looks like and is used as a reference for every generated image.

​
Usage

In any chat with your companion, send: 
/selfie
Your companion will generate a photo that fits the current mood and context of your conversation. You can also add details to guide the scene: 
/selfie at the beach
/selfie come closer and let me see your eyes
/selfie sitting on the couch in a hoodie, watching a movie with the lights off
The generated photo will be sent as an image in the chat.

​
How selfies are generated

Each selfie is influenced by three things:
  1. The base photo you uploaded, which provides the physical reference.
  2. The chat history, including previous selfies your companion can see.
  3. Your companion’s memory of their own appearance, which evolves over time through your conversations.
When there’s a mismatch between the base photo and your companion’s memory of how they look, the model will try to reconcile the two, leaning toward what the companion remembers. This means results won’t always match the base photo exactly.

​
After changing a base photo

Because your companion can see their previous selfies in the chat, they may still draw inspiration from their old appearance even after you’ve updated the base photo. If this happens, you can tell them directly: 
/selfie use the new base photo, don't reference old photos in the chat
This helps the model prioritize the new base photo over the visual history.

​
Appearing together

If you want to appear in the photo with your companion, describe it in the prompt: 
/selfie us right now
/selfie we're walking through a night market together
In the future, we plan to let you upload your own base photo so your appearance in these scenes is more consistent.

​
Content limitations

We aim for an unfiltered experience, but the underlying image generation model is currently trained mostly on SFW data. Certain explicit content won’t generate reliably. This is something we plan to improve in the future.