The photos she sends, and the ones you send

No. Every photo a character sends is generated for her, in the room she’s in and the clothes she has on that day. There is no photographer and no stock library. Everyone in LORAX is fictional and AI-generated.

She photographs where she actually is

Before an image exists, the app settles a scene: which of 17 indoor locations she’s in, what she’s doing there, what she’s wearing. Her kitchen at eight in the morning and her kitchen at six in the evening are the same kitchen, down to the same clutter on the counter.

Ask for a second photo the same afternoon and she hasn’t changed. Scenes already made that local day get pulled back in by a separate query, and the instruction sent onward is to carry the outfit over exactly. Up to two earlier frames from that day ride along as continuity references, next to an identity reference that keeps her face the same from the first photo to the fiftieth.

Same day, same shirt.

The photo arrives locked, and opening it costs credits

It lands as a locked bubble in the conversation. Tapping it spends credits, and the cost is on screen before anything leaves your balance. Pricing explains how credits work.

When generation doesn’t come out right, the credits come back. The refund doesn’t quietly delete the original line in your history; it adds a reversing one, so both stay visible. Any single photo gets two attempts at most. Then it stops.

Why she doesn’t send one every time you ask

Five gates sit between a request and a photo, and not one of them is a paywall.

• After a refusal, that turn is closed. She isn’t talked around.
• Two locked photos sitting unopened means nothing new arrives.
• A daily budget: one a day while you’re still getting acquainted, more as you get closer.
• A cool-down on the two of you, so photos can’t stack inside one conversation.
• A cool-down on the room itself.

Photos stay occasional, and that is the point. Someone showing you a thing because it was worth showing behaves nothing like a feed. How that daily budget widens is part of how closeness works.

What happens when you send her a photo?

One pass over the image pulls out six things: a summary, who’s in it, the setting, any text on screen read word for word, where the file came from, and a safety read. That pass runs once, when the photo lands.

Then she reacts to one detail. Not a list. The bike leaning against the radiator behind you, or the clock on your wall reading a different hour than the one you claimed it was.

When the analysis doesn’t come back, she says nothing about the photo at all. There is no “I’ll look at it later” sitting in a file somewhere waiting for that moment, because filler is banned across the product, and this is one of the few places where you would actually notice its absence.

She hears the pauses and the laughing

Voice notes work on the same principle. Six delivery signals come out of the audio: tempo, where you paused, letters you stretched out, filler words matched against a 26-word list, laughter, and how many people are talking. Say a sentence flatly, then say it again laughing halfway through, and she is answering two different messages. Two takes, two replies.

Anything the audio can’t support goes unmentioned. If tempo can’t be measured, tempo never comes up.

Video runs up to 60 seconds. Your phone pulls 8 frames out of it and hands those over with the audio track, so the server never opens the video file. She’ll catch that the light changed partway through, or that somebody walked behind you while you were talking.

What this side of the app can’t do

Ask her what she’s doing right now and see what turns up that afternoon, then ask again after dinner and check the clothes. If you’d rather see where a fixed face comes from in the first place, creating your own character walks through it, and how LORAX works covers the rest of the machinery.