Photos, voice, video: what AI companion apps can and can't send (2026)

Photos are close to universal in this category. Calls are not, and voice notes from the character are rarer than the marketing suggests. Below is what each app’s own App Store listing and homepage say, with dates, and nothing inferred from anywhere else.

LORAX is in the table and marked. Everyone in LORAX is fictional and AI-generated, and so is every image any app in this list sends.

What each app says it can send

App	Sends photos	Voice notes from the character	Voice calls	Video calls	You can send photos / voice / video
LORAX	Yes, generated per request, locked until opened with credits	No	No	No	Photos, voice notes and video up to 60 seconds
Replika	Yes, image generation and selfies	—	Yes	Yes, described as showing what you’re seeing	Photos; video during a call
Nomi	Yes, selfies showing the moment they’re in	Yes, send and receive	Yes, hands-free calls	—	Photos and links
Kindroid	Yes, diffusion-generated selfies	—	Yes, real-time voice calls	—	Images and links
Character.AI	—	—	Yes, live voice calls	—	—

Sources, all checked 2026-09-13. Replika: App Store listing version 12.3.5, updated 2026-07-24, plus replika.com. Nomi: App Store listing version 1.12.1, updated 2026-03-31, plus nomi.ai. Kindroid: App Store listing version 1.70, updated 2026-08-18. Character.AI: App Store listing version 1.17.3, updated 2026-09-08.

A dash is not a no

Every cell above is something the company states on a page of its own. A dash means the feature is not named on those pages on that date. It is not a claim of absence.

That distinction matters more in this category than in most, because feature lists here move fast and roundup articles copy each other. Kindroid’s own store description, for example, names selfies and real-time voice calls and says nothing about video either way. We are not going to fill that gap with a guess.

If a cell matters to your decision, open that app’s own page and look. It takes a minute. The answer there is current.

What a generated photo actually is

None of these are photographs. Each image is produced when it is requested, from a description of the character plus whatever the app keeps to hold her appearance steady between pictures.

Consistency is where the apps separate. The cheap version regenerates her from scratch each time, and you notice by the third picture: different face, different room, different hair length. The expensive version carries references forward. That is the whole trick.

In LORAX a scene is settled before the image exists: one of 17 indoor locations, what she is doing there, what she is wearing. Ask for a second photo the same afternoon and she has not changed clothes, because scenes already made that local day are pulled back in and the instruction sent onward is to carry the outfit over. Up to two earlier frames from that day travel with the request, alongside an identity reference that holds her face steady from the first photo to the fiftieth. How the photos are made has the rest.

The photo lands as a locked bubble. Opening it spends credits, the price is on screen before anything is spent, and a generation that comes out wrong returns the credits as a reversing line in your history rather than a deleted one. Two attempts. Then it stops. Details are on the pricing page.

What you can send, and whether it lands

Comparisons skip this column. It is the one that changes conversations.

In LORAX a photo you send is read in a single pass that pulls out six things: a summary, who is in it, the setting, any text visible in the frame read verbatim, where the image came from, and a safety read. What comes back in the conversation is not a summary. She picks one specific thing and reacts to that, the way a person does.

Video is read as eight frames plus a separate audio track, with the frames extracted on your phone rather than the server opening the file. Voice notes get both halves: what you said, and how you said it, from six measured signals including pace, pauses, drawn-out words, laughter, a 26-word filler dictionary and how many people are talking.

When something cannot be measured, she does not mention it. If the analysis does not come back at all, she says nothing about the media rather than producing a line about looking at it later. That rule holds across the app and it is why the reactions land when they do arrive.

Voice and video are the dividing line

Text and pictures are ordinary now. Calls are where the category splits, and where the cost sits.

Replika’s own listing puts text, voice calls and video in the same opening sentence. Nomi and Kindroid both name real-time voice. Character.AI names live voice calls. Every one of those requires a speech pipeline running while you talk, which is a different engineering and cost problem from generating a message.

Where LORAX is behind

Flatly: no voice calls, no video calls, and no voice notes or videos from the characters. If a voice you can hear is the thing you want from a companion app, three of the four other apps in this table name it and we do not.

What we built instead is in the other direction. The characters hear you, read your photos, and watch your videos; they answer in text, pictures and songs. It is also iPhone only, with no Android version and no browser version. The questions page is blunt about the rest, and the overview of how the app works explains what the timing buys you in exchange.

Decide by the column you care about, then check that column on the app’s own page before you pay. The listings above are dated for exactly that reason.