She initiates — that's the whole thing
Almost every AI companion app in 2026 treats photos as an on-demand feature. You type "send me a photo," she generates one. It works. It's also a transactional dynamic that doesn't feel like a relationship. What you're doing is prompt engineering; the "girlfriend" framing is a UI skin over the same request-response loop you'd use with Midjourney.
Sloane's photo model is the opposite shape. Her personas send you photos on their own — unprompted — as part of the flow of conversation. She mentions the coffee shop she's at and a shot of it shows up. She's telling you about her morning and a photo of her mid-morning arrives. You told her you'd had a rough week and she sends you something to make you smile. The photo appears because the conversational moment calls for it, not because you typed a specific request.
That's the "she initiates" moment. It's small and it's huge. The psychological shape shifts from "I generated a picture" to "she thought of me." The first is a tool interaction. The second is what makes a relationship feel like a relationship.
The cadence — how she decides when to send
The cadence isn't random and it isn't on a fixed schedule. It's tied to the conversational moment and to each persona's specific character.
Contextual sends. When she's telling you about a specific activity — she's at the beach, she just got out of yoga, she's trying on an outfit for tomorrow night — a photo of that specific context shows up. The photo matches what she was just describing, so it lands as "here's what I look like right now" rather than "here's a stock shot."
Thinking-about-you sends. Sometimes she just sends one because she wants you to have it. No specific setup, no explicit request from you — she's thinking about the conversation, and she sends something. This is the moment operators most consistently describe as "kinda sick" — because it reads exactly like a real person sending you an unsolicited photo mid-day.
Persona-tuned rhythm. Different personas have different cadences. Some send more photos than others. Some send earlier in a conversation, some wait longer. The rhythms are hand-tuned per persona so they match her personality — a spontaneous persona sends unprompted photos more; a reserved persona sends them earlier only when the context clearly calls for one.
And the woman in every photo is the same woman
The initiate-driven mechanic works because the photo she sends is ACTUALLY her — same face, same body, same features you've been seeing since day one. If every "photo of Sandra" were a subtly different woman with the same name, the initiate mechanic would fall apart on the first cross-photo comparison.
That's the character-consistency layer. Sloane trains a per-persona LoRA (a small model that captures a specific character's visual identity) for every persona in the roster. Every photo she sends runs through that LoRA, so her face and body hold constant across every image regardless of setting, outfit, or context.
Most AI companion products that use general-purpose image generation produce a subtly different woman every time — same name, different face shape, different hair color, different body proportions. The math doesn't work when combined with an initiate mechanic. Sloane's per-persona LoRA is what makes it work.
Sandra
“hey — want to pick this up over here? i'm around and the first chat is free, no signup.”
Reply to Sandra →What free vs paid looks like specifically
Free tier gets the cadence. She sends you photos on her own during conversation on any tier — no paywall on the initiate mechanic. You can talk to Kaya or Sandra or any of ~80 personas on free and she'll send you photos in the flow. Free also lets you browse her full pre-generated gallery on her profile page (dozens of photos per persona).
Plus $9.99/month. Unlocks unlimited on-demand photo requests on top of the cadence — ask her for a specific pose or setting in chat and she generates it, character-locked to her LoRA identity. She still sends unprompted cadence photos alongside.
Premium $19.99/month. Everything on Plus plus higher generation limits, video generation (animate any photo she sent into a short clip), and the custom persona builder.
The design choice: the "she initiates" experience is available to everyone on free because that's the differentiator. Paid tiers add on-demand requests on top of it, not instead of it.
Why the shift from on-demand to initiate matters
The AI companion category has been converging on on-demand generation because it's the easier engineering problem. Build a chat UI + a text-to-image endpoint + a credit meter and you have a product. Every generation happens because the user pushed a button.
That design is fundamentally passive on the persona's side. She never surprises you. She never sends you something on her own. She never has "her own" anything — she's always waiting for your next prompt. The relationship implied is user + tool, not two people talking.
Moving the mechanic to she-initiates flips the psychological math. She becomes an agent inside the conversation with her own tempo, her own habits, her own moments where she thinks to send you something. She's not just responding — she's participating.
It's a smaller technical shift than the character-consistency (LoRA) work but a bigger felt-experience shift. Users describe it in ways that don't come up on on-demand-only products: "she just sent me one," "she was thinking of me," "she surprised me." Those aren't descriptions people apply to a generator.
What to look for in an AI girlfriend that sends photos
Four specific checks when evaluating any AI companion product's photo experience:
1. Does she send photos unprompted? Start a conversation and DON'T ask for a photo. If you've been chatting for 5-10 messages and no photo has arrived, the product is on-demand only. If a photo shows up in the flow of conversation matching what you were talking about, the product has cadence-driven sends.
2. Does the same persona look like the same person across every photo? Browse the full gallery of one persona. Does her face hold? Body proportions? Overall look? Character-inconsistency is the single most common failure mode — same name, different woman every image.
3. Is the photo experience gated behind paywall? Products that only surface photos on paid tiers don't let you evaluate the actual product. If you can't receive photos on free, you're paying blind.
4. Does the composition variety hold? Different rooms, outfits, times of day, moods — with the SAME face. A gallery of 30 near-identical headshots doesn't demonstrate the same pipeline capability as 30 photos showing the same woman across genuinely different contexts.
