How Realistic Are Lovora’s AI Girlfriends? Face, Photos, Voice and Memory, Measured

• 131 ready-made companions: 98 realistic women, 22 anime characters and 11 men (About Lovora).
• Companions you design in a builder with fixed choices and no photo upload, in a realistic or an anime style; the minimum age of a character is 23 (Designing your own companion).
• Every one is a fictional adult. Nobody can be a copy of a real person.

The most visible failure of AI companions is a face that changes from picture to picture. On Lovora a realistic companion’s face is pinned in writing: her profile holds a detailed description of it — nationality and face shape, nose, lips, brows and one distinguishing mark — and that description goes into every image generated of her. For a companion you design, the same kind of description is built automatically from the choices you make in the builder.

• It is a description, not a face swap. No face is cut from one image and pasted into another. This keeps the skin and light natural, and it is why the likeness is strong but not pixel-identical.
• What can drift: hair styling, make-up, lighting and how old she seems from one shot to the next. Whatever the description does not fix, the generator chooses.
• How to test it: open her profile, ask for three photos in different places, and compare the brow line, the nose and the distinguishing mark first, not the hair.

• A set per realistic girl. Her profile carries five photos: the cover, a second profile shot, a lying pose, a view from behind and a story photo. Which of them are shown to a visitor depends on the law where he is; see Age verification.
• Speed, measured. The image worker produced a photo in about 6 seconds with the worker already running, and in about 2 minutes when it had to start from cold (measured on 3 October 2026). A photo in the chat includes the preparation around the generation, so it usually arrives in under a minute and can take several minutes at quiet times.
• Size: chat photos are portrait, 832 × 1216 pixels (Photos and videos).
• Re-use: to save time, a photo that was already generated of her, that matches your request and that you have not received before, can be delivered instead of a new one. If nothing matches, a new one is made.
• Check before delivery. Finished images are screened automatically; one that fails is discarded and not charged.

• 24 voices: 14 women’s voices, 6 anime-style voices and 4 men’s voices.
• The voices are synthesised from clips designed for each voice and each language; no real person’s voice is cloned.
• Voice notes and push-to-talk calls work in all nine languages. One call exchange takes about 0.3–0.7 seconds to transcribe what you said, about 1 second to write her reply and about 3 seconds to speak it in English (September 2026).
• Quality is not equal across languages: best in English and German, weakest in Arabic. The measurements are in Languages on Lovora, and the call details in Voice notes and voice calls.

Realism in conversation is mostly remembering. She reads the last 40 messages of the chat word for word (24 in a voice call) and receives up to 15 lasting facts about you, refreshed every 5 of your messages. Memory is kept per companion and erased with your account. The mechanism and its limits are in How memory works.

• Realistic: photographic faces and bodies, a pinned face description, the set of photos described above.
• Anime: 22 ready-made characters and an anime style in the builder. They keep one consistent character design rather than a photographic face. See Anime AI girlfriends.
• The same chat, memory, voice and prices apply to both; only the look of the pictures differs.

• Messages are unlimited with a plan, with a fair-use ceiling of 500 a day.
• A photo is 12 credits (a variant or regeneration 8) and a short video 150 credits. Photos are never unlimited.
• A plan includes 500 credits a month, the 24-hour pass 150. A free account sees photos blurred and can play 2 voice notes.
• Full table: Plans and credits; what is free: Free access.

• They are AI images. Hands, jewellery, text on clothing and background details can still be wrong in a generated picture; ask again if one is.
• The likeness is strong, not guaranteed. The face is pinned by words, so expect the same woman, not the same pixels.
• Video is shorter and rougher than photos. A clip lasts a few seconds and is animated from a still; it takes minutes, not seconds.
• Voices have an accent of their own. They are synthetic, and in some languages a listener will notice it.
• Memory is a short list written by a model and can be wrong; it is corrected by telling her.
• No real people. Nothing here can depict a real person, and what can and cannot be depicted is set out in the Acceptable Use Policy.

Five questions make a claim of realism comparable between apps:

• Is the face fixed by a method, and which one — a description, a stored face, or nothing?
• How long does one photo take, measured warm and cold?
• How many photos does a character come with, and how many can you ask for?
• Is the voice designed or cloned, and is its quality measured per language?
• What does the app remember, in numbers?

For Lovora: a written face description; about 6 seconds warm and about 2 minutes cold; five profile photos per realistic girl and credit-metered photos in chat; designed synthetic voices with published measurements; 40 messages and 15 facts.