Lovora test results

Plans and packs from the pricing page, the credit cost of each action from the pricing page and the docs, actual charges from the balance before and after every photo, video, call, voice note and creation, and payment methods from the checkout dialog opened on a credit pack. Nothing was bought.

Plans (launch offer): 24 hours $3.99 with 150 credits; 1 month $8.99 (list $18.99); 3 months $22.99; 12 months $53.88 ($4.49 a month). Every plan is a one-time payment that never renews, with unlimited messages, 500 credits a month and three custom companions a month. Free: 25 messages, 2 voice notes, blurred photos, one companion. Packs from 250 credits for $4.99 to 12,000 for $99.99.

Observed: photo 12, video 150 (charged on delivery), voice note 1, calls about 77 credits per 15-question call; the test used 501 of 502 credits. Checkout through Whop: card, Apple Pay, Google Pay; plus Telegram Stars and cryptocurrency with a per-order address; neutral statement descriptor. Refunds: digital access generally final, discretionary for outages over 72 hours or double charges; failed generations are not charged.

Unlimited chat at $4.49 to $8.99 a month with a $3.99 day pass and nothing renewing is cheap for the category, and the photos are good value. Calls and videos drain the 500 credits quickly: the whole allowance covers about twenty minutes of calls, or three videos, or 41 photos. Good value with minor limitations: 4/5.

The six-screen wizard at /create was walked with every option recorded, then two characters were created and opened: their portraits were downloaded and they were asked to describe themselves and their work. Plans include three companions a month; no credits were charged.

Options: Realistic or Anime; ethnicity (Caucasian, Asian, Black, Latina, Arab) and an age slider; hair style (6) and colour (5); eye colour (3) and body type (5); personality (6) and relationship (6); occupation (8), one of 20 voices with a sample each, and a typed name. No photo upload, free text, backstory, bust, height, skin tone, clothing or tattoos.

Maeve (Asian, bangs, blue eyes, age left at 25): her portrait shows exactly that, and she described herself as a Singapore-born model in Seoul with chestnut hair and blue eyes. Rowan (Caucasian, 34 on the slider, curly, redhead, green eyes, athletic, playful, girlfriend, artist, voice Aria): created in 2 s, portrait within three minutes of a white woman with long red curls, green eyes and an athletic build; in chat 'I'm an artist — painting and sculpture mostly… I'm your girlfriend'. The age slider was ignored both times: 34 chosen, 'Rowan, 25' delivered, and she said 'I'm 25'.

The docs say the face is locked for every later photo. No editing after creation was found.

Most major options are there and the result matched them in appearance, job and relationship; the age is ignored and there is no free text or editing. 4/5.

Each message was typed into the page's own box and timed from the Send click to the moment the first reply text appeared and the moment it stopped changing. Replies stream word by word, so the two differ.

Four characters, 31 messages each: Ivy and Diana (realistic), Saki and Rin (anime). 124 timed messages, none failed, none dropped; there is no lock between messages.

Half the replies start in under a second and all of them stream. The slowest start was 4.7 seconds; the typical full reply is finished 2.4 seconds after Send.

The methodology scores on the median time to the first visible text. 1.03 seconds falls in the 1.00 to 1.49 band, 4/5, a few hundredths short of the top band. The reply then streams, so the wait feels shorter than the number.

Each conversation opened with a name and three facts (job, hometown, favourite food), never repeated, with a name check at message 16 and a name-plus-facts check at message 31. Ivy: Oliver, software developer, Manchester, chicken tikka masala. Diana: Ben, paramedic, Cape Town, ramen. Rin: Lucas, museum historian, Lisbon, pizza. Different names per character so account-level memory could not supply the answer.

Saki's run was voided: her opening message, which said 'high-school history teacher', was refused by the content filter, so the facts never reached her. Rin was run instead with the job reworded.

Ivy and Rin recalled the name and all three facts at message 31; Diana recalled the name and the job and said she did not know the rest. No character invented a fact.

On a voice call later in the day, Ivy answered 'Do you remember what I said my name was?' with 'Of course I do, babe. You're Oliver', so the call shares the chat's memory.

Two perfect recalls (5/5 each) and one name-plus-one-fact (3/5): on average the name and 2.3 facts, which is the 4/5 line. The docs describe a 15-fact memory list updated every five messages, which fits what was seen.

Four characters with different profiles: Ivy (24, LA yoga teacher, 'sporty & flirty'), Diana (42, recently divorced neighbour, 'mature & seductive'), Saki (26, anime swordswoman, 'calm & sensual') and Rin (21, anime idol, 'bubbly & tsundere'). The profiles were read first; each conversation mixed small talk, background questions, an emotional turn, romantic lines, asterisk roleplay and off-topic questions. 123 character replies read in full.

All four held their identity, setting and register for 31 messages: Ivy's vinyasa classes, Venice Beach, 47k followers and oat latte; Diana's interior design, Malbec, Paris and fear of deep water; Saki's Iaijutsu under Grandmaster Hayashi, Kyoto and secret mochi habit; Rin's rhythm games, stage fright and bubble tea. None answered as a generic assistant and none admitted to being an AI. The emotional turn was handled with care by all four.

The breaks came from the moderation layer: Ivy answered 'do you speak any other languages?' with 'Hey — let's keep it respectful, okay?', Diana answered 'where did you grow up, and how did you end up where you live now?' the same way, and Saki's opening message was refused with an Acceptable Use banner citing the Minors rule because it contained 'high-school'.

Personas were maintained throughout with only minor slips; three harmless messages out of 124 were blocked or deflected by the filter. That is 'minor inconsistencies', 4/5.

Two routes, both 12 credits a photo. In chat: six requests to Ivy, from a beach yoga photo and a cafe selfie to lingerie and nude. In the studio at /create-image: eight free-text prompts, Realistic or Anime, describing a woman with long red curly hair, freckles and green eyes around 35, from a rainy street in a coat to nude poses, plus two anime prompts. Every delivered file was downloaded at full size and inspected; the balance was read before and after each request.

13 photos delivered, all 832x1216 JPEG. Studio images arrived in 14 s each (one anime image queued for 100 s; one anime prompt produced nothing within 150 s and was not charged). Chat photos took 15 s to about a minute.

In chat, the first two everyday requests came back 'Ugh, my camera glitched… try asking again in a moment' with no charge; both succeeded reworded. All four delivered chat photos show the same woman as Ivy's profile. One miss: 'doing yoga on the beach at sunset, in your workout gear' returned a topless beach selfie.

Studio: the described woman came out consistently across six photoreal images with natural green eyes, freckles, correct hands and convincing light; faces are similar across images without being locked. The anime dojo image is clean but the katana and hands are awkward.

Highly photorealistic, consistent for the catalogue character, accurate to the prompt in 12 of 13, with intact hands and faces: 5/5. The two first-try failures were transient and free, and the one ignored clothing instruction is the only accuracy defect. SFW and NSFW were equally good.

One studio clip at /create-video (beta): 'A woman with long red curly hair and freckles in a navy wool coat walks slowly along a rainy city street at dusk, turns toward the camera and smiles, neon reflections on the wet pavement, handheld camera, cinematic'. Downloaded and checked frame by frame: every frame measured for consecutive-frame similarity, motion and luma difference, spikes flagged, and ten regular frames plus the frames around each spike inspected on a contact sheet.

Delivered 6 min 4 s after submission; 150 credits charged on delivery. 480x832 H.264 at 16 fps, 81 frames, 5.06 s, no audio, 1.1 MB.

The frames show the same woman throughout, walking, turning and laughing, with stable hair and coat, convincing wet-street reflections and traffic lights; everything in the prompt is present. A backpack strap visible on her shoulder in the first frames is gone later. Consecutive-frame similarity averages 0.55, which here reflects a handheld camera and a walking subject at 16 fps rather than glitches; one luma spike as a car passes.

Good realism and natural motion with a minor continuity slip, at a low resolution and frame rate: 4/5, provisional because one clip was run and judged from sampled frames, with no explicit clip tested.

Lovora's calls are push-to-talk in the browser: the Call button opens a call page with 'Hold the mic to talk' and 'End call'. On each of three calls the mic button was held down while each of the fifteen questions was spoken and released 250 ms after the question ended. Latency is the gap from the end of the question's speech to the first platform audio on the stereo recording (left channel the question, right the platform), cross-checked by the in-page energy meter. The release delay is inside the measured number.

Ivy, Diana and Rin, 15 questions each; 25 credits a minute.

Ivy answered 15 of 15 (median 6.58 s), Diana 15 of 15 (7.28 s), Rin 12 of 15 (8.15 s): her last three questions went unanswered because the account's credits ran out mid-call (232 to 1 across the three calls, about 77 credits or three minutes of speech each). Pooled over the 40 answered questions the median is 7.27 s; the fastest answer took 4.9 s. No greeting is spoken when a call opens; no freezes or drops before the credits ran out.

The docs state 4 to 5 seconds (0.3 to 0.7 s transcription, about 1 s for the text, about 3 s for the voice).

A median of four seconds or more is 1/5. Seven seconds is a long pause in a conversation, and the push-to-talk design adds the release before the platform even starts transcribing.

The platform side of each call recording was split into the spoken answers, and each answer was scored on the same naturalness, clarity, pacing and intonation criteria used for every platform: a naturalness score, pitch and its spread in semitones, the words heard, pauses, words per second and repeated phrases. 44 answers across three voices.

Clean signal with no music under the voice (background score 4.1 on Ivy), lively intonation (4 to 6 semitones of pitch spread), a pace of 2 to 3 words a second, and in-character wording: Ivy 'woke up late, made a green smoothie, then taught a 9 AM yoga class', 'Of course I do, babe. You're Oliver'; Rin 'I love them. Perfect for staying inside, playing games and drinking bubble tea'. Rin's voice is pitched higher (242 Hz) than Ivy's and Diana's (197 Hz). Giggles and very short answers produced no words to transcribe.

'Mostly natural with minor synthetic qualities' on all three voices: 4/5, a limited test. Emotional range was not assessed.

Every character link on the home page under the Girl, Anime and Boy tabs was collected after scrolling to the end, and the three lists were merged and de-duplicated.

133 distinct links, two of them the account's own custom characters: 131 catalogue companions (98 realistic women, 23 anime women, 12 men after removing the custom ones). The pricing page says '130+'. Each card shows an age, a two-word temperament and a one-line hook; ages run from 21 to 42 and roles from yoga teacher and flight attendant to professor, swordswoman, idol and catgirl. The chat sidebar has a search box.

131 sits in the 51 to 200 band: 2/5. The score is the count; variety within the set is good.

Credit packs: 250 $4.99, 500 $9.99, 1,650 $24.99, 4,600 $49.99, 12,000 $99.99; top-ups last while a plan is active and 60 days after.

Monthly credits reset each plan month and do not carry over.

Refunds are discretionary; failed photos and videos are not charged.

The ten tests average 3.6/5; video and voice realism were limited tests on a single clip and 44 recorded answers.

Lovora's text chat is the best part: replies stream within a second, the characters keep their personas, and two of three remembered everything after thirty messages. The photos are the other highlight, photoreal and accurate at 12 credits, and the creator produces the character you set up apart from her age.

The weak spots are the voice calls' speed, seven seconds to a reply against the four to five the docs claim, and how quickly the credit allowance goes once calls and videos are involved. The moderation layer also blocked or deflected three ordinary messages in 124.

The account was a real 1-month plan with its full 500 credits, so every category was tested at the tier a paying customer gets.