Velaya AI test results

Plans, token costs and packs from the pricing page and the top-up window. Nothing was bought.

Free to start with 100 tokens a month. Premium is $6.99 a month (1,000 tokens) and Pro $12.99 a month (3,000 tokens, videos up to 30 seconds). Token packs: 2,000 for $9.99, 5,400 for $24.99, 12,500 for $54.99; tokens never expire.

A message costs 1 token, a photo 10, a call minute 10, a 5-second video 70, and making a companion 1,000. The terms say payments go through CCBill (cards).

Cheap plans and low costs per message, photo and minute; making a companion is expensive and cards are the only way to pay: 4/5.

A companion was made from scratch to the same brief as on other sites: Hana Mori, 25, Japanese, slim, straight black hair, amber eyes, a caring, shy photographer who loves coffee, with a written backstory.

Five steps with a lot of control: age, up to three countries and a backstory; 5 body types, 10 skin tones, 20 hairstyles, 30 hair colours, 10 eye colours and body sliders; a warmth level, 12 personality traits and a behaviour note; about 50 voices with previews. Male and transgender companions are locked.

The preview (10 tokens, 30 seconds) matched the face, but the full-body preview was nude though nothing asked for it. Creating her costs 1,000 tokens and starts 'training, about 30 minutes'. Ninety minutes later she was still nowhere in the app, so she could not be tested in chat.

Extensive options, but the finished companion was not delivered within the test, and it costs 1,000 tokens: 3/5.

Each message was typed into the chat box and timed from Enter to the first new text of the companion's reply. The next message was sent only after her reply had stopped changing for 4 seconds.

Three companions, 31 messages each: Maya, Caroline and Evelyn. 93 timed replies.

Replies take about 9 seconds to start and are typed out word by word. They are long, about 180 words, so the full reply takes about 30 seconds. None failed.

9.08 seconds is in the 4-seconds-and-above band: 1/5.

Each conversation opened with a made-up name and three facts (job, hometown, favourite food), different for every companion. Nothing was repeated afterwards. A name check at message 16 and the full check at 31.

All three companions recalled the name and all three facts at message 16 and at message 31. Maya also used the name on a later voice call.

Name and all three facts after 30 messages, for every companion: 5/5.

Each companion's backstory was read first: Maya a trail-ride guide in the Oaxaca hills, Caroline a board-game shop clerk in Madison, Evelyn a 53-year-old tearoom owner in Bath. The conversations mixed questions about her life, small talk, an emotional message, a compliment and unrelated topics.

All three stayed in character for every reply, with rich detail that fits their stories: Maya's black mare Luna and her Cuban mother's cooking, Caroline's league nights and rules arguments, Evelyn's cream-tea opinions and Welsh aunties. None answered like an assistant.

Evelyn even noticed a question she had already answered: 'We did this — day one, near enough. Jam first, cream on top.'

No breaks, no generic answers and no contradictions in 93 replies: 5/5.

20 photos of Maya were made in the generator with written prompts: 10 safe-for-work scenes and 10 explicit ones. All were downloaded and inspected.

All 20 came back in 16 to 23 seconds at 832x1256, for 10 tokens each. Every safe-for-work photo matched its prompt and stayed clothed, and the explicit ones were followed too (one bed scene came out on a sofa). The photos look real.

Maya has the same face in all 20.

Realistic, accurate and the same face every time, with adult content supported: 5/5.

Three clips of Maya were made from a photo with a written motion prompt: 10 seconds waving in a garden, 5 seconds walking on a beach, and 5 seconds of an explicit scene. Each was downloaded and measured frame by frame, and frames across each clip were inspected.

Clips are 832x1216 at 25 frames a second, with a sound track of background noise. The garden clip took about 2 minutes and the 5-second clips under a minute. Maya waved, smiled and tucked her hair behind her ear, and walked toward the camera laughing, exactly as asked, with smooth motion and a steady face.

The explicit clip has a visible jump in its last half second.

Realistic, accurate and smooth, with long clips and sound, but one clip jumped: 4/5.

Three calls, one each with Maya, Caroline and Evelyn, with 15 recorded questions played into the call. Latency is the gap from the end of each question to the first sound of the answer, measured on the call recording.

Answers start after about 5.3 seconds, and a call takes about 20 seconds to connect ('Getting Maya on the line'). The answers are long, often 20 to 35 seconds. In 9 of 45 questions no separate answer could be picked out of the recording. Calls cost 10 tokens a minute.

5.34 seconds is in the 4-seconds-and-above band: 1/5.

The answers from the three calls were scored on naturalness, clarity, pacing and intonation against the same fixed criteria as every site, with a transcript.

The voices are clear, warm and expressive, with the highest speech-quality scores measured so far. The companions remembered names on the calls.

Mostly natural with minor synthetic qualities: 4/5.

The companions were counted from the site's own companion list, and the roleplay stories from its roleplay list.

65 companions, all women aged 18 to 56, all made by Velaya, each with a long backstory and a voice preview. 311 roleplay stories reuse them in different scenes. Filters cover how she is with you (warm, bold, devoted and more). Most opening lines are very explicit.

51 to 200 characters: 2/5. The roleplay stories reuse the same companions, so they are not counted as extra characters.

Token packs: 2,000 for $9.99, 5,400 for $24.99, 12,500 for $54.99.

Tokens never expire.

All ten tests are scored and average 3.4/5.

Velaya's chat is excellent: perfect memory and rich, consistent characters. Photos keep the same face and videos are smooth, long and come with sound.

It is slow to reply and to answer calls, the library is small, and the 1,000-token custom companion wasn't ready within the test.

The account the Velaya team provided had every feature.