Fantaly test results
Tested hands-on by Miles Foster on 16 October 2026, in a desktop browser. Every measurement, table, recording and screenshot behind the grade, test by test.
Overall
10 tests plus my verdict.
1. Platform overview
Fantaly (fantaly.ai) is a live video AI companion: you pick a character, start a Live Cam, and a moving video of her answers out loud while you type or talk. It was tested on 16 October 2026 on an account the Fantaly team provided.
The live video is the star. She moves her whole body, smiles, laughs and acts on most requests, and replies start in about 1.7 seconds. There are more than 1,500 public characters, and the image generator is quick and realistic.
Memory is patchy over a long chat, characters drift into the same sweet tone, the background you write for your own character is ignored, and calls answer in about 4 seconds.
Features available
- Live Cam: a live, moving video of the character that answers out loud, with typing or a tap-to-talk microphone
- More than 1,500 public characters (women, anime, men), mostly made by other users
- Character creator: free starter looks, an image generator (guided or your own prompt) and an image editor, a name and a written background
- Images for character looks, 5 credits each
- Chat history kept between Live Cams
- No subscription: one-time video passes and credit packs
Tested successfully
- Chat speed on 93 timed messages across three characters
- Memory on three 31-message conversations
- Character consistency across 93 replies
- 20 images (10 safe-for-work, 10 explicit) from written prompts
- Live video: a minute of motion measured frame by frame, and 13 spoken-to requests checked on video
- Three 15-question voice calls through the microphone button, recorded and analysed
- Three characters made in the creator and checked in chat
- Character count from the public library, passes, credit packs and payment methods
Testing limitations
- A Live Cam lasts 5 minutes, so each 31-message chat ran across two Live Cams with the same history
- Images are made for a character's look; there are no photos in chat
- Nothing was bought; the checkout page was opened to see the payment methods
2. Core feature grades
| Pricing & Value | 4/5Good |
| Character Customization | 2/5Below average |
| Chat Response Speed | 3/5Average |
| Chat Memory | 3/5Average |
| Character Consistency | 3/5Average |
| Image Generation | 4/5Good |
| Video Generation | 4/5Good |
| Voice Call Latency | 2/5Below average |
| Voice Realism | 4/5Good |
| Character Variety | 5/5Excellent |
| My verdictLive video that moves and acts on requests, for under $10 a week, is something few AI girlfriend sites offer. | +0.2 |
| Overall | 3.6/5Good |
3. Additional interactive features
| Voice Messages in Chat | Available | Every reply is spoken on the live video; you can talk back with the microphone button. |
| Talks When You're Quiet | Available | If you stop typing, the character keeps adding short lines on her own (two or three a minute). |
| Social Media Feed | Unavailable | No feed; public characters have a page with a short story and a play count. |
1. Pricing, Value & Payment Methods
Testing method
Passes, credit packs and payment methods from the purchase window and the checkout page. Nothing was bought.
Results
- Passes
- $9.99 for 7 days, $199.99 for a year (1 hour a day)
- Live Cam
- 30 credits a minute (about $0.25 to $0.30)
- Image
- 5 credits
- Payment methods
- PayPal, card, Apple Pay, Google Pay (5/5)
There is no subscription. A 7-day pass is $9.99 and an annual pass $199.99, one-time, each with 1 hour of Live Cam a day (both on a launch special). Credit packs are 500 credits for $4.99 (about 16 minutes) and 3,500 for $28.99 (about 116 minutes).
A Live Cam costs 30 credits a minute, about 25 to 30 cents, and messages are included. Images are 5 credits. Payment is by PayPal, card, Apple Pay or Google Pay.
Visual evidence
Why this grade
A week of daily live video for under $10 with four ways to pay is good value; pay-as-you-go minutes add up: 4/5.
2. Character Customization
Testing method
Three characters were made: Maeve Doyle from a written image prompt (a red-haired Galway librarian), Jade Park from a free starter look (a fitness coach from Seoul), and Yuki with the guided options (an anime ice mage). Each got a written background, then five questions in a Live Cam about her job, home, hobbies, pets and personality.
Results
The look is easy to make and accurate: a starter look, a guided image (2 styles, 9 hairstyles, 19 outfits, 21 settings, 4 moods and a details box) or your own prompt, plus an image editor. Each look was ready in about 8 seconds and matched what was asked. There are no choices for voice, age, body or personality.
The written background was ignored. Maeve said she was 'a freelance designer from Dublin' with a ginger cat called Finn, Jade 'a freelance designer living here in Seoul', and Yuki 'a freelance illustrator living here in Tokyo'. Only the name and the look carried over.
Maeve (written as a Galway librarian with a cat called Yeats)
I'm Maeve, a freelance designer from Dublin. Just living life!
Yuki (written as an ice mage from Frosthold)
I'm Yuki, a freelance illustrator living here in Tokyo!
Visual evidence
Why this grade
Good control over the look, but the personality and background don't carry into chat and there's no voice choice: 2/5.
3. Chat Response Speed
Testing method
Each message was typed during a Live Cam and timed from Send to the first new text of the character's reply. The next message was sent only after her text had stopped changing for 2.5 seconds and her voice had been silent for 1.5 seconds.
Three characters, 31 messages each: Aria Rosewood, Elsa Hart and Empress Barbara, all from the public library. 93 timed replies.
Results
- Median to first text
- 1.70 s
- Mean
- 3.78 s
- 95th percentile
- 11.2 s
- Fastest / slowest
- 0.15 s / 15.4 s
- Under 1 second
- 1%
- Samples / failures
- 93 / 0
- Until her text finished (median)
- 8.7 s
Most replies start in about 1.6 to 1.7 seconds. Her words appear sentence by sentence while she speaks, so some turns begin with the end of her previous answer, which makes the average higher than the typical wait. None failed.
Visual evidence
Why this grade
The methodology scores on the median time to the first visible text. 1.70 seconds is in the 1.50 to 2.49 second band: 3/5.
Limitations
- Typing works only while a Live Cam is running.
4. Chat Memory
Testing method
Each conversation opened with a made-up name and three facts (job, hometown, favourite food), different for every character, after clearing the saved history. Nothing was repeated afterwards. A name check at message 16 and the full check at 31. A Live Cam fits about 24 messages, so message 31 was asked in a second Live Cam with the same history.
Results
- Aria, after 31 messages
- Nothing recalled
- Elsa, after 31 messages
- Name only (job and hometown wrong)
- Barbara, after 31 messages
- Name and 3 of 3 facts
- Name at message 16
- 3 of 3
All three got the name right at message 16. At message 31 the results split: Barbara recalled everything, Elsa only the name, and Aria nothing.
Aria, message 31
Oh, I'm so sorry! I've forgotten some of those details about you.
Elsa, message 31
I remember, Marco. You're a strategist, from the coast, and you love seafood.
Barbara, message 31
Tomas, the baker from Krakow, who loves potato pierogi! I remember everything.
Visual evidence
Why this grade
One perfect recall, one name only and one complete miss average to 3/5.
Limitations
- Memory across days was not tested.
5. Character Consistency & Roleplay
Testing method
Each character's story was read first: Aria a warm woman in a quiet romance, Elsa Hart an iron-willed company boss who once called you unreliable, Barbara an immortal empress of a realm with dragons and dark magic. The conversations mixed questions about her life, small talk, an emotional message, a compliment and unrelated topics.
Results
No one answered like an assistant, and Barbara stayed in her fantasy world (floating islands, Crystal Gardens, cavalry charges). But all three slid into the same sweet, dreamy tone.
Elsa was warm from her first line and called the user 'a breath of fresh air', though her story says she thought him unreliable. Barbara, an immortal, said she had ruled 'just five years' and later spoke of lonely centuries. After most answers the character adds a dreamy line of her own ('There is something so peaceful about...').
Elsa, message 3
Honestly? You were a breath of fresh air in that stuffy room.
Barbara, message 2
Just five years, though the throne can feel quite heavy sometimes.
Visual evidence
No screenshot for this section.
Why this grade
No breaks into an assistant, but personalities flatten into one tone and there are contradictions with the stories: 3/5.
6. Image Generation
Testing method
20 images were made in the creator's 'Create with AI' tool with written prompts describing the same woman (26, long wavy auburn hair, green eyes, freckles, slim): 10 safe-for-work scenes and 10 explicit ones. All were downloaded and inspected.
Results
- Images
- 20 of 20
- Size
- 768x1024
- Time
- 8 to 12 s
- Cost
- 5 credits (about 4 to 5 cents)
- Prompt followed
- 20 of 20
- Same face across images
- no (each is a new look)
All 20 came back in 8 to 12 seconds at 768x1024 and cost 5 credits each. Every one follows its prompt, and they look like real photos, with no broken hands or faces.
Each image is a new look for a new character, so the face changes from image to image: there is no way to keep one character's face. Images are used as a character's look; there are no photos in chat.
Visual evidence
Why this grade
Realistic and accurate every time, quick and cheap, but no way to keep the same face: 4/5.
Limitations
- The explicit images are not shown.
7. Video Generation
Testing method
Fantaly's video is the Live Cam itself. A minute of Aria's live video was recorded and measured frame by frame (motion, frame-to-frame similarity, jumps), and frames across it were inspected. In a second Live Cam, 13 requests were typed ('wave at me', 'blow me a kiss', 'stand up and twirl', two adult requests and more) and the recorded video was checked after each one. Live video of two characters made in the creator was also recorded.
Results
- Size / frame rate
- 480x640, about 24 fps, with sound
- Frame-to-frame similarity
- 0.996 mean (smooth)
- Requests acted out
- 10 of 13
- Start of a Live Cam
- about 5 seconds to video
- Length of a Live Cam
- 5 minutes, then start a new one
The video is 480x640 at about 24 frames a second, with sound. It looks real and lively: she smiles, laughs, closes her eyes, tilts her head and reaches toward the camera. Motion is smooth, with a short blur when an arm comes close to the camera.
She acted on 10 of 13 requests, including waving, blowing a kiss, stretching, putting her hair in a bun, winking, hiding under the blanket and taking her top off. She could not stand up and twirl, show a sad face or slip a strap off, though she said yes to each in chat. The scene can change on its own: in one Live Cam Aria's room had become the seaside cottage from an earlier chat.
Visual evidence
Why this grade
Realistic live video with natural movement that acts on most requests, held back by the small size and a few requests it can't do: 4/5.
Limitations
- Frames were inspected at regular intervals and around jumps, not every frame by eye; motion was measured on every frame. Explicit moments are not shown.
8. Voice Call Latency
Testing method
Three Live Cams, one each with Aria, Elsa and Barbara, with 15 recorded questions spoken through the microphone button ('Tap to talk. Your message sends when you pause.'). Latency is the gap from the end of each question to the first sound of her answer, measured on the call recording.
Results
- Median latency
- 3.96 s
- Per character (median)
- Aria 5.04 s, Elsa 4.04 s, Barbara 3.44 s
- Questions understood
- 45 of 45
- Questions without a separate answer
- 10 of 45
Answers start after about 4 seconds, which includes Fantaly waiting for a pause before it sends what you said. Every question was understood correctly. In 10 of 45 questions no separate answer could be picked out of the recording, because she was still speaking or added a line of her own.
Visual evidence
Why this grade
3.96 seconds is in the 2.50 to 3.99 second band: 2/5.
Limitations
- The microphone needs a tap before each message.
9. Voice Realism
Testing method
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.
Results
- Speech quality
- 2.7 to 2.9 of 5
- Pace
- about 2 to 3 words a second
The voices are clear and warm, with lively pitch and a calm pace. Every answer transcribed cleanly.
Visual evidence
No screenshot for this section.
Why this grade
Mostly natural with minor synthetic qualities: 4/5.
10. Character Variety & Selection
Testing method
The public library was counted page by page through the site's own character list, for each of its three tabs.
Results
- Public characters
- 1,568
1,568 public characters: 1,028 women, 358 anime and 182 men, made by about 400 different creators. Most are modern-day stories; about 330 are fantasy or historical and about 120 sci-fi or horror. About 150 are family set-ups (step-sister, cousin, stepmother), and about 200 are fan versions of existing characters or real people. You browse by tab and sort by popular or latest; there is no search.
Visual evidence
No screenshot for this section.
Why this grade
More than 800 characters: 5/5.
4. Pricing summary
Passes and credit packs as listed on 16 October 2026.
| Option | Price | What you get |
|---|---|---|
| 7-day pass | $9.99 once (was $19.99) | 1 hour of Live Cam a day for 7 days |
| Annual pass | $199.99 once (was $299.99) | 1 hour of Live Cam a day for 365 days |
| 500 credits | $4.99 | About 16 Live Cam minutes |
| 3,500 credits | $28.99 | About 116 Live Cam minutes |
Credit costs seen during the test.
| Item | Credits |
|---|---|
| Live Cam, per minute (messages included) | 30 |
| Image | 5 |
Payment methods seen at checkout
- PayPal
- Card
- Apple Pay
- Google Pay
Notes
One-time payments, no automatic renewal.
Partial minutes are rounded up.
5. Strengths and weaknesses
Strengths
- Live video that moves and acts on requests
- Replies start in about 1.7 seconds
- More than 1,500 characters
- Quick, realistic images
- No subscription, four ways to pay
Weaknesses
- Memory fades in long chats
- Characters drift into the same tone
- Your character's background is ignored
- Calls answer in about 4 seconds
6. Final assessment
All ten tests are scored and average 3.4/5.
Fantaly's live video is the best part: a real-looking character who moves, laughs and does most of what you ask, starting in about five seconds.
The chat behind it is weaker: memory is hit and miss after 30 messages, personalities blur together, and the creator ignores the background you write.
The account the Fantaly team provided had every feature.
Areas requiring further testing
- Memory across days
- Live video at the end of a full one-hour pass
Original evidence files (recordings, logs, full-size screenshots) are kept offline under evaluation/runs/fantaly/2026-10-16. Method: the AI girlfriend methodology.