Proof
Measured, not claimed.
We measured 14 public sample videos from Arcads, Breakout Clips and Creatify. Below is what came back, the bar every Apex Flow AI ad has to clear before it is shown to anyone, and the commands to repeat every number yourself.
Their samples, our bar
Measured values in competitor sample clips next to the requirement for an Apex Flow AI ad| Measured | In the 14 sample clips | Every Apex Flow AI ad must reach |
|---|
| Picture size (short side) | Best 936 px. 0 of 14 clips reach 1,080 px. | 1,080 px or more |
|---|
| Soundtrack present | 7 of 14 clips carry any sound at all. | Designed voice, effects and music on every ad, each on its own recorded track |
|---|
| Loudness | −23.4 to −11.5 LUFS across the 7 clips with sound. 1 lands in our window. | −15.5 to −12.5 LUFS |
|---|
| Loudest moment | Up to 0.0 dBFS (full scale). 2 of 7 peak above −1.5 dBFS. | −1.5 dBFS or lower; true peak −1.0 dBTP or lower |
|---|
| Lips follow the voice | 4 clips show a talking face. Best score +0.39; 1 of 4 pass the basic lip check and 0 reach our bar. | Beat the best competitor: +0.39 or more over a shuffled voice, on every talking ad, automatically |
|---|
The right-hand column is what our release gate enforces: an ad that misses any of it is never shown. It is a requirement, not a boast. A film’s own measured numbers appear further down only after that film has cleared the gate.
How many characters you can pick from
12,699 different character designs, every one with its own file, across 62 families in 11 style groups, counted from the same list the Universe page loads. The largest library any competitor states on its own site is 2,000+.
Their figures are what each vendor states about its own avatar library, read on 2026-09-18. Ours counts character designs you can choose from. A design is a starting point: each one is shown as a design until a film with that character has cleared every check.
Lips, clip by clip
Of the 14 clips, 4 show a talking face. The score is how much better the mouth follows the real voice than it follows the same voice shuffled. Our bar is +0.39, set just above the best clip here, so an ad that clears it beats every competitor clip we measured.
- Arcads
ai-video_1.webm−0.28
fails the basic check - Arcads
ai-video_2.webm+0.04
fails the basic check - Arcads
header_1.webm+0.02
fails the basic check - Creatify
blond_lady.mp4+0.39
passes the basic check, below our bar
The check is not built to fail everyone: Creatify’s real talking avatar passes the basic lip check, and our bar sits just above it. Breakout Clips’ templates are short silent animations with no speech to match.
Repeat it yourself
Download any sample from the vendor’s own site and run the same commands. Measured 2026-09-19.
- Size
- ffprobe -v error -select_streams v:0 -show_entries stream=width,height -of csv=p=0 clip.mp4
- Loudness and loudest moment
- ffmpeg -i clip.mp4 -vn -af volumedetect,ebur128=peak=true -f null - (read "I:" and "max_volume" at the end)
- Lips
- InsightFace 106-point landmarks give the inner-lip opening on every frame at 12 fps. The same clip's voice gives a loudness curve. Both are smoothed over 0.4 s and correlated within ±0.25 s. The result is compared with the same voice shuffled, so a mouth that moves for no reason scores the shuffle level.
Every clip and its numbers
| Vendor | Clip | Size | Seconds | LUFS | Peak dBFS |
|---|
| Arcads | ai-video_1.webm | 600×1068 | 11.6 | −23.2 | −9.2 |
| Arcads | ai-video_2.webm | 600×1068 | 11.6 | −18.3 | −4.2 |
| Arcads | create-your-own-ai_1.webm | 600×1066 | 4.8 | no sound | no sound |
| Arcads | header_1.webm | 936×1666 | 11.6 | −11.5 | 0.0 |
| Breakout Clips | bakery-example.mp4 | 720×720 | 5 | no sound | no sound |
| Breakout Clips | coffee-example.mp4 | 640×640 | 5 | no sound | no sound |
| Breakout Clips | how-it-works.mp4 | 800×1000 | 49 | no sound | no sound |
| Breakout Clips | squirrel-example-2-720p.mp4 | 720×720 | 6.2 | no sound | no sound |
| Breakout Clips | trex-example.mp4 | 720×720 | 5 | no sound | no sound |
| Creatify | blond_lady.mp4 | 704×720 | 6.3 | −16.9 | −1.8 |
| Creatify | ngl.mp4 | 1080×720 | 20.4 | −15.0 | −0.2 |
| Creatify | podcast.mp4 | 1080×720 | 35.1 | −23.0 | −9.2 |
| Creatify | product_car.mp4 | 720×576 | 5 | no sound | no sound |
| Creatify | ugc_lady_3.mp4 | 1080×720 | 8.5 | −23.4 | −7.4 |
What stops a bad ad from shipping
Every ad passes 49 separate checks covering picture, sound, captions, search readiness, faces, style and lips. A check that has never been seen to fail is worthless, so each one is tested against a film broken on purpose in exactly the way it should catch. The self-test runs 73 such cases and is green for the exact checks in use. Change one check and the green mark is void until the whole self-test runs again. A person also looks at the finished ad and signs the review before it can appear on this site.
Ads that have cleared the gate
Frank Dreams Up An Ad
53.7 seconds · 55 of 55 checks passed
- 1080x1080 (need at least 1080x1080)
- -14.1 LUFS, peak -2.4 dBFS
- mouth moves (movement 0.022, open 47%) and matches the voice at +0.49 against +0.02 for a shuffled voice (need >= +0.20 over shuffle and >= 0.30); open while speaking 59% and closed in silence 93% (balanced 0.76, need >= 0.68)
- picture cuts vs designed cuts: worst 40 ms off, 19 cuts (limit 102 ms)
- picture cuts 33 ms from the music's accent grid (median), 100% within 100 ms; pulse 156.4 bpm
- narrator over everything else: median 24.0 dB, quietest tenth 11.3 dB (need >= 14 and >= 7)
- music 12.4 dB under the narrator in the delivered file (need >= 8.5); already down at 23 of 23 first words
- mono fold loses 0.0 dB (limit 2.5), L/R correlation 0.99
- quiet-passage noise -120 dBFS (limit -50)
- longest silent gap between phrases 0.70s (limit 1.6)
What this does not show
These are each vendor’s own public sample clips, picked by them to look good, not a typical customer’s output. 14 clips is a small sample. The lip check compares mouth opening with voice loudness, which catches a mouth that ignores the voice but cannot judge vowel shapes. Vendors change their products; a number here is a measurement of that clip on that day.
See the formats in action → · Make yours in the Studio →
