Checked November 12, 2026. I build bamabound.app, which turns one photo of one person into a vertical clip of them holding a red I'M BAMA BOUND banner. The test clip in this post is synthetic (a red box swaying over a green field, with a hum and some beeps), not a real person.
A question I keep seeing: "My Bama Bound video sounds way quieter than the clips before and after it. Can I make it louder without it distorting?"
The clip keeps the original soundtrack of the reference (FAQ), and that audio was never mastered for your feed. Once you trim it, add a caption track, or drop it into a slideshow, the level can end up anywhere. "Turn it up" is the wrong instinct. Measure it first, then apply the gain you need.
Two numbers that matter
- Integrated loudness (LUFS). How loud the whole clip sounds on average, weighted the way ears work. Streaming and social apps normalise playback around a target. Values around -14 LUFS are commonly quoted for music streaming. Platforms change this without announcing it, so treat -14 to -16 as a sensible range, not a spec.
- True peak (dBTP). The highest point the waveform reaches between samples, after reconstruction. Keep it at or below about -1 dBTP so the app's own AAC re-encode doesn't clip.
If you raise the gain until the loudness hits the target, the peaks go up by the same amount. Short sharp sounds (a beep, a clap, a shout) hit the ceiling long before the average is loud enough. That is the trade-off this whole post is about.
Measure: ebur128
ffmpeg's ebur128 filter prints a summary with integrated loudness (I), loudness range (LRA) and, with peak=true, true peak:
ffmpeg -hide_banner -nostats -i clip.mp4 -map 0:a:0 \
-af ebur128=peak=true -f null - 2>&1 | sed -n '/Summary:/,$p'
The script
loudness.py wraps that and adds a fix step. The fix step uses loudnorm in two passes: the first pass measures, the second applies one linear gain using the measured values. The video stream is copied, never re-encoded.
#!/usr/bin/env python3
"""Measure integrated loudness (LUFS), loudness range and true peak of a clip
with ffmpeg's ebur128 filter, and say how far it is from a target.
python3 loudness.py clip.mp4 [--target -14] [--fix out.mp4]
"""
import argparse, json, re, subprocess, sys
def measure(path):
r = subprocess.run(["ffmpeg", "-hide_banner", "-nostats", "-i", path,
"-map", "0:a:0", "-af", "ebur128=peak=true",
"-f", "null", "-"], capture_output=True, text=True)
if r.returncode != 0 or "Summary:" not in r.stderr:
sys.exit("no audio stream, or ffmpeg failed:\n" + r.stderr[-400:])
s = r.stderr.split("Summary:")[-1]
num = lambda k: float(re.search(k + r":\s+(-?[\d.]+|-inf)", s).group(1).replace("-inf", "-999"))
return {"I_LUFS": num("I"), "LRA_LU": num("LRA"), "TP_dBFS": num("Peak")}
def fix(src, dst, target, tp=-1.0):
# pass 1: measure with loudnorm, pass 2: apply linear gain with those numbers
r = subprocess.run(["ffmpeg", "-hide_banner", "-i", src, "-map", "0:a:0",
"-af", f"loudnorm=I={target}:TP={tp}:LRA=11:print_format=json",
"-f", "null", "-"], capture_output=True, text=True)
m = json.loads(r.stderr[r.stderr.rindex("{"):r.stderr.rindex("}") + 1])
af = (f"loudnorm=I={target}:TP={tp}:LRA=11:linear=true:"
f"measured_I={m['input_i']}:measured_TP={m['input_tp']}:"
f"measured_LRA={m['input_lra']}:measured_thresh={m['input_thresh']}:"
f"offset={m['target_offset']}")
subprocess.run(["ffmpeg", "-v", "error", "-y", "-i", src, "-map", "0", "-c:v", "copy",
"-af", af, "-ar", "48000", "-c:a", "aac", "-b:a", "160k", dst], check=True)
if __name__ == "__main__":
ap = argparse.ArgumentParser()
ap.add_argument("clip"); ap.add_argument("--target", type=float, default=-14.0)
ap.add_argument("--fix")
a = ap.parse_args()
m = measure(a.clip)
m["delta_LU"] = round(m["I_LUFS"] - a.target, 1)
print(json.dumps(m))
if a.fix:
fix(a.clip, a.fix, a.target)
n = measure(a.fix); n["delta_LU"] = round(n["I_LUFS"] - a.target, 1)
print(json.dumps(n))
-ar 48000 matters: loudnorm upsamples internally to 192 kHz, and without it you get a 192 kHz AAC track some phones refuse.
What happened on the test clip
The synthetic clip is 8 s, 720x1280, with a quiet 220 Hz hum and five short loud beeps. Measured on my box:
| Step | Integrated | True peak |
|---|---|---|
| Original | -18.0 LUFS | -3.3 dBTP |
| Fix, target -16 | -16.1 LUFS | -0.9 dBTP |
| Fix, target -14 | -15.2 LUFS | -0.9 dBTP |
Target -16 works: a 2 dB lift, and the peaks stay under the ceiling.
Target -14 misses by 1.2 LU. To get there, a 4 dB lift would push the beeps from -3.3 to about +0.7 dBTP, past full scale and well over the -1 ceiling. loudnorm won't break the ceiling, so it gives up on linear gain and compresses. You get a slightly squashed track that is still short of the target. Running plain one-pass loudnorm=I=-14 (dynamic mode) gave -15.5 LUFS, which is no better.
So the honest answer is that this clip can be about 2 dB louder without distorting, not 4 dB. If you want more, you need a limiter in front of it, which changes how the beat feels. For a meme clip that's usually not worth it.
A practical routine
- Export your edit.
python3 loudness.py edit.mp4 --target -16- If
delta_LUis worse than about -2, run it with--fix out.mp4and check the second line of output. - If the fixed true peak is above -1, lower the target, not the ceiling.
FAQ
Will the app normalise it anyway? Many apps turn loud uploads down. Fewer turn quiet ones up. A quiet clip next to loud ones just sounds quiet.
Does this touch the picture? No. -c:v copy keeps the video bit-for-bit.
Can I change the soundtrack on bamabound.app? No. The preset keeps the original soundtrack (FAQ). Swap it in your editor if you want, as long as you have the rights to the new track.
What does a clip cost? On the live pricing page today: 480p $7.98, 720p $9.98, 1080p $16.98, 4k $24.98 for one generation, tax added at checkout, no subscription. Resolution changes the picture, not the audio, so for this problem the cheapest tier behaves the same as 4k.
Top comments (1)
Measure first, then apply one linear gain is the right order, and copying the video stream matters since people often re-encode the picture just to fix the audio. Two gotchas in the second loudnorm pass: if the target would push true peak past the ceiling, loudnorm falls back to dynamic mode, so check normalization_type in its JSON output and re-measure the result with ebur128 instead of trusting the target; and loudnorm resamples to 192 kHz internally, so add -ar 48000 on the output or some players and editors get an odd file. For clips with sharp beeps or claps, a gentle alimiter before the gain step usually lets you reach -14 without the peaks forcing you to stop near -18.