autotrend

AI-powered YouTube video assembler. Point it at a folder of clips/photos and a text script, get a 5–10 minute MP4 back.

autotrend ./media/ script.txt -o video.mp4

---

what it does

1. segments your script by paragraph into voiceable chunks
2. generates TTS audio for each segment (Microsoft Edge TTS — no API key needed)
3. assigns your media files round-robin to each segment, trimming or looping as needed
4. pads silence between segments automatically if the total would fall under 5 minutes
5. assembles everything into a single 1920×1080 MP4 via ffmpeg
6. optionally mixes in background music at a low volume

it has a full terminal UI: animated progress bars, media inventory table, segment plan table, per-stage spinners, and a final stats panel.

---

requirements

• python 3.11+
• uv — used to run the script with inline deps
• ffmpeg + ffprobe in PATH

# arch
sudo pacman -S ffmpeg uv

# debian/ubuntu
sudo apt install ffmpeg && pip install uv

# mac
brew install ffmpeg uv

python dependencies are installed automatically by uv on first run: rich, typer, edge-tts

---

install

curl -o ~/.local/bin/autotrend https://raw.githubusercontent.com/numbpill3d/autotrend/main/autotrend
chmod +x ~/.local/bin/autotrend

or clone and symlink:

git clone https://github.com/numbpill3d/autotrend
ln -s $PWD/autotrend/autotrend ~/.local/bin/autotrend

---

usage

autotrend [OPTIONS] MEDIA_DIR SCRIPT_FILE

arguments

argument	description
MEDIA_DIR	folder containing your video clips and/or images
SCRIPT_FILE	plain text file — the narration script

options

flag	default	description
-o, --output	autotrend_output.mp4	output file path
-v, --voice	en-US-AriaNeural	edge-tts voice name
-m, --music	—	background music file (any format ffmpeg accepts)
--music-vol	0.12	background music volume, 0.0–1.0
--min-dur	300	minimum target duration in seconds (5 min)
--max-dur	600	maximum target duration in seconds (10 min)
--ken-burns	off	slow zoom effect on images (cinematic, but slower to render)
--preview	off	show the segment plan without rendering anything
--keep-tmp	off	keep temp files after render (useful for debugging)

---

script format

write your narration as plain text. separate scenes/segments with a blank line:

This is the first scene. It will be read aloud and paired with your first media file.

This is the second scene. It gets the next media file, looping back around if you
have fewer files than paragraphs.

Short paragraphs get merged automatically. Very long ones get split at sentence
boundaries so nothing sounds rushed.

the tool targets 5–10 minutes. if your script is shorter, inter-segment silence is added proportionally. if it's longer, you'll get a warning but it'll still render.

---

examples

# basic
autotrend ./footage/ narration.txt -o my_video.mp4

# specific voice + background music
autotrend ./clips/ script.txt -o out.mp4 \
  --voice en-US-GuyNeural \
  --music lo-fi.mp3 --music-vol 0.10

# photos with ken burns zoom effect
autotrend ./photos/ script.txt --ken-burns -o cinematic.mp4

# preview the segment plan before committing to a full render
autotrend ./media/ script.txt --preview

# longer target (15 min)
autotrend ./media/ script.txt --min-dur 600 --max-dur 900

---

supported media formats

video: .mp4 .mov .avi .mkv .webm .m4v .flv

image: .jpg .jpeg .png .webp .bmp .tiff .gif

videos are preferred over images when both exist in the folder. all output is normalized to 1920×1080 @ 30fps, letterboxed/pillarboxed if needed.

---

voices

list all available TTS voices:

edge-tts --list-voices

some good defaults:

voice	style
en-US-AriaNeural	friendly, natural (default)
en-US-GuyNeural	male, clear
en-GB-SoniaNeural	british female
en-AU-NatashaNeural	australian female
en-US-JennyNeural	conversational

---

license

MIT