Douyin talking-head
A hook in the first three seconds, sentences under twenty characters, a clear call to action at the end.
- Counter-intuitive opener
- Second person, spoken
- Ends with an action
Windows 10 / 11 · Runs locally
AI short video
Give it a topic.
AI short video
Give it a topic.
It handles the rest
Script, voiceover, visuals, subtitles and the final vertical cut — one pipeline, start to finish. Visuals render on your own GPU: no stock subscriptions, no per-clip charges. Everything runs on your machine. Your script and video never leave it.
◆Script segment 3
✎Voice Yunxi
◧Visuals local GPU
Rendering · 62%9:16 vertical · 1080×1920
Four templates, applied in one click
Pick a category and the script rules, voice pace, visual style and subtitle layout are configured for you.
Novel recap
Third-person narration, a hook every ten seconds, and the last line stops right before the reveal.
- Opens mid-scene
- Truth saved for last
- Restrained tone
Product video
Pain point first, two or three concrete selling points, then a natural call to action without banned claims.
- 15 to 30 seconds
- Specific details
- Avoids banned words
Knowledge explainer
Starts from something you see in daily life, explains the mechanism by analogy, ends with an open question.
- Plain language
- Jargon gets a gloss
- Closes on a question
Why use it
It automates the two slowest parts: finding visuals and matching them to your script.
🔒
One topic to a finished video
Enter a topic and press generate. Script, voiceover, visuals, subtitles, background music and the final cut all happen automatically.
🌍
Visuals that match the words
Keywords are full scene descriptions rather than two-word search queries, so each image corresponds to the line being narrated.
📚
Batch production
Queue up to ten topics at once. Every task lands in the task manager where you can watch progress and download results.
🔌
China-based models built in
Kimi, DeepSeek and Qwen connect directly without a proxy. Local Ollama is supported too and costs nothing at all.
🗣️
Clean licensing
Subtitles use Noto Sans SC and background music is CC0 public domain, so finished videos are safe to publish commercially.
⚡
Automatic VRAM handoff
It releases VRAM before rendering and hands it back to the language model afterwards, so one GPU can serve both without manual steps.
FAQ
Does generating videos cost extra?
No. Visuals render on your own GPU and the voiceover uses the built-in Edge TTS, so both are free.
You only pay if you choose to plug in a cloud service such as premium TTS or a cloud text-to-video provider.
Will it work without a dedicated GPU?
Yes. Without the rendering engine it falls back to free stock footage from Pexels and Pixabay and still produces a video.
Stock footage matches the script less precisely, so some lines will not line up with what is on screen.
How big is the rendering engine?
About 8 GB, including the SDXL model and the CUDA runtime. It downloads from inside the app with a progress bar.
After that, rendering is offline and free — roughly three to five seconds per image.
Which language models are supported?
Kimi, DeepSeek, Qwen, Zhipu, MiniMax and MiMo connect directly from China. OpenAI, Claude, Gemini and local Ollama are supported too.
The first launch walks you through picking one and entering an API key.
Can I use the videos commercially?
Yes. The subtitle font is SIL OFL licensed and the background music is CC0 public domain, both safe to redistribute with your video.
If you swap in your own fonts or music, check the licence on those yourself.
Where are videos saved?
In the task folder inside the app data directory. The task manager can open that folder for you.
Does it upload my script or data?
No. Scripting, rendering, voiceover and editing all happen locally. Only when you use a cloud language model does the prompt go to the provider you picked.
With local Ollama, nothing leaves your machine at all.
Download AI Short Video Studio
Windows 10 / 11 · x64
AI Short Video Studio 1.0.0
288 MB · guided setup on first launch
Double-click to install. Windows may warn about an unknown publisher — choose More info, then Run anyway. SHA-256：9153e7264a444844b46532def6415d12646433cd720b92621babd3ff214aa1c7
SystemWindows 10 / 11 (64-bit)
Installer sizeAbout 288 MB
Rendering engineOptional, about 8 GB
GPUNVIDIA 6 GB or more (optional)