Images, video, and speech
Turn these on in the init wizard, or later in
yourai.toml followed by yourai rebuild. Each one has its own model,
chosen with a Recommended pick for your computer:
- Images —
[image], with stable-diffusion.cpp (SD-Turbo for speed, FLUX.1 schnell for quality). - Short videos —
[video], with Wan2.1. Needs a GPU and takes minutes per clip. - Speech to text —
[speech_to_text], with whisper.cpp. - Text to speech —
[text_to_speech], with Qwen3-TTS.
The engines start the first time you use them and unload after
server.idle_unload_minutes (10 by default) without use, so they only take memory
while you need them. Only one of image or video is loaded at a time. Files are saved in the
project's outputs/ folder.
Example — Duy makes a slide picture and listens to a summary
$ yourai config set image.enabled true
$ yourai config set image.model sd-turbo
$ yourai config set speech_to_text.enabled true
$ yourai config set text_to_speech.enabled true
$ yourai rebuild
$ yourai image "a cute cartoon plant cell diagram"
Creating the image (the first one also loads the model)…
✔ saved outputs/images/a-cute-cartoon-plant-cell-diagram-1793519664.png
• took 8s
$ yourai speak "Photosynthesis turns light into chemical energy."
✔ saved outputs/audio/photosynthesis-turns-light-into-chemical-energy-1793519668.wav (4.5s)
$ yourai transcribe lecture-week3.wav --out lecture-week3.txt
✔ wrote lecture-week3.txt (21.3s)
$ yourai video "waves on a beach at sunset" --frames 9 --steps 4
✔ saved outputs/videos/waves-on-a-beach-at-sunset-1793519820.webm
Useful options: image --size 768x512 --n 2 --seed 7 --negative "text",
video --frames 33 --fps 16, speak --language de --out hi.wav, and
transcribe --language vi. Run yourai <command> --help for all of them.
yourai status shows which engines are loaded and how much memory they use. The
same features are available to apps through the API and to the
chat agent.
Questions or problems? support@bdtren.com · Contact page.