Turbo Studio

Images in seconds, video clips in about a minute, all on your own Mac. Drop any picture in as inspiration. No cloud, no account, no queue.

View on GitHub Quick start Say hello

Typing a prompt, generating an image, then using it as the inspiration for a video clip. Recorded on a 16 GB M1 Pro.

Prompt and settings on the left, preview and history on the right. Drag an image anywhere onto the window.

What it does

Text to image

SDXL-Turbo in 1 to 4 steps, 512² up to 1024². About five seconds per picture.

Text to video

AnimateDiff-Lightning, 8 to 32 frames at 512². About a minute for a 16-frame clip.

Inspiration image

Drag, browse or paste an image. It seeds image-to-image for pictures and steers videos through IP-Adapter, each with its own slider.

Preview, history, save

Looping playback, a strip of everything you made this session, save as PNG, JPEG, MP4 or GIF, and one click to reuse a result as inspiration.

Fits in 16 GB

One model in memory at a time, fp16 throughout, and a chunked self-attention processor that keeps 16-frame clips from running out of memory on Apple's MPS backend.

Yours

MIT licensed. Models download once into a plain folder you can see, never into a hidden cache.

Quick start

git clone https://github.com/ShimonNavon/turbo-studio.git
cd turbo-studio
scripts/setup.sh          # .venv, install, download the models (~15 GB)
.venv/bin/turbo-studio    # opens the window

scripts/make_app.sh --install /Applications   # optional: a double-clickable Mac app

Needs macOS on Apple Silicon (tested on an M1 Pro with 16 GB), Python 3.10 to 3.14, and about 15 GB of disk for the four models. The README covers manual installs, configuration and troubleshooting.

Under the hood

Model	Role	Size
SDXL-Turbo	images, fp16	6.9 GB
epiCRealism	SD1.5 base the video model runs on	4.3 GB
AnimateDiff-Lightning	4-step motion adapter	0.9 GB
IP-Adapter	lets an image steer the video	2.6 GB

Each model carries its own license on its model card; SDXL-Turbo is non-commercial unless licensed from Stability AI.

Let's build this together

Turbo Studio was made for one Mac and would love to work on yours. NVIDIA cards, Intel Macs, Linux boxes, 8 GB machines, UI ideas, new engines: all of it helps. The good first issues are scoped to be approachable, Discussions is open for questions and show-and-tell, and CONTRIBUTING.md has the two-minute setup.