01
Style
Train a character, product, edit style, or visual habit without setting everything up from scratch each time.
Train it yourself
AITK Studio
Turn your own images, clips, or audio into LoRA and LoKr adapters you can reuse. Run the job on your machine, check the samples, and keep the result.
The basic idea
A folder of references can become a model you come back to.
Why use it
Teach it what you mean.
Use the UI for the parts you do often. Open YAML when a run needs exact settings. Your dataset does not have to leave your machine.
02
Motion
Use Wan and LTX when the thing you want has timing, not just a good still frame.
03
Sound
Ace Step support gives audio experiments a place in the same toolkit.
04
Control
Start with presets. Then change phases, optimizers, checkpoints, and encryption when the job gets serious.
Model range
Models you can actually train.
Image, edit, video, and audio support live in one place. Choose a tab, pick a family, and get to work.
- FLUX.1 / FLUX.2-dev
- FLUX.2 Klein
- AsymFLUX.2
- Flex.1 / Flex.2
- Chroma / Zeta Chroma
- Lumina Image 2.0
- Qwen-Image
- GLM-Image
- i1-3B
- HiDream / HiDream-O1
- PRX Pixel
- Ideogram 4
- OmniGen2 / Z-Image
- SDXL / SD 1.5
- ERNIE-Image
- Nucleus-Image
- Juggernaut Z
Actual workspace
Here's the real UI.
No fake product mockups here. These are the screens you use to set up jobs, inspect samples, and keep configuration close.
Edit captions, tune object boxes, and move through dataset images without leaving the studio.
Build Ideogram prompts with structured controls, canvas elements, palette choices, and live JSON preview.
Generate samples while you work, then compare what changed before you move on.
Set the model, adapter, dataset, optimizer, and run profile from one place.
How it goes
Bring data. Train. Reuse.
The first run can be simple. The next one can be tuned. Either way, you end with an adapter you can use again.
Bring the dataset
Add images, video, or audio with the captions and notes that explain what matters.
Choose the target
Pick a model family, adapter type, preset, and phase plan that fit the result you want.
Check the run
Watch checkpoints, samples, and loss while the worker does the slow part.
Keep the result
Export the adapter, generate with it, and use it in the rest of your setup.
Install
Install it locally.
You will need Python 3.10+, git, and an NVIDIA GPU. Blackwell / RTX 50-series users should use the Blackwell torch requirements file.
Runs on your GPU
Opens at localhost:8675
Use the UI or edit YAML
Linux setup
git clone https://github.com/BadAtCaptchas/AITK-Studio.git
cd AITK-Studio
python3 -m venv venv && source venv/bin/activate
pip3 install -r requirements_torch_legacy_cu128.txt
pip3 install -r requirements.txt