Train a model on your own clips and images so it learns a style, a product or a character and holds on to it from shot to shot. Bring the footage, pick a base model, compare what it learned, and call it through the same API as the catalogue.
01/Prepare your data
Upload the clips and still images that show what the model should learn: a visual style, a product, a character. Add files straight from your Library, too. Every item gets a caption, drafted for you and yours to edit, because the captions are how the model connects your words to what it sees.
02/Train
Choose the model to build on and start a run. Fine-tuning trains a LoRA adapter, a small set of weights on top of the base, instead of a whole new model, so the base keeps everything it already does well. Leave the settings at their defaults or set steps and learning rate yourself, and follow the run as it goes.
03/Compare
Render one prompt with the same seed at each checkpoint, next to the untuned base model, and see what the model has picked up and where it starts to overdo it. Then keep the checkpoint that looks right, which is not always the last one.
04/Use it
Your fine-tune gets a model id on your account. It appears in the playground and under My models, takes the same request as the model it was built on, and is called with the API keys you already have. Nobody outside your account can see it or use it.
Tell us what you want a model to learn and what you would build on it, and we will get you set up.
Requests go to hello@fluxion-sys.ai. In the meantime, every model in the catalogue is available today.