Trained models
Trained models are LoRAs the AI latches onto. They produce more consistent results than reference photos alone. 150 credits, takes 5 to 15 minutes.
Train one#
-
Gather your reference set. 5 to 20 photos for products, 10 to 30 for styles, 10 to 20 for faces.
-
Open /app/train and click Create New Model.
-
Pick the type. Product, Style, or Face.
-
Upload the references. Drag and drop. The trainer validates count and format.
-
Name and start training. 150 credits. Comes back in 5 to 15 minutes.
-
Use it in any prompt.
@your-handleand combine with Studios.
Status meanings#
- Training. The model is being prepared.
- Ready. Usable via
@in prompts. - Failed. Training did not converge. Credits refunded automatically.
Tip: Pick memorable handles. You will type these dozens of times.
My trained model drifts off-product#
Mention the model twice in long prompts (early and late). If it still drifts, retrain with a tighter, cleaner reference set.
Training failed#
Usually a reference set problem: too few images, mixed subjects, harsh lighting. Credits are refunded automatically. Clean up the set and retrain.