Train a LoRA: teach the AI a character

A LoRA is a small add-on you train yourself so the AI learns one specific person or character. Once trained, you can place that same identity in any scene, outfit or pose, with a consistent face across every render. Training happens on your own GPU and the photos never leave your machine.

1. Prepare your photos

Gather 10 to 20 images of the same subject: varied angles and expressions, good light, the face clearly visible, no other people in frame. Variety matters more than quantity. See Prepare your photo dataset for the details that make or break a LoRA.

2. Create the LoRA and import

Open the training panel, create a new LoRA, give it a name, and drop your photos in. tendre.AI handles the captioning and sizing for you.

3. Train

Press Train and let it run. The duration depends on your GPU and the number of images; it runs unattended. Training uses the full GPU, so it is best not to run other heavy generation at the same time.

4. Use it

When training finishes, your LoRA appears in the model list. Add it to a generation and prompt as usual with the name you gave it. The subject now stays consistent across scenes, outfits and camera angles.

To go further, read Character consistency and Character duo to place two trained characters in the same image.