Train a LoRA: teach the AI a character

A LoRA is a small file trained on a character so it keeps the same face and body across every generation afterward. Open Characters > My characters, pick a character and choose LoRA training, or create a new character straight from the training screen.

1. Prepare your photos

Gather 10 to 15 real photos of the subject: varied angles, expressions and a couple of outfits, face clearly visible, no other people in frame. See Prepare your photo dataset for what makes a clean set.

2. Set the trigger word and type

Import the photos, then check the trigger word (the word you will type in prompts to call this character, filled in from the name by default) and the type (Woman, Man, Person, Creature, Animal, Object, or Custom). The type anchors training to the right family, which matters for non-human subjects.

3. Choose Local or Cloud, then train

Pick Local (your own GPU, about 30 to 60 minutes, pauses TendreEngine to free VRAM) or Cloud (runs on tendre.red's servers for credits, about 15 to 20 minutes, your PC stays free). Press Start training and let it run in the background.

4. Use it

Once training finishes, the LoRA is ready in the LoRA generation tab with its trigger word. Add it to a prompt and the character's identity carries through every scene, outfit and pose.

For the next steps, read Character consistency and Character duo to place two trained characters together.