Keep a character consistent

Once a character is trained, tendre.AI keeps that same face and identity across as many images as you want. This relies on the LoRA you trained, backed up by InstantID.

Attach the LoRA

Open a new generation and add the character's LoRA from the model list. This is what locks the face; without it, each generation invents a new person even with a similar prompt.

Prompt around the character

Describe the scene, outfit and pose the way you normally would, and let the LoRA carry the identity. Change anything in the text (location, clothes, lighting, camera angle) and the face still comes out as the same person.

Stays consistent across outfits and poses

Because the LoRA encodes the face rather than a full image, changing the outfit or pose does not break the likeness. This is what lets you build a whole gallery of one character in different situations without retraining.

InstantID as a backup

tendre.AI also runs InstantID to reinforce the identity on shots where the LoRA alone drifts a little, especially on hard angles or unusual poses. Nothing to configure, it works alongside the LoRA automatically.

Tips for best results

Keep the LoRA weight in a moderate range so the face stays accurate without looking stiff or repeating a training pose. If a character keeps drifting, the fix is usually the source material, revisit your photo dataset or read Train a LoRA again before retraining.