DEVELOPER DOCUMENTATION
Custom Character LoRA
One character identity across scenes and content formats.
Custom Character LoRA gives a character a dedicated visual identity that can be used across a content workflow. The service is available with a platform deployment or as a standalone project.
One character, any scene
The aim is a recognizable face and body across scenes, outfits and lighting. The character gallery shows Amora in four different settings.
A trained identity is one component of the generation workflow. Model choice, reference inputs, scene prompts and review still affect the final result.
Example training dataset
A character dataset can use approximately 50 curated images, for example:
| Frame type | Example share |
|---|---|
| Portrait | 50% |
| Full body | 30% |
| Mid-body upwards | 20% |
Vary angles, lighting, expressions and composition while keeping the intended identity consistent. The final dataset and training approach are tailored to your character and selected model.
Service workflow
- Agree the character brief and intended uses.
- Prepare an AI-designed or approved creator-based reference set.
- Curate the dataset and review identity consistency.
- Train and evaluate the dedicated character model.
- Configure generation recipes for agreed scene types.
- Connect the approved workflow to your content operations.
Content applications
Use the character identity in photo sets, generated clips, paid-content catalogs and social-content workflows. Video generation may require a reference image produced by the character workflow rather than accepting a LoRA directly.
Talk to our team about your architecture, character and launch.
Discuss your integration ↗