utility | Kling AI
Elements
Train the AI to recognize your character, object, or costume permanently — upload reference photos once, then reference them by name in any Kling video or image generation
utility
Custom Elements
Kling AI
Elements is your character and object library. Instead of uploading the same reference photos every time you generate content, you train the AI once — upload a few photos of a person, object, costume, or scene — and the AI learns what it looks like. From that point on, you just reference the element by name in your prompts and the AI knows exactly what to generate.
Two training modes are available. Image Reference lets you upload 1 frontal photo plus up to 3 additional angles — the AI learns the visual appearance and can reproduce it across any generation. Video Reference lets you upload a short video clip (3 to 8 seconds) — the AI learns both appearance and movement style, and can even extract the voice automatically if the person speaks in the clip.
Once an element is created, it works everywhere across the Kling ecosystem. In Omni Image, reference it as <<<element_name>>> in your template prompt. In Kling Video, reference it the same way. The element carries consistent identity across all generations — same face, same proportions, same distinctive features — without you needing to upload photos again.
Tags let you organize your library by type: Character, Animal, Item, Costume, Scene, or Effect. Creating elements is free — costs only apply when you use them in image or video generation.
Two training modes are available. Image Reference lets you upload 1 frontal photo plus up to 3 additional angles — the AI learns the visual appearance and can reproduce it across any generation. Video Reference lets you upload a short video clip (3 to 8 seconds) — the AI learns both appearance and movement style, and can even extract the voice automatically if the person speaks in the clip.
Once an element is created, it works everywhere across the Kling ecosystem. In Omni Image, reference it as <<<element_name>>> in your template prompt. In Kling Video, reference it the same way. The element carries consistent identity across all generations — same face, same proportions, same distinctive features — without you needing to upload photos again.
Tags let you organize your library by type: Character, Animal, Item, Costume, Scene, or Effect. Creating elements is free — costs only apply when you use them in image or video generation.
Best results
Clear Frontal Photo is Essential
The first image must be a clean, well-lit frontal shot — face looking straight at the camera with no obstructions. This is the primary reference the AI uses to learn the identity. Everything else builds on this image.
Add Multiple Angles
Upload up to 3 additional photos from different angles — side profile, three-quarter view, slightly above or below. More angles give the AI a fuller understanding of the face and body from every direction.
Video Reference for Movement and Voice
If you want the AI to learn how a character moves and sounds, use Video Reference mode. Upload a short 3 to 8 second clip of the person speaking or moving — the AI extracts appearance, motion style, and voice together.
Use Descriptive Names
Give your element a clear, memorable name — Maeva, Red Dress, Beach Scene — not element_001. You will type this name in prompts across other tools, so it should be easy to remember and type.
Create Once, Use Everywhere
An element works in Omni Image and Kling Video. Train your character as an element and you never need to re-upload reference photos — just type <<<element_name>>> in any prompt to include them.
Free to Create
Creating and managing elements costs nothing. You only pay credits when you generate images or videos that use the element. Build your entire character library without spending anything.
Guides
Elements
🛠️
Utility
Elements — Technical Guide
Train the AI to recognize your character, object, or costume permanently — upload reference photos once, then reference them by name in any Kling video or image generation