Visual DNA

Visual DNA is Kolbo.AI's character-consistency system. Build a profile once for any character, product, or environment, then every model on the platform uses it automatically. It is free to create, unlimited on every plan, and you can combine several profiles in a single generation.

Overview

The hardest problem in AI content is consistency. You get a great character in GPT Image 2 or Nano Banana 2, then cannot reproduce them tomorrow: different face, different outfit, different vibe. Even tools that accept reference images tend to drift after a few generations.

Visual DNA is built to prevent that drift. It is not a single model, it is a smart approach that uses every underlying model better and improves as those models improve. Build a profile once and reuse it across your whole library of work.

Free and unlimited

Create as many Visual DNAs as you want on every tier. There is no per-DNA charge, no creation fee, and no monthly cap. Your characters, products, and brand environments: build a library and keep growing it.

How to build a Visual DNA

Open the Visual DNA panel in any image or video tool (left sidebar), then add:

1. Reference images - upload up to 4. The more angles and contexts, the more reliable the consistency.
2. Character sheet (optional but powerful) - a multi-angle composite of the subject. Kolbo uses this as the canonical reference for high-fidelity generations.
3. Reference video (optional, a few seconds) - for models that support video reference, this provides motion patterns, gestures, and body language.
4. Voice sample (optional) - for character DNAs used with voice-capable tools, this provides voice consistency in audio outputs.

Saving the profile takes seconds. From there it appears in every tool that accepts references.

Works with every image and video model

One Visual DNA, used across the platform:

• Image models: GPT Image 2, Nano Banana 2, Seedream, and every other image model
• Video models: Seedance 2, Kling, Veo, Sora, and the rest
• Elements generation: locks the character into multi-element compositions
• Creative Director: multi-scene generations stay character-consistent end to end
• Lipsync and voice-capable video: uses the voice sample if attached
• Video Editing tools: character replacement in existing footage

You pick the DNA from a dropdown and the model picks up the references automatically. No re-prompting, no re-uploading.

Combine up to 8 in one generation

Need two characters in the same scene, a product held by your brand mascot, or a character in a specific environment? Select multiple Visual DNAs in the same generation:

• Image generations: combine up to 8 Visual DNAs in one prompt
• Video generations: combine up to 3 Visual DNAs in one shot

This is what makes Visual DNA a system rather than a single feature. Compose entire scenes from your library and the characters, products, and environments all stay consistent across every model and every generation.

Reference by @name

In project chat and on the Cast & Assets panel, every Visual DNA has a mention name. Reference it with @name (for example "put @Sophia on a rooftop at dusk") and Kolbo knows exactly which profile to apply. Pair it with Moodboards, which are referenced by #name, to combine a subject with an art direction in one instruction.

Video and audio references

Newer models accept video and audio as references, not just images, and Visual DNA carries those automatically:

• If a DNA has a reference video attached, models that support it use it for motion and body language.
• If a character DNA has a voice sample, voice-capable models use it for the character's voice.

Attach the references once when you build the DNA and every supported model just works.

Shareable profiles and team libraries

Visual DNA profiles can be shared via link. Send a profile to a teammate, client, or collaborator and they can generate with the same character. Teams build shared libraries of approved characters and products that everyone in the organization uses, which is essential for keeping brand work consistent across every deliverable.

Tips

• Upload varied reference angles and contexts; more coverage means more reliable consistency.
• A character sheet gives the highest-fidelity results, so add one when exact identity matters.
• Let the DNA carry identity. Avoid re-describing a character's appearance in your prompt, since that competes with the saved profile and causes drift.

Related