Character Consistency

Overview

Character consistency is one of the hardest problems in AI video generation. Melodious AI solves this using Gemini 3.1 Flash Image with dedicated character reference conditioning.

How it works

1. You upload up to 4 character reference photos (different angles help)
2. The AI creates scene descriptions from your vision, the song's mood, and lyrics
3. During image generation, all your reference photos are sent alongside each scene prompt
4. Gemini 3.1 generates images matching the scene while preserving your exact likeness
5. Each keyframe becomes the first frame of the video clip, carrying your appearance through

With vs without reference photos

| | Without reference | With reference | |---|---|---| | Character appearance | Different person each scene | Consistent across all scenes | | Facial features | AI-generated, varies | Preserved from your photo | | Best for | Abstract/landscape videos | Videos featuring a specific person |

Multi-reference support

You can upload up to 14 reference images total:

• 4 character references for maintaining likeness (yourself, band members)
• 10 object/style references for visual style, props, and settings

All references are used for every scene in the storyboard, so your appearance stays consistent across the entire video.

Scene regeneration

If a specific scene doesn't look right, you can regenerate just that scene's keyframe. Hover over the scene thumbnail and click the regenerate button. Your reference photos are applied again with a fresh generation.