Reference Photos

What are reference photos?

Upload photos so the AI generates images that preserve your likeness and style across every scene. Without a reference, each scene shows a different person.

Melodious uses Gemini 3.1 Flash Image with dedicated character and object reference slots for high-fidelity consistency.

How to upload

1. Click + in the input bar
2. Select Photos
3. Choose your images

Your photos appear in the conversation and are used for all image generation.

Types of references

You can upload up to 14 reference images per project:

| Type | Limit | Best for | |------|-------|----------| | Character photos | Up to 4 | Photos of yourself or band members. The AI preserves facial features, skin tone, hair, and clothing across all scenes. | | Object/style references | Up to 10 | Album artwork, mood boards, screenshots from videos you like, specific props or settings. The AI matches the visual style and includes objects with high fidelity. |

You don't need to label which is which. The AI understands from context whether an image is a person reference or a style reference.

Tips for character photos

• Use a clear, well-lit portrait with your face clearly visible
• Front-facing works best for likeness preservation
• Multiple angles (front + side profile) improve consistency
• For bands: upload one separate photo per member (up to 4 members)
• The AI places each person in the right scenes automatically

Solo photos vs group photos

| Photo type | How the AI uses it | |---|---| | Solo portrait (one person) | The AI can place this person in specific scenes independently. Best for individual character control. | | Group photo (multiple people) | The AI reproduces the entire group together. It cannot extract individual people from a group shot. |

For best results with multiple people: upload separate solo photos of each person rather than one group photo. This lets the AI put each person in the right scenes (e.g., a solo verse scene vs a full band chorus scene).

If you only have a group photo, the AI will use it for scenes where the whole group appears together.

Tips for object and style references

• Objects/props (instrument, car, specific location): included in scenes where they appear
• Album artwork or cover images: set the overall aesthetic across scenes
• Screenshots from music videos you admire: guide the visual mood
• Specific settings (neon city, forest, studio): the AI includes them in matching scenes

Smart scene assignment

The AI director analyses each reference you upload and decides where it belongs in the storyboard:

• A solo portrait of the lead singer appears in their solo scenes
• A guitar photo appears in performance scenes
• A neon city backdrop appears in scenes with that setting
• A group photo appears in ensemble scenes

You don't need to specify this manually. The AI reads your creative vision and matches references to scenes intelligently. If you want to adjust which references appear in which scenes, just ask in the chat.

When can I upload?

You can upload at any point during the conversation. References accumulate across messages, so you can add more as the conversation progresses.

How it works technically

1. You upload reference images (solo portraits, group photos, objects, style refs)
2. The AI analyses each image and classifies it (character / group / object / style)
3. When creating the storyboard, the AI assigns specific references to each scene based on the scene content
4. During keyframe generation, only the relevant references are sent to Gemini 3.1 for that scene
5. Each keyframe becomes the first frame of the video clip, carrying the reference through to the video