SD 2.0 Omni Reference Training: Training

Train a reusable character from a reference photo. Once complete, reference the character in Omni Reference video prompts using @omni-character:<request_id> to generate videos featuring that character consistently.

📝

Overview

About this model

Train a reusable character identity from a single reference photo. The trained character can then be used across multiple Omni Reference video generations by referencing it with @omni-character:<request_id> in your prompt — maintaining consistent appearance, style, and identity across different scenes and motion types.

1Character Consistency: Create a recurring character for a video series or story, keeping their appearance identical across all generated clips.
2Brand Mascots: Train a brand character once and reuse them in unlimited video campaigns.
3Storytelling: Build a cast of characters and direct them in different scenes without re-uploading reference images each time.
💰

Pricing & Value

Cost analysis

muapiapp$0.50 per character

One-time training cost per character. Trained characters can be reused in unlimited Omni Reference video generations.

Fal.aiNot available

Character training for Omni Reference-style generation is not available on Fal.ai.

ReplicateNot available

Character training for Omni Reference-style generation is not available on Replicate.

* Competitor pricing is estimated based on similar model architectures and usage tiers.

⚙️

Technical Details

Configuration schema

Reference Imagestring

Reference photo of the character (JPEG/PNG/WebP). One clear face portrait works best.

Default Valuehttps://d3adwkbyhxyrtq.cloudfront.net/webassets/videomodels/seedance-v2.0-omni-reference.png
Character Namestring

Name for this character. Use this name with @omni-character:<request_id> in Omni Reference prompts.

Default ValueJuno
Descriptionstring

Optional description of the character.

Default ValueA brave explorer with piercing blue eyes
📖

Implementation Guide

Developer documentation

How to Use Omni Reference Train Character

  1. Upload a reference photo: Provide a clear, well-lit portrait photo of the character. A single face photo works best — avoid group shots or heavily obscured faces.

  2. Set a character name: Give the character a memorable name (e.g. 'Alex'). This name is for your reference only and does not affect generation.

  3. Submit and wait: Training takes approximately 20–60 seconds. You'll receive a request_id when the job completes.

  4. Use in Omni Reference: In any Omni Reference video prompt, reference the trained character with @omni-character:<request_id>. Example: @character:abc123ef walking through a neon-lit city at night.

  5. Combine with other references: You can combine character references with image, video, and audio references in the same prompt.

Common Questions

Frequently asked

How do I use the trained character in a video?

After training completes, use @omni-character:<request_id> in any Omni Reference video prompt. Replace <request_id> with the request_id returned by this endpoint. The character will be automatically included as an image reference.

What makes a good reference photo?

A clear, front-facing portrait with good lighting works best. Avoid group photos, heavy shadows, extreme angles, or images where the face is partially obscured.

Can I use the same character in multiple videos?

Yes. Once trained, you can reuse the character's request_id in as many Omni Reference prompts as you like — you only pay the $0.50 training cost once.

How long does training take?

Training typically completes in 20–60 seconds. The request will timeout after 5 minutes if training does not complete.