How to Create Realistic AI Avatars (Full Guide)
Demonstrates a robust workflow for locking avatar facial consistency using split-frame character reference sheets before feeding multiple keyframes into a single 30-second multi-scene video generation.
What this lesson covers
Youri van Hofwegen demonstrates how to build a photorealistic AI avatar from a multi-angle reference collage and a split-frame character sheet in Higgsfield. He then shows how to generate keyframes across distinct environments and animate all scenes into a continuous 30-second shot using Seedance 2.5.
Key takeaways from the creator
AI-extracted notes, not independently verified product claims. Timestamp links let you check each point in the original video.
- 02:16 ↗
Using a multi-angle photo collage as an image-to-image reference enables the model to capture facial depth and structural details better than a single image.
- 02:41 ↗
Generating a split-frame character sheet containing both a full-body shot and a tight chest-up closeup creates a solid master anchor for subsequent generations.
- 03:11 ↗
Setting a pure white, empty background on character reference sheets prevents video generation models from confusing foreground subjects with background elements.
- 04:38 ↗
Prompting two contrasting light sources with different color temperatures (such as warm firelight and cool daylight) effectively separates the subject from the background.
- 06:28 ↗
Seedance 2.5 can ingest multiple reference location images to render automatic scene cuts and transitions within a single 30-second generation call.
- 07:18 ↗
Pacing on-camera dialogue for a 30-second video generation requires approximately 75 words in the prompt script to match normal speaking speed.
Workflow outlined in the video
- Upload a collage of personal photos from multiple angles into the image generation workspace.
- Generate a split-frame character sheet (full body on left, chest-up close-up on right) on an empty white background using detailed skin texture prompts.
- Use the character sheet as an image reference to generate keyframe stills across different environments, explicitly naming two distinct light sources for depth.
- Open the video workspace, select Seedance 2.5, set the duration to 30 seconds at 1080p 16:9, and attach all location stills as reference frames.
- Write dialogue directly into the prompt audio section, specifying that the lines are spoken on camera with matching mouth movements at a ~75-word pace.
Before you use this workflow
These notes describe the source video at its publication date. Model access, pricing, connectors and interfaces may have changed. Check the original source and the provider’s current documentation before installing an add-on, connecting an account or spending credits.
ReelStack has not independently tested this workflow. Preview one representative shot and check motion, continuity and output quality before applying it to a full production. No result, cost saving or model capability is guaranteed.
Explore the tools
How to interpret the numbers
16,318 views captured 25 September 2026. ReelStack recommendation score: 51/100. Formula: radar-v5. This is ReelStack’s calculation, not a YouTube rating or a measure of factual accuracy.
Show the calculation
- 65% performance: 0.77× views vs 9 other channel videos observed at a similar age. View-evidence factor 94% (views / (views + 1,000))
- 25% freshness: 96/100 with a 14-day half-life
- Momentum pending: collecting daily snapshots; its 20% weight goes to observed performance, not free points
- 10% engagement: 6/100; likes + 4× comments, smoothed with a 500-view neutral prior