I Made a Viral AI Love Story (500M Views) — Steal My Prompts

What this lesson covers

Adil from Higgsfield details the end-to-end production pipeline behind a viral narrative AI short film generated using Claude and Seedance 2.5. He covers practical character consistency hacks using hybrid photo-asset sheets, location staging selection, and physics-focused prompting to fix AI motion issues.

Key takeaways from the creator

AI-extracted notes, not independently verified product claims. Timestamp links let you check each point in the original video.

1. 03:35 ↗

   Pre-structure project directories with scene-by-scene asset folders in Cinema Studio before generating to maintain order across hundreds of iterations.

2. 05:00 ↗

   Fix generic 'lookalike' faces by generating an AI costume sheet, erasing the generated head, and compositing the real subject's photo directly into the portrait reference panel.

3. 06:18 ↗

   Select background locations based on functional staging space (clearing lanes for sprinting or physical collisions) rather than visual complexity that turns into generation artifacts.

4. 08:15 ↗

   Prevent video models from rushing subtle romantic beats into hugs or kissing by breaking down emotional encounters into granular physical mechanics like momentum, shoulder contact, and planted foot turns.

5. 08:31 ↗

   Stop background extras from freezing like static NPCs by explicitly prompting every peripheral character to be mid-action in every frame.

Workflow outlined in the video

1. Set up dedicated project folders for each scene, character, and prop before running any generation batches.
2. Generate a costume and body silhouette sheet in Cinema Studio, then paste the real person's photo onto the face panel to build a hybrid character reference.
3. Batch generate location plates and filter for environments with wide paths and clean staging areas that leave room for actor motion.
4. Translate narrative scene beats into Claude-assisted prompts that describe physical weight, contact points, and momentum instead of high-level emotional concepts.
5. Add explicit instructions for background crowds to remain in continuous motion across all frames.
6. Generate final shots in Seedance 2.5 at 1080p, feeding the hybrid character sheets and location elements.

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

169,058 views captured 27 September 2026. ReelStack recommendation score: 42/100. Formula: radar-v5. This is ReelStack’s calculation, not a YouTube rating or a measure of factual accuracy.

Read our methodology and limitations →