🧠 Built with AI: AI Video Generation with SkyReels | 888 Day 7

Cindy Adem
May 24
3 min read

Challenge: Create a fully coherent, emotionally grounded AI-generated video using a real childhood memory; with zero code, just smart prompting.

Tools Used:

SkyReels (AI video generation)
ChatGPT (Prompt development and testing)
My mind and memory (a powerful combination)

Objective

On Day 7 of my 8 products in 8 days challenge, I set out to demonstrate the power of prompting in AI video generation by recreating a vivid, layered scene from my childhood.

This wasn’t just a test of AI tools; it was a test of how far creative technology has come in allowing ordinary users to translate simple prompts into consistent visuals.

The scene: A skinny Black girl (me) in an African suburb, home from school on a school day. My “mission” that morning: to catch birds using a bucket trap, a stick, and a long string that led to my hiding spot.

The tension. The cleverness. The ethical ambiguity (protein or kite?). The joy and anticipation wrapped into a single morning. That was the story I wanted to bring to life.

Method 1: Piece-by-Piece Prompting

I started by feeding SkyReels the story bit by bit, breaking it down into “frames” or short prompts to maintain control and consistency.

I assumed this would help with character persistence, scene matching, and overall pacing. I used ChatGPT to break the story into well-scoped visual prompts while maintaining tone and continuity.

The result?

Each scene looked decent on its own.
But character consistency broke. Backgrounds shifted.
The little girl changed appearances; from her face to her clothing (or lack thereof).
Most disturbingly, the child character appeared unclothed which made me uncomfortable.

✅ Pros: More control over detail in each moment ❌ Cons: Scenes felt stitched, not seamless; characters were not consistent

❌ Major Fail: Misrepresentation of a child in sensitive visuals

Method 2: Full Prompt, Single Shot

For round two, I went all in. I pasted the entire narrative (the full text below) into a single prompt.

“It takes place when I was young, home because my folks couldn’t afford my fees (but happy because I hated school anyway). I'm approx 7-8 years old. Skinny Black girl with cornrows catching birds…”(Full story continues…)

My hope was that SkyReels would use this as a screenplay seed, breaking it into coherent scenes while preserving tone, character, and continuity.

The result?

SkyReels broke the story down, yes.
But it introduced new inconsistencies in visual tone, character attire, and geography.
The pacing felt robotic- not the fluid, emotional progression I’d envisioned.

❌ Unexpected Outcome: No scene-to-scene continuity, tone variance ✅ Pro: Slightly smoother transitions than piecemeal

🧠 Key Takeaways (a Product Eye on AI)

Even with clear vision, sharp prompting, and tool fluency, the results were inconsistent and unreliable. And that’s the real story here.

As someone within the target audience for these tools, prompt-savvy, non-programmer, creative - I hit wall after wall. These weren’t just bugs. They were UX gaps that break trust and creativity.

What This Reveals About GenAI Today:

Prompting ≠ Predictability: Even the most beautifully structured prompt can’t override a tool’s current training limitations.
Visual AI Needs Guardrails: Especially when rendering children. Accuracy and appropriateness must be non-negotiable.
Continuity Is Still an Afterthought: Most genAI tools are built to wow in the short-term, not tell stories that unfold over time with consistency.
Real Users = Real Edge Cases: Tools like SkyReels may perform fine with generic fantasy prompts. But try feeding it something personal, culturally specific, and emotionally layered, it cracks.

Why This Still Matters

To me this was not a failure. It was a stress test of AI’s creative edge, and the results are lessons worth documenting.

Because as AI tools evolve, the goal isn’t just speed; it’s sensitivity, continuity, coherence, and care.

This was Day 7 of my #888Challenge - 8 products in 8 days, 8 hours each. Perfection isn't the goal, exploration is.

🔥 What is 888?

8 Hours. 8 Days. 8 Products. A personal challenge to build 8 functional AI-powered tools in 8 consecutive days — dedicating just ~8 hours to each build.

This is not about perfection — it’s about speed, creativity, and clarity of product thinking.

Every project is an exploration: solving real problems, testing ideas fast, and documenting the build process openly.

📌 How to Follow the Flow

Each day, I’ll share:

A short demo video
A concise build recap post (on LinkedIn & Twitter)
A full write-up in comments or blog (with lessons, tech stack, and what worked/didn’t)
And once ready, a live link to the deployed product (or prototype status if not yet deployed)