From prompt to video: my complete AI video workflow for Instagram

I have a confession: I used to stare at a blank AI video prompt for 10 minutes before typing anything.

"What kind of video should I make?" "What angle?" "Is this a good idea?"

That analysis paralysis cost me more time than the actual video generation. So I built myself a system. A repeatable workflow that takes me from blank page to published Instagram Reel in under 20 minutes.

Here's the exact workflow I use every time.

Phase 1: The pre-write (5 minutes)

I don't open any AI tool until I've answered three questions on paper:

1. What's the one thing I want someone to remember after watching?
2. What problem does this solve for the viewer?
3. What's the single visual that proves it?

For a recent video about a collapsible silicone colander, my answers were:

1. "This colander takes almost no space in your cabinet"
2. "Your kitchen drawers are overflowing with bulky gadgets"
3. "Me, folding the colander flat and sliding it into a tiny drawer"

Those three answers become my entire video strategy. Everything else is execution.

Phase 2: Scene mapping (3 minutes)

With those answers in hand, I map out 3-4 scenes. Each scene is one sentence.

Here's the scene map for that colander video:

Scene 1 (2s): A bulky metal colander falling out of an overstuffed kitchen cabinet. Disorganized, cramped feel.

Scene 2 (3s): A silicone colander being unfolded on a counter. Bright lighting, clean background. The action of opening it up.

Scene 3 (3s): Someone rinsing vegetables in the colander under a faucet. Water flowing, greens bouncing. Product in use.

Scene 4 (2s): The colander folded flat, sliding into a shallow drawer. Text overlay: "Your cabinet, but with space."

Total: 10 seconds. That's Instagram Reel territory.

I write these in a simple notes app. Nothing fancy. Just a bullet list.

Phase 3: Prompt building (2 minutes)

Now I turn each scene into a prompt. The formula is simple:

For Scene 2 of the colander video, I wrote:

I don't overcomplicate this. If the prompt is longer than 20 words, I trim it. Longer prompts don't mean better results. They usually mean the AI gets confused.

Phase 4: Generation (5 minutes)

I use MakeClipAI for this, but the approach works anywhere:

1. I paste all 4 scenes into a new project
2. I set each scene to 2-3 seconds
3. I pick a model that handles product shots well (Kling for this one)
4. I generate 3 versions

While the AI works, I scroll Instagram for 3 minutes. It's my "research time." I look at what other stores in my niche are posting and note any trends.

When I come back, I have 3 versions ready. I pick the best one based on:

• Does the product look consistent across scenes?
• Is the motion smooth?
• Does the story make sense without text?

Phase 5: Polish (3 minutes)

The AI output is the raw material. It almost always needs some polish. Here's what I do:

1. Add text overlay on frame 1 — "Bulky colander ruining your cabinet?" or something similar
2. Add a simple caption — the problem + solution format
3. Crop to 9:16 if the output isn't already vertical
4. Add trending audio — I pick a track that matches the pacing

I do this in the Instagram app itself. No need for a separate editor. The text tool and audio library are good enough for what I need.

Phase 6: Publish (2 minutes)

I upload the video with this caption structure:

Example:

I add 2-3 relevant hashtags. No more. Instagram says use more, but my data says 2-3 gets the best reach.

The full timeline

What I've learned from 30+ videos using this workflow

I've been running this workflow weekly since February. Here's what's consistent:

• The pre-write is the most important step. When I skip it or rush through it, the video always underperforms. The 5 minutes I spend answering those three questions saves me from generating random clips I won't use.
• 3-4 scenes is the sweet spot. Two scenes is too simple. Five+ scenes gets messy. Three or four scenes gives me a clear story arc in under 10 seconds.
• Generation time is my research time. I use those 3-5 minutes productively. If I sit and watch the progress bar, I get impatient and make bad decisions.
• The first version is rarely the best. I always generate at least 3 versions. Version 2 or 3 is usually the winner.

When the workflow breaks

This workflow works great for standard product content. But it breaks when:

• The product is very technical (needs more than 10 seconds to explain)
• The product has no clear visual benefit (insurance, software, services)
• I'm trying to be funny or trendy (those videos need a different structure)

For those cases, I adapt the workflow — longer scenes for technical products, more emphasis on lifestyle shots for services, and a completely different approach for trend-based content.

Related reading

• AI video for social media: what actually works — see which video styles drive engagement on each platform
• Stop rewriting AI video prompts from scratch every time — build reusable templates from your best workflows

A final thought

The biggest lesson from running this workflow for months: consistency beats brilliance. A good video posted every day outperforms a perfect video posted once a week.

I don't try to make every video amazing. I try to make every video clear. Clear sells. Clear gets shared. Clear builds trust.

If you're staring at a blank prompt right now, just start with the three questions. The rest will follow.

Start your first video on MakeClipAI — it's free and takes less time than you think. If you want the tool to break a larger idea into scenes for you, use AI Director mode.