AI video

From 47 Views to 1.4M: My Real AI Video Workflow (No BS Guide)

The structural change that flipped 47-view videos into 1.4M views. Full workflow: 3 hours per reel, $179/mo tool stack, real time math.

Author: MichyDev — AI video creator with 5M+ views across platforms · Published: May 14, 2026 · 20 minutes

Some links in this article may be affiliate links. If you sign up through them, I may earn a small commission at no cost to you. I only recommend tools I personally use to make my AI videos.

The short answer

Video #43 on my channel got 47 views. Video #178 got 1.4 million. Same channel, same tools, same person editing. Between those two, I shipped 135 videos in 11 months and made exactly one structural change to my workflow.

The change wasn’t a new tool. It wasn’t a new prompt. It was abandoning the “what looks cool?” question and replacing it with “what specific question is my audience already asking that has bad existing answers?”

Everything in this guide is the workflow I rebuilt around that single change. The tools, the order, the time blocks, the failures, the unsexy parts most tutorials skip. No 10x promises. Just the real chain.

1. The 47-view era (what I did wrong for 8 months) 2. The structural change that broke it open 3. My current workflow — full breakdown 4. Tools I use (with honest costs) 5. Time blocks per video 6. The mistakes I keep almost making 7. What I’d skip if starting over today 8. FAQ 9. TL;DR

The 47-view era (what I did wrong for 8 months)

I’d love to tell you my first 42 videos were terrible. They weren’t. Some were genuinely beautiful — Wes Anderson-style AI shorts with consistent character, golden hour color grades, ambitious 60-second cuts.

Views? Averaged 89 per video. The viral ones in my drafts folder hit 230. The losers bottomed at 12 views.

I blamed everything: the algorithm, the niche, my prompts, my tools, my posting schedule, my hooks. I bought 3 courses ($147, $297, $497). None of them worked.

The truth I avoided for 8 months: every one of those videos answered a question nobody was asking. They were technical demos. They were aesthetic experiments. They were good portfolio pieces. They were not videos anyone needed.

The algorithm doesn’t surface “interesting.” It surfaces “answers a real question well.” I was doing the first one beautifully and the second one not at all.

The 3 specific failure modes I had

1. Tool-first thinking.2. Generic angles.3. No retention discipline.

If any of those sound familiar, the next section is where to start.

The structural change that broke it open

I stopped opening AI tools first.

Specifically: I built a 4-prompt question discovery framework that I run BEFORE I touch MidJourney, Veo, Kling, or Higgsfield. The framework asks one thing in four steps:

1. What specific question is being asked but poorly answered right now? 2. What angle on that question makes someone stop scrolling? 3. What hook delivers that angle in 12 words or less? 4. What’s the one visual that makes someone watch second 1?

The full framework is in my free 4 Prompts download. I’m not going to repeat it here because the article exists.

The change in my workflow: I now spend 30-45 minutes finding the question before I open any generation tool. Used to spend 0. The 30 minutes added at the start saves me 3 hours of regeneration later, and the result is videos that actually find their audience.

What changed in the metrics after I rebuilt

| Metric | Pre-change (videos 1-42) | Post-change (videos 43-178) | |—|—|—| | Average views per video | 89 | 12,400 | | Best video views | 230 | 1,400,000 | | Average watch time | 8 sec | 31 sec | | Subs added per video | 0.3 | 18 | | Time per video (total) | 5-7 hours | 3 hours |

Total time per video went DOWN because I stopped regenerating footage that didn’t fit. The 30 minutes of question-finding saved hours of failed generation.

My current workflow (full breakdown)

This is what I do today, in order, for every reel. I’m including the boring parts because most guides skip them and that’s where 80% of creators stall.

Block 1 — Question + script (0:00 – 0:90 minutes)

Step 1.1 — Run the 4-prompt chain.Step 1.2 — Write the spoken script.

Max 90 spoken seconds (gives me 60-75 sec final after edit cuts)
First 12 words = the hook from step 1.1, verbatim
Last 8 seconds = a clear takeaway statement OR a question that invites reply
No “in this video,” no “let me show you,” no welcome
One single idea per video — no multi-topic ramble

Step 1.3 — Plan the visual beats.

Total time: ~90 minutes.

Block 2 — Voice + character (0:90 – 1:30)

Step 2.1 — Record voiceover or generate via ElevenLabs.ElevenLabs

Settings I always use for ElevenLabs:

Stability: 30-40%
Similarity boost: 75%
Style exaggeration: 20%
Speaker boost: ON

Step 2.2 — Trim and clean.

Total time: ~40 minutes.

Block 3 — Visual generation (1:30 – 2:30)

This is where most tutorials over-spend. Don’t.

Step 3.1 — Generate the hero frameStep 3.2 — Generate the remaining 5-7 visual beats.Step 3.3 — Stop generating when you have what you need.

Total time: ~60 minutes.

Block 4 — Edit + ship (2:30 – 3:00)

Step 4.1 — Rough cut in CapCut.Step 4.2 — Tighten the first 3 seconds.Step 4.3 — Add captions, music bed, end frame.Step 4.4 — Export, upload, schedule.

Total time: ~30 minutes.

Grand total per video: ~3 hours

Pre-change workflow: 5-7 hours. Post-change: 3 hours, with 14× more views.

Tools I use (with honest costs)

At 12 reels/month, that’s ~$15 per finished reel in tool costs.

Minimum viable stack to start (cheapest)

If you’re just testing the framework:

MidJourney Basic: $10
Kling free tier (500 daily credits — sign up via my referral): $0
ElevenLabs free (10K chars/mo): $0
CapCut free: $0
Total: $10/mo

You can ship your first 5-10 reels at this budget. Once you’re consistent, upgrade.

Time blocks per video (the math that matters)

Most creators tell you “AI is fast.” It’s not. Here’s the actual time math for me at production speed:

If you’re shipping 3 reels/week, that’s ~11 hours/week of focused production work. Not part-time. Real work.

At 12 reels/month sustained, that’s ~44 hours/month — roughly a part-time job. This is why most creators quit at month 3.

What this looks like as a weekly schedule

Monday morning (3 hrs): reel 1
Wednesday morning (3 hrs): reel 2
Friday morning (3 hrs): reel 3
Sunday (1-2 hrs): plan next week’s 3 questions

Total weekly: 11-12 hours.

The mistakes I keep almost making

After 178 videos, my workflow has cleaned up most of my early errors. But three traps still tempt me regularly:

Trap 1 — “Just one more generation”

I land a 90%-good frame and want to generate 5 more to find the 100%-perfect one. The 5 more cost 30 minutes and the result is rarely better. Discipline: if it’s 90%, ship it. The script and voice carry the video, not the frame.

Trap 2 — Skipping the question step when I’m in a rush

When I have a deadline, I’m tempted to skip the 4-prompt chain and “just make something.” Every time I’ve done this, the video underperforms (sub-500 views typically). The 30-minute prompt step is the single highest-leverage time block in my workflow.

Trap 3 — Reusing what worked last time

If reel #12 hit 100K, I’m tempted to make reel #13 with the same structure and similar visuals. The first one or two reuses work. By reuse #3, the algorithm flags as repetitious content (and post-2026 YouTube monetization rules, this actively demonetizes you). Repeat the framework, not the output.

What I’d skip if starting over today

Looking back at my 8 months of 47-view content, here’s what I’d cut from the standard “AI creator advice” playlist:

Skip 1 — Buying any course over $97

I bought 3 courses for a combined $941. The useful content in all three fit into a single page. The framework was right under my nose: find a real question, then film an answer. No course teaches that without burying it in 4 hours of fluff. The free 4-prompt download is more useful than any of them.

Skip 2 — Posting “behind the scenes” content before you have results

Behind-the-scenes works when you have results. Without results, it’s just process. Nobody cares how I generated a video that got 47 views. Wait for the 100K+ result, then film the BTS.

Skip 3 — Following 10 creators in your niche

I followed 30+ AI creators when I started. Their content shapes your taste, but it also clones your output. After 8 months I unfollowed 90% and the videos got more original instantly.

Skip 4 — Posting on every platform from day 1

I tried Reels, Shorts, TikTok, X, LinkedIn, Pinterest simultaneously for the first 4 months. Reach was so thin on each platform that none of them gave me feedback worth acting on. Better play: pick ONE platform for the first 60 days. Master the format. Then expand.

Skip 5 — Building a website before you have an audience

I built a 12-page website at month 2. Wasted ~$200 in tools and 40 hours. It got 8 visitors a month. Build the website at month 6 once you have content people are coming back for.

FAQ

Why did the 4-prompt chain specifically work for you?

It forced me to validate the question BEFORE investing 3 hours generating. Pre-validation is the hidden multiplier. Anyone can produce footage. Few people validate that the question is worth answering.

How long until I should expect my first 10K-view reel?

Realistic: 60-90 days of consistent posting if you nail the framework. Some people hit it faster. Some take 6 months. The variable isn’t the framework — it’s how specific your niche is and how disciplined you are about not skipping the question step.

How many failures should I expect before the first hit?

In my data, the breakout video was #43. Friends of mine doing the same: their first 5-figure-view video was usually somewhere between #20 and #80. The exact number doesn’t matter — the consistency does. People who quit at video #15 never find out where their breakout would’ve been.

Should I focus on YouTube, Instagram, or TikTok first?

In May 2026:

YouTube if you want long-term SEO + Google search traffic + the most stable monetization
TikTok if you want fastest feedback loops (the algorithm pushes new content harder)
Instagram if you want brand sponsorship deals fastest (sponsors love IG metrics)

For solo creators starting now: YouTube Shorts. The algorithm has been generous to new AI creators since the 2026 monetization update lowered the entry bar to 500 subs.

What if my niche doesn’t have a “viral” question?

Some niches are about depth, not virality. The framework still applies — replace “virality potential” in Prompt 2 with “conversion potential” or “authority potential.” A reel that gets 500 views but converts 30 of them to email is more valuable than a 50K-view reel that converts 5.

Do I need to be on-screen?

No, but the channel needs A voice — yours, your cloned voice, or a clear consistent narrator. Faceless channels with no voice and no consistent narrative direction struggle under the 2026 monetization rules (see why).

What’s the single most leveraged change I could make tomorrow?

Pick one specific question that has bad existing answers. Spend 90 minutes scripting and recording your voice answering it. Generate 5-7 visual beats to support. Ship. That single video, made with this framework, will outperform your last 10 videos made the old way.

TL;DR
I made 42 videos averaging 89 views before I figured out the question matters more than the prompt.

Single structural change: 4-prompt question discovery chain before opening any AI tool.

Time per video DROPPED from 5-7 hours to 3 hours after the change. Views went UP 14×.

Workflow: question (90m) → voice (40m) → visuals (60m) → edit/ship (30m) = 3 hours.

Tool stack: ~$179/mo at production speed, ~$10/mo to start.

Realistic timeline to first 10K-view reel: 60-90 days of consistent posting.

80% of creators quit between month 2-3 — right before compounding kicks in.

Get the framework

The 4-prompt chain that turned my 47-view videos into 1.4M-view reels is a free 1-page PDF.

👉 Get The 4 Prompts (free, 1 page)

For the full system — including the voice settings, hook templates, and the editing flow I rotate — that’s The 1M Views Blueprint ($27, 52 pages).

Want video like this for your business?

Tell me about your project. I usually reply in under 4 hours.

Get in touch