How I Made a 10-Minute Animated YouTube Video with Claude AI

From 19 minutes of raw voice notes to a finished animated video
I used AI video editing with Claude to turn my raw voice recordings into a 10-minute animated YouTube video. Every scene moves, every word has a caption, and every claim is backed by a real screenshot. I did not open a traditional video editor once, and the whole video was done in one evening.
The video tells a true story. In 2022, Fiverr banned my seller account permanently. I appealed four times and got copy-paste answers. In 2026, I signed up again with the same email, and it simply worked. I first wrote it as a blog post: Fiverr Banned Me "Permanently." Then the Market Crashed, and Now It Lets Me Back In With the Same Email. Then I wanted to tell the same story in a way people would actually watch.
In this post, I show you exactly how I made it:
- how one blog post became a script, a video and posts for every social network,
- how I cleaned noisy audio and cut out my mistakes,
- how the video was planned scene by scene,
- how the animations were built with code instead of a timeline,
- what I did, and what Claude did.
If you are a creator, a coach or a business that needs explainer or story videos, this post also shows you what I can build for you.
The result: a 10-minute animated story video
Here is the finished video:
| Detail | Value |
|---|---|
| Length | 10:02 |
| Format | 1920×1080, 30 frames per second, MP4 |
| Raw voice recorded | 4 files, about 19 minutes |
| Voice after editing | 9:59 (mistakes and long pauses removed, played 1.12× faster with natural pitch) |
| Scenes | 64 planned scenes in 10 chapters |
| Real screenshots used as proof | 10 |
| Captions | 1,453 words, burned in, with the spoken word highlighted in green |
| Subtitles | English and Bangla SRT files, plus a Bangla title and description on YouTube |
| Music | Original background track, composed for this video, so no copyright claims |
| Thumbnail | Made separately with ChatGPT image generation from one of my photos |
| Final render time | About 18 minutes on my own PC |
| Total time | One evening, from the finished voice recordings to the finished video |
Before the video: one story, written once
The video did not start with video. It started with a blog post.
I gave Claude (the normal Claude chat app) my real story and screenshots of every email: the ban email, my appeals and Fiverr's replies. Claude drafted the article from those facts. Then I edited it myself. I added my own parts, and I published it on my website.
Next, Claude turned the published article into a voiceover script. English is not my first language, so I asked for short, simple sentences that are easy to say on camera. The script also had notes for which screenshot to show at each moment.
My AI video editing workflow with Claude, step by step
The whole project ran inside Claude Code on my Windows PC, through the Code tab in the Claude desktop app. I had already made animated videos this way before, so all the tools were installed. I gave one written brief, and Claude followed it step by step. It stopped at every point where I wanted to approve something.
Step 1: Write a clear brief
Before anything else, I prepared a one-page brief, a file called VIDEO_BRIEF.md, with help from Claude in the chat app. I put it in one project folder together with four subfolders: my voice files, my screenshots, my photo and the script.
The brief described the look (light background, green for good news, red for bad news, a yellow highlighter), the rules (something moves every 2 to 4 seconds, screenshots stay on screen for only 1.5 to 2.5 seconds) and the outro with my hiring links.
I also listed my approval points: the noise samples, the list of cuts, the speed samples, the full cleaned voice and the scene plan. This brief was the most important part of the project. A clear brief is what makes AI output look professional instead of random.
Step 2: Record the voice, mistakes included
I recorded the story on my phone in four parts, about 19 minutes in total. My first language is Bangla, but I spoke in English because I want this story to reach freelancers all over the world. I say this at the start of the video too.
I did not try to be perfect. When I made a mistake, I just said the sentence again. Reading a script line by line is hard for me, so I also changed many lines while speaking. There was also road noise in the background.
None of that was a problem. Claude treated what I actually said as the truth, and used the script only as a guide.
Step 3: Clean the audio
Claude used FFmpeg to remove the background noise. It learned the noise from a quiet moment in each recording, removed it, and lowered the sound in the pauses between sentences.
It made three samples with light, medium and strong cleaning. I listened and picked the strong one. Then it set all four parts to the same loudness, the level YouTube expects (-16 LUFS).

Step 4: Turn speech into text with Whisper
Claude ran Whisper (large-v3, an open-source speech-to-text model) on my own graphics card. This gave a transcript with the exact time of every word. That timing is what later made the captions and the animations line up with my voice.
Step 5: Approve the list of cuts
Claude found every repeated take, false start, filler word and long pause. It did not cut anything on its own. Instead, it wrote an "edit decisions" list with a number for every cut, so I could reply like "3.5: keep the last take" or "1.11: keep".
After my approval, it made the cuts at the quietest point between words, so nothing sounds chopped. Then it checked the result with Whisper again and fixed about 20 cuts that had clipped the end of a word.
Step 6: Set the pace
I speak slowly, so I listened to samples at normal speed, 1.12× and 1.20× and chose 1.12×. The pitch stays natural, so it still sounds like me. The voice went from 11:10 to 9:59. I listened to the full cleaned voice once more and approved it before any animation work started.
Step 7: Plan 64 scenes before building anything
Claude wrote a scene plan: 64 scenes in 10 chapters, each with its time, my words and what appears on screen. For example: on the word "Permanently", a red stamp hits and the word shakes. On each "same" in "same email, same Payoneer, same name", a green tick appears in a checklist.
I changed a few things and approved it. Fixing a plan takes minutes. Fixing a finished animation takes hours.
Step 8: Build the animation in code with Remotion
This is the part most people don't expect. The video is not edited on a timeline. It is a React app built with Remotion, a framework that turns code into video.

Claude wrote reusable building blocks: text that slams in, a counter that counts up to $20,000, a stock chart that draws itself going down, a robot with a name tag, a door that slams shut, confetti and an appeals timeline. Every animation starts at the exact moment the word is spoken, because the code reads the times from the transcript.
To check its own work, Claude rendered test frames from all over the video and looked at them. It fixed layout problems before I ever saw them.
Step 9: Use screenshots as proof, not as a slideshow
I wanted an animated video, not a slideshow. So each real screenshot (the ban email, my appeals, the replies, the 404 page) appears as a short "PROOF" card. It slides in, zooms to one line, gets a yellow highlight and leaves after about 2 seconds.

Step 10: Add captions that follow the voice
The captions are burned into the video, two lines at most, with the current word in green. Claude corrected the transcript first, for example names like Fiverr, Payoneer and my new username, so the captions are spelled right. It also made a matching SRT file for YouTube subtitles.
Step 11: Add quiet sound effects and original music
Claude generated eight small sound effects in code: a whoosh, a pop, a stamp, a slam and a few more. They sit about 13 dB below my voice, so they add energy without being annoying.
For music, most free tracks I found were registered with YouTube Content ID, which can trigger copyright claims. So Claude composed an original background track in code. It is calm and thoughtful during the story and turns hopeful at the "welcome back" part. It sits about 19 dB under my voice and dips a little more while I speak.
Step 12: Preview, approve, render
I watched the full preview in Remotion Studio and approved it. The final render took about 18 minutes on my PC. Claude then checked the file: the size, the frame rate, the loudness and sample frames from the start, middle and end.
Step 13: Make the thumbnail
The thumbnail was not made in Claude Code. I asked Claude for a detailed image prompt, gave that prompt and my photo to ChatGPT, and it generated the thumbnail: my face on the right, a cracked red "PERMANENTLY DISABLED" stamp, the headline "NOW THEY WANT ME BACK" in bright green, and a sad robot holding a "PLEASE COME BACK" sign. Then Claude reviewed it and suggested changes for phone-size viewing.
Step 14: Add Bangla subtitles, title and description
Many of my viewers speak Bangla. So Claude translated the English subtitle file into natural Bangla, keeping every timing the same, and wrote a Bangla title and description. I uploaded them in YouTube Studio under Subtitles. Now viewers can choose Bangla captions, and people whose YouTube app is set to Bangla see the Bangla title.
Step 15: Publish the story everywhere
One story became content for every platform. Claude wrote:
- the YouTube title, description, hashtags and tags,
- a Facebook caption that is emotional, a bit funny, and mentions that the video was made with Claude,
- a short Bluesky post under 300 characters,
- a more professional LinkedIn post with lessons for freelancers, with the links in the first comment,
- a Pinterest pin title and description that link back to my blog post.
What I did vs. what Claude did
I want to be honest about this, because it matters to clients. Claude did the technical work and drafted the writing. I lived the story, made the creative and editorial decisions, and nothing went out without my approval.
| Part of the project | My part | Claude's part |
|---|---|---|
| Story and article | Lived the story, supplied every email and screenshot, edited the article and added my own parts | Drafted the article from my story and screenshots (in the Claude chat app) |
| Script | Asked for simple sentences, changed lines freely while recording | Turned the article into a voiceover script; Claude Code used it as a guide only |
| Voice | Recorded it, chose the cleaning level and the speed | Removed noise, matched loudness, changed speed with natural pitch |
| Editing | Approved every cut, chose the best takes, approved the full cleaned voice | Found every mistake and pause, made clean cuts, checked them |
| Look and scenes | Approved the brief, approved and changed the scene plan | Wrote the brief with me, wrote the 64-scene plan and built every animation in code |
| Proof | Supplied the real screenshots and chose what to show | Turned them into short animated proof cards |
| Captions | Checked names and spelling | Transcribed, corrected and timed every word |
| Sound | Chose quiet effects and approved the music | Made the sound effects and composed the music |
| Thumbnail | Gave my photo and the prompt to ChatGPT, picked the result | Wrote the image prompt and reviewed the result (ChatGPT made the image) |
| Bangla version | Uploaded the subtitles, title and description | Translated the subtitles and wrote the Bangla title and description |
| Publishing | Posted on my website, YouTube and social networks | Wrote the posts for each platform |
| Quality | Watched the preview and gave the final OK | Rendered test frames, fixed layout, checked the final file |
Tools used
- Claude Code (Anthropic), through the Claude desktop app: the AI assistant that planned, wrote the code and ran every step on my PC.
- Claude (chat app): drafted the article, the script, the brief, the Bangla subtitles and the social posts.
- Remotion with React and TypeScript: builds the video from code.
- FFmpeg: audio cleanup, loudness, speed change and the final music mix.
- Whisper large-v3 (via faster-whisper): speech-to-text with word timings, on my own graphics card.
- Python: small helper scripts, for example the cut list and the captions file.
- ChatGPT image generation: the thumbnail.
7 lessons for creators using AI for video
- The brief is the product. The clearer your rules (colors, pace, what to show and when), the more professional the result.
- Approve before building. Check noise samples, the cut list and the scene plan before the long work starts. Changing a plan is cheap. Changing a render is not.
- Don't trust the first cut. Automatic cuts can clip the end of a word. A second listen, or a second transcription, catches it.
- Proof beats claims. Short screenshot cards with one highlighted line make a story believable without turning it into a slideshow.
- Watch the music license. Many "free" tracks are registered with YouTube Content ID. Original or carefully checked music avoids claims on your video.
- Keep the human voice. AI handled the technical work, but the story, the voice and the decisions are mine. That is what viewers connect with. You don't need perfect English or a perfect recording. Your mistakes can be cut; your story can't be replaced.
- Write once, publish everywhere. One true story became a blog post, a video, subtitles in two languages and posts for five platforms. Start with one strong piece and let AI help you adapt it.
Hire me: AI-powered animated video editing
I now offer this same workflow as a service. You send your voice recording, and I deliver a finished animated video with captions, ready for YouTube.
What I can make for you
- Story and explainer videos for YouTube, with kinetic text, icons and simple animations.
- Talking-head or voice-over videos turned into animated videos, so viewers stay longer.
- Case studies and testimonials with your screenshots shown as clean "proof" cards.
- Course and tutorial videos with animated key points and word-by-word captions.
- Audio cleanup only: noise removal, cut mistakes and pauses, even loudness.
- English and Bangla subtitles, plus a translated YouTube title and description.
- Script writing: send me your story and I'll turn it into a script that is easy to record.
How it works
- You send your recording, script (optional), screenshots and brand colors.
- I send back cleaned audio samples and a list of cuts for your approval.
- I send a scene plan: what appears on screen at each moment.
- You review a preview and ask for changes.
- You get the final MP4 (1920×1080), an SRT caption file and, if you want, a thumbnail frame.
Pricing
Basic: $40
Audio cleanup, mistakes and pauses cut, word-by-word captions, and an SRT file. For recordings up to 10 minutes.
Standard: $120
Full animated video up to 5 minutes, with captions, proof cards, sound effects and music.
Premium: $220
Full animated video up to 10 minutes, plus a thumbnail and subtitles in English and other language.
Add-ons:
Each package includes 2 rounds of changes. Extra rounds cost $15 each.
Extra minute beyond 10 minutes: $20 per minute
Script writing from the client’s story: $30
Rush delivery within 24 hours: 30% extra
Get in touch
- Upwork: upwork.com/freelancers/~013916a584fddc7c31
- Freelancer: freelancer.com/u/abdullahtofa
- Website: Send me a message on abdullahtofa.com
Tell me what your video is about and how long your recording is. I'll reply with a plan and a price.
Frequently asked questions
Can Claude AI edit videos?
Claude does not edit video in a timeline like Premiere Pro. Instead, it writes and runs code. With tools like Remotion and FFmpeg, it can clean audio, cut mistakes, add captions and build a fully animated video from code.
Do I need to know how to code?
No. I am not a programmer. I described what I wanted in plain words, approved each step and Claude handled the code. For clients, I handle the whole process.
How long did the whole project take?
The video itself was done in one evening, from my finished voice recordings to the final file. The final 10-minute video rendered in about 18 minutes on my own PC. In total, it took me 5 hours.
Is the music copyright-free?
Yes. The background music was composed for this video, so it cannot trigger a YouTube Content ID claim.
Does the video still sound like me?
Yes. The voice is my real recording. It was cleaned, the mistakes were cut and it plays 1.12× faster with the pitch kept natural.
My English is not perfect. Can I still make a video like this?
Yes. English is not my first language either. I repeated lines and changed words while recording, and the mistakes were cut out. Captions also help viewers follow every word.
Can you make a video like this for me?
Yes. Send me your voice recording and screenshots on Upwork, Freelancer or through my website, and I'll send you a plan and a price.


