How to build viral stickman YouTube videos with Claude Code and free AI tools

You’ve probably seen them: simple stickman videos with basic drawings, no animation, and millions of views. It feels unfair when you’re stuck at zero, especially if you think you need expensive tools, pro editing skills, or complex software to compete.

The truth is, you can build these videos from scratch using only free AI tools—and a smart workflow. This guide breaks down a complete system that goes from idea to published YouTube video using Claude on the free plan, free text-to-speech, free transcription, and free image generation.

The core idea: rhythm beats visuals

The secret behind these viral stickman videos isn’t advanced animation. It’s rhythm—the way the visuals match the natural flow of the voiceover.

Most people build videos in this order: write a script, generate images, then record or generate a voiceover and try to sync everything at the end. On paper it sounds logical, but it creates a mismatch. The images and audio weren’t built for each other, so the pacing feels off. Viewers can’t always explain it, but they feel it—and click away.

The winning workflow flips this completely:

Voiceover first. Scenes second. Always.

By building scenes around the natural pauses in the voice, every image change feels intentional. That’s what keeps people watching—even if your visuals are just stick figures.

Step 1: Generate viral video ideas and script with Claude (free plan)

You’ll start with a single, well-engineered prompt designed for long-form YouTube content. The idea is to feed Claude a prompt that asks for multiple viral topic ideas and then a full script for the one you choose.

On the free Claude plan:

1. Open Claude in your browser.
2. Paste the master prompt (the workflow assumes you’ve prepared one that asks for multiple viral ideas and a long-form script).
3. Claude returns five viral topic ideas tailored for YouTube growth.
4. Pick one idea and reply with its number.

Claude then generates a complete long-form script—typically 5–15 minutes of content—ready for voiceover. It also prepares a downloadable text file so you can easily reuse or edit the script later.

If you want to go deeper into using Claude for YouTube workflows, check out this guide on building viral YouTube Shorts with Claude and free AI tools.

Step 2: Turn the script into a natural voiceover

Next, you convert the script into a voiceover. The workflow uses 11labs on the free plan, which gives you 10,000 credits per month—enough for multiple long videos.

Here’s how:

1. Go to the 11labs website and create a free account.
2. In the left sidebar, open Voices and search for the voice named Raunak (or a similar natural, relatable voice). Add it to your library.
3. Go to Text to Speech in the sidebar.
4. Select the Raunak voice.
5. Paste your full script into the text box.
6. Click Generate, then download the audio file.

This audio is now the backbone of your video. Every scene you create will be built around the timing and pauses in this voiceover.

Step 3: Use AI transcription to find perfect scene boundaries

Now you need to know exactly where each scene should start and end. Instead of guessing, you’ll use an AI transcription tool that also detects pauses with timestamps.

The workflow uses faziscribe.ai:

1. Visit faziscribe.ai and sign up for a free account.
2. Upload your voiceover file.
3. Choose the Accuracy mode (since every frame matters for timing).
4. Click Transcribe audio.

The tool doesn’t just turn speech into text. It marks every pause with exact timestamps—like pauses at 2s, 4s, 7s, and so on. Each pause becomes a natural boundary for a new scene.

It also auto-detects the language (English, Hindi, Urdu, etc.), so this workflow works across languages.

Once it’s done, either download the timestamped script or copy it directly. You’ll see lines of dialogue tied to precise timestamps. This is what you’ll feed back into Claude.

Step 4: Ask Claude to write image prompts for every scene

Go back to your Claude chat and paste the full timestamped script. Ask Claude to generate a detailed text-to-image prompt for each timestamped line.

Claude will usually respond in batches:

1. It takes the first set of timestamped lines and returns a prompt for each one.
2. When it stops, reply with “next” to continue.
3. Repeat until every line has a corresponding image prompt.

When all prompts are ready, Claude can bundle them into a single downloadable text file. You’ll end up with a file that includes:

– Scene text
– Timestamp
– A detailed image prompt per scene

This becomes your blueprint for image generation.

Step 5: Generate stickman scenes with Flow and Nano Banana 2

To turn prompts into images, the workflow uses flow.google (Google’s Flow) with the Nano Banana 2 model. This model gives solid results without hitting generation limits too quickly.

To generate a single image manually:

1. Go to flow.google and create a new project.
2. Turn Agent mode off.
3. Select Image mode.
4. Set aspect ratio to 16:9 (YouTube standard).
5. Set Output per prompt to 1.
6. Choose Nano Banana 2 as the model.
7. Paste the prompt for scene 1.
8. Click Send to generate the image.

You can repeat this for every scene, but if you have 100+ prompts, doing it manually will take hours. That’s where automation comes in.

Step 6: Bulk-generate all images with the Zappy Flow Chrome extension

To speed things up, the workflow uses a free Chrome extension called Zappy Flow, built specifically for bulk image generation in Flow.

Here’s how to set it up:

1. Search for “Zappy Flow Chrome extension” on Google.
2. Open the Chrome Web Store result.
3. Confirm the developer name is zappywala.ai to avoid fake versions.
4. Add the extension to Chrome and pin it.
5. Open Flow in a new tab and create a new project.
6. Turn Agent mode off. The extension should show two messages indicating it’s connected and agent mode is off.

Now configure the bulk generation:

1. In Flow, set mode to Image, aspect ratio to 16:9, output per prompt to 1, and model to Nano Banana 2.
2. In the Zappy extension, paste all your prompts into the prompt box, with one blank line between each prompt. Alternatively, upload the prompt text file you downloaded from Claude.
3. Confirm the total number of prompts (e.g., 103 prompts loaded).
4. Turn Include serial number in file name off (not needed for this workflow).
5. Choose your download folder.
6. Enable Auto save settings and make sure the first option is turned off (as recommended in the workflow).
7. Optionally adjust the random delay between generations, or leave it as is.
8. Click Generate.

The extension will now send each prompt to Flow, generate the image, and download it automatically—one by one. You can minimize the window or switch tabs; just don’t close the browser, the Flow tab, or the extension.

If one or two prompts fail (for example, scene 15 at 39 seconds), just regenerate those manually in Flow using the same settings.

Step 7: Sync images to the voiceover in your video editor

With all your images and the voiceover ready, it’s time to edit.

Open your preferred video editor (any editor with a timeline will work):

1. Import the voiceover file and drag it onto the main audio track.
2. Import all your generated images.
3. Drag the images onto the video track in the correct order.

The key is syncing each image to the right timestamp. The filenames (or your timestamp list) tell you exactly where each scene should start.

For each scene:

1. Look at the next scene’s start time (e.g., 4 seconds).
2. Move the playhead to exactly that timestamp.
3. Trim the current image from the right so it ends at that point.
4. Repeat: if the next scene starts at 7 seconds, move the playhead to 7 seconds and trim the next image there.

You may need to nudge a few cuts slightly to match the feel of the pauses. Take your time here—this is where the rhythm is created. When every image change lands on a natural pause, the video feels smooth and intentional, even with simple stickman drawings.

Once everything is aligned, watch the full video from start to finish. If the pacing feels right, export the video in a YouTube-friendly format (e.g., 1080p, 16:9).

Step 8: Generate viral metadata and thumbnails with Claude and Zappy

With the video exported, you still need strong metadata and a scroll-stopping thumbnail.

Back in your Claude chat, you can ask it to generate:

– A high-CTR YouTube title
– A compelling description
– Relevant tags

Because Claude already knows your script and structure, it can tailor the metadata to your exact video.

For thumbnails:

1. Ask Claude for five high-CTR thumbnail prompts for your video.
2. Tell it to put each prompt on its own line and wrap them in a copyable code block.
3. Copy all five prompts.

Then use the Zappy extension again:

1. Open Flow in image mode with the same settings (16:9, Nano Banana 2, output per prompt 1).
2. Open the Zappy extension and toggle it once to reconnect.
3. Paste the five thumbnail prompts.
4. Choose your download folder.
5. Click Generate.

Within a short time, you’ll have five different thumbnail options. Pick the one that best matches your video’s hook and stands out at small sizes.

Step 9: Upload, publish, and repeat the system

Now you’re ready to publish:

1. Go to your YouTube channel and click Upload.
2. Select your exported video file.
3. Paste in the title, description, and tags generated by Claude.
4. Upload your chosen thumbnail.
5. Set visibility and publish.

You’ve just gone from a blank page to a fully produced stickman-style YouTube video using only free AI tools—no paid subscriptions, no design skills, no team.

Once you’re comfortable with this system, you can reuse it for multiple channels, niches, and languages. And if you want to push even further into AI-powered YouTube automation, you may also like this guide on cloning any YouTube channel’s style with Claude Code.

The key is consistency: keep iterating on topics, hooks, and thumbnails while letting this workflow handle the heavy lifting. The visuals can stay simple—what matters most is rhythm, storytelling, and how well your content keeps viewers watching.