/avatar-make + /tik5-avatar

Hey, it's Jennie.

In 2022, an app called Gas went viral. It was a simple anonymous compliment app for high school students. The team? Just 4 people. Marketing budget? $0. But when teens started posting their experiences on TikTok, daily downloads hit 300,000.

Three months later — $6 million in revenue. Discord acquired them.

This wasn't a fluke. Locket Widget was built by a solo developer. It went viral on TikTok, hit #1 on the App Store, and now makes over $13.5 million a year. Duolingo handed their TikTok to a 23-year-old employee — she grew it to 16 million followers, and annual revenue jumped from $161M to $748M.

All three share the same pattern: TikTok UGC content drove explosive growth. Zero paid ads.

But here's where it gets hard.

You see these stories and think, "I should be posting on TikTok too." But actually making face-on-camera videos consistently? That's a different story.

Set up the camera, adjust the lighting, record, edit. Doing it solo, one video takes 2 hours. Post 3 times a week, and that's 6 hours gone — just on production. No time left for actually building your app or planning content.

So I decided to go with an AI avatar. HeyGen lets you upload photos and generates a talking avatar that reads your script.

But another problem showed up.

Upload just 1-2 photos and the avatar looks stiff. It only learns one angle, one background — so expressions are flat and gestures feel robotic. The official guide says you need at least 10 quality photos across different environments to get a natural-looking avatar.

Generating 10 photos one by one with AI means writing 10 separate prompts and fighting to keep the face consistent. Every new image, the eye color shifts, the jawline changes.

On top of that, you need a fresh script every time. Find a source, adapt it, translate it, organize the files. From finding a single video source to getting a finished HeyGen video — it took over 2 hours.

So I built 2 skills.

The first: /avatar-make — 10 Avatar Photos in One Batch

Pick your gender, pick your orientation (horizontal or vertical), and 10 situational photos drop in one batch. It runs on Google Gemini Imagen 4.0 API.

Studio with gray backdrop, cafe, inside a car, office, living room with bookshelves, podcast studio with a mic — all different backgrounds. But the face is identical across all 10. The prompt locks in facial features (age, skin tone, hair color, eye color) and only swaps the background and clothing.

Run it once, and in under 2 minutes, 10 photos land in your folder. Upload them straight to HeyGen. Avatar registration done.

The second: /tik5-avatar — Script Adaptation Workflow

This skill turns any source into a script your avatar can read.

Throw in a YouTube transcript, a topic, or raw text, and the pipeline kicks off:

Every phase waits for your confirmation before moving on. It doesn't blast through automatically. Source table — confirm. Adaptation plan — confirm. Script draft — confirm. Say "make the tone more casual" and it adjusts right there.

I added a Korean-to-English translation phase because I think in Korean. But that's my workflow. You can add, remove, or customize any phase to fit yours. Running a Spanish channel? Add a Spanish translation phase. Already writing in English? Skip translation entirely.

Connect the two and here's the full pipeline:

The avatar is a one-time setup. After that, it's just /tik5-avatar every time you need a new video.

From finding a source to getting a finished HeyGen video: 2 hours to 20 minutes. 6x faster.

Post 3 times a week and you're saving 5 hours every week. Use that time to build your app, create more content, or just rest.

One thing I learned from building this.

I originally tried to make one mega-skill that did everything — photo generation, script writing, video rendering — all in a single command.

But avatar photos are a one-time thing. You make them once and use them for months. Scripts need to be written fresh every time. Bundling tasks with completely different cycles into one skill made it unnecessarily complex.

Tasks with different cycles need separate skills. One-time tasks get a one-time skill. Recurring tasks get a recurring skill.

Gas made $6 million with 4 people because the product was simple. Content creation works the same way. Two simple tools you can snap together like Lego blocks beat one complicated robot every time.

Whatever language you run your channel in, whatever topic you cover — this structure applies. Build a face with the avatar skill, pump out content with the script skill. Customize the phases to fit your channel.

Jennie

Get the Skills

Want to try these skills yourself? Here are the full source files:

• /avatar-make — AI Avatar Photo Batch Generator (10 photos for HeyGen)
• /tik5-avatar — TikTok Avatar Script Adaptation Workflow

How to use

1. Click Raw on the Gist page and copy the entire content
2. Save it as a SKILL.md file in your project:
   .claude/skills/avatar-make/SKILL.md
.claude/skills/tik5-avatar/SKILL.md
3. Open Claude Code and type /avatar-make or /tik5-avatar to run