← Back to Lessons
Claude Skills

Recreate Any Viral Reel With Your Own Face & Voice

Hand Claude a viral short and a few selfies. It rebuilds the whole thing as you, your face, your voice, captions on the beat, a custom music bed. You never touch the terminal.
⏱ Setup: ~12 min · a few $ in credits

What you'll walk away with. A short-form reel that looks like you filmed and edited it: your face, your voice, captions snapping to the beat, a custom music bed under it. You don't make any of that by hand. You hand Claude a viral reel you like and a few photos, and it rebuilds the whole thing as you.

You won't touch the terminal or edit anything. You paste one prompt and Claude walks you through it step by step. First it asks for your keys, one at a time ("Okay, what's your ElevenLabs key?", then Gemini, then Kie), then your reel link and your photos. Once it has those, it's off to the races: it rebuilds a reel like the one in this lesson, start to finish, and checks with you before each step that costs money. All you do is answer the questions.

Heads up, this costs a few dollars

I'm not going to pretend it's free. The image, video, voice, and music models charge per use, so figure a few dollars in credits per reel. It runs in steps and stops for your OK before anything that costs money, so you're never surprised.

1
Make three accounts and grab a key from each
6 min

You need three keys. Don't do anything technical with them. Just make each account, copy the key somewhere handy (your notes app is fine), and Claude will ask you for them later.

1. ElevenLabs · your voice

Get your ElevenLabs key →

Sign in, then Settings → API Keys → Create, and copy it. Voice cloning needs a paid plan (the cheapest one is fine).

2. Gemini · your face + reads the reel

Get your Gemini key →

Sign in to Google AI Studio, click Get API key → Create, and copy it. Turn on billing for the project (about $0.13 an image, no free tier).

3. Kie · motion + music

Get your Kie key →

Sign in at kie.ai, open API Key → Create, and copy it. Add a few dollars of credits under Pricing; watch your spend at kie.ai/logs.

Three keys, copied somewhere. That's all you do with them. Claude takes it from here.

2
Get your face, your voice, and your reel ready
4 min

Three things to have ready before you start, all in one folder:

3 to 6 selfies. Even lighting, a few different angles, the same look (beard, glasses, hair) you actually want on camera. The more consistent they are, the more it looks like you.

A one-minute voice recording of you just talking. No music, no background noise. Drop it in the same folder. Claude clones your voice from this on its own, you don't upload anything yourself.

A reel you want to remake. Pick a viral short whose format you like, the structure and the energy, not the exact words. Copy the link. You're borrowing the shape and saying your own thing, that's what keeps it yours.

Easiest way to hand over the folder

When Claude asks for the folder, you don't have to type the path. Just drag the folder straight into the chat. If you're using the VS Code extension, hold Shift while you drag so it drops the path in as text.

3
Paste this one prompt into Claude Code
the fun part

Open Claude Code in any folder. Hit Copy on the prompt below, paste it in, and press enter. You don't change a single word of it, and you don't even need to read it, just answer the questions it asks you.

Show the full prompt (or just hit Copy)
You're going to remake a viral reel starring me, with my face and my voice. I'm not technical, so run this like a friendly guided setup: ask me for everything you need in plain language, ONE thing at a time, and wait for my answer before moving on. Never paste a key into a file or commit one; just hold each key in this session. Start by collecting these from me, one at a time, in this order: 1. Ask: "Okay, what's your ElevenLabs API key?" When I paste it, set it as ELEVENLABS_API_KEY for this session. 2. Ask: "Great. What's your Gemini API key?" Set it as GEMINI_API_KEY. 3. Ask: "And your Kie API key?" Set it as KIE_API_KEY. 4. Ask: "Paste the link to the viral reel you want to remake." 5. Ask: "In a sentence or two, what's your spin on it, the message you want to deliver?" 6. Ask: "Last thing: put your 3-6 selfies and your one-minute voice recording into a single folder, then drag that folder into the chat (or paste its path)." Wait for the folder before continuing. Then build the reel as a GATED pipeline: do each step, show me the result, ask me ONE short question, and STOP for my OK before anything that costs money. We're billed per clip. 0. VOICE CLONE (one-time, tiny). Clone my voice from my voice recording using the ElevenLabs REST API directly (do NOT use an MCP tool, it mis-sends the upload): POST https://api.elevenlabs.io/v1/voices/add with header "xi-api-key: $ELEVENLABS_API_KEY" and multipart fields name="my reel voice" and files=@<my voice file>. Read voice_id from the JSON response and reuse it for every voice swap below. If this fails because my plan doesn't allow cloning, tell me before generating anything paid. 1. BREAKDOWN (cheap). Download the reference reel (yt-dlp; transcode VP9 to H.264 so it isn't audio-only). Send it to Gemini (gemini-3-flash-preview) for a second-by-second breakdown: every shot, each look (warm-desk vs white-void), the on-screen text word for word, the song feel, and the beat hits. Show me the breakdown. Ask: "This is the format. Adapt the script to your angle?" 2. SCRIPT + TEXT MAP (cheap). Rewrite the monologue in my angle, keeping the original's structure and pacing. Give me the spoken script plus the word-pop text map (which words flash on screen per beat) and the end card. Ask me to lock or tweak the wording. 3. LIKENESS STILLS (paid, ~$0.13 each). With Nano Banana Pro (Gemini model gemini-3-pro-image-preview) and my selfies, generate two stills: a hero desk shot (front, hands steepled) and a white-void contrast shot (rim-lit). Put a strong identity-lock + skin-texture / anti-beautification instruction at the END of each prompt. Open both. Ask me to lock the face or adjust. 4. CLIPS (paid, ~$1-1.5 each, ONE AT A TIME). For each beat, use Kie's Seedance (model "bytedance/seedance-2", Jobs API at https://api.kie.ai/api/v1/jobs/createTask, poll recordInfo) to animate the right still into a clip: talking clips from the desk still, white-void stings from the void still. Generate ONE clip, save its taskId, and let me eyeball it on kie.ai/logs before the next. Download every result with a browser User-Agent (Mozilla/5.0), or the CDN returns 403. 5. VOICE (paid, tiny). For each talking clip, use ElevenLabs scribe_v1 to get word-level timestamps, then eleven_multilingual_sts_v2 speech-to-speech to convert the clip's audio to my cloned voice (the voice_id from gate 0), keeping the original timing. Stings stay silent. 6. CAPTIONS (free, preview). Burn the red word-pop captions synced to those word timestamps, one locked font size across all clips, inside a safe margin. Show me a preview. 7. MUSIC (paid, cents). Use Kie's Suno to make an instrumental that matches the format (e.g. drift phonk), then detect the drop and kick times. 8. ASSEMBLE (free). Cut the clips and stings to the beat grid: opening line dry, the music drops on the first hard cut, a continuous bed under the voice, and a DEBUT-style end card. Export the reel. 9. QA (free, blocking). Sample frames and verify nothing overflows the safe margin or runs off canvas. Fix and recheck before you call it done. Then show me the final and ask if I want any changes. Start by asking me for my ElevenLabs API key.

From here you're just chatting. Claude asks for your keys one at a time, asks for your reel link and your folder, then builds the reel step by step and checks with you before each paid step. Approve it, tweak it, or have it redo a step until you love it.

What powers each part

A quick map of which key does what, so you know where the cost and the magic come from.

Part of the reel Model Key
Reading the original, shot by shot Gemini (video understanding) Gemini
A photo-real you Nano Banana Pro (gemini-3-pro-image-preview) Gemini
Your photo, now moving and talking Seedance via Kie (bytedance/seedance-2) Kie
A clone of your voice (from your recording) ElevenLabs Instant Voice Cloning ElevenLabs
Your real voice on the clips ElevenLabs speech-to-speech ElevenLabs
Caption timing ElevenLabs Scribe (word timestamps) ElevenLabs
The music bed Suno via Kie Kie
4
If something's off

Claude didn't ask for a key

Tell it: "ask me for my ElevenLabs, Gemini, and Kie keys one at a time, then continue." It'll pick the flow back up.

A Gemini billing or quota error

The face model is paid. Turn on billing for the key's Google project (see the keys step) and try again.

A clip fails to download (a 403)

Kie's download blocks the default agent. The prompt already tells Claude to download with a browser User-Agent, so just tell it to retry that step.

The voice clone fails

Cloning needs a paid ElevenLabs plan. Check your plan, then tell Claude to retry the clone step.

The face drifts between clips

Use more consistent selfies (same look, even lighting). Tell Claude to keep the identity-lock line at the very end of the photo prompts.

Costs creeping up

Stay one step at a time and check kie.ai/logs before you approve the next clip.

Questions?

Made by AI Service Engine · what is a skill? · generate images with Nano Banana Pro