Your face. Your voice.
Your AI clone.
The workflow from my reel, ready to use. Set up Claude + Higgsfield, write a script, and make your first talking-head video.
Free guide + skill. Instant access. Generation costs depend on your account.
Claude plans. Higgsfield creates. You review.
Start the guideSet up once.
Start with one reel.
You’ll need Claude, a connected Higgsfield account, and a few clear references of yourself.
Open HiggsfieldUse my campaign link to check the current offer and eligibility. Confirm the cost before you render.
Gather your face + voice references
Keep these ready in one folder:
- 3 photos: front, left, and right. Clear face, even light, no filters.
- About 25 seconds of your voice: natural speech, quiet room, no music.
- Optional reference clip: helpful for your gestures and delivery.
These are reference inputs. A permanently trained avatar may need more photos; the selected model’s requirements come first. Use your own likeness and voice, or get permission.
Connect Higgsfield to Claude
In Claude, choose Fable 5.2 if available. Open Customize / Settings → Connectors, find Higgsfield, connect your account, and enable it in the chat.
If it isn’t listed, add a custom connector using https://mcp.higgsfield.ai/mcp. Test it by asking: “Check my Higgsfield credit balance.”
Install the skill
Download the clone skill ZIP above. In Claude, open Customize → Skills → + → Create skill → Upload a skill, upload the ZIP, and enable it. Turn on Code execution and file creation under Settings → Capabilities.
Older layouts may show Skills under Settings → Capabilities. This download is adapted for your own scripts; it does not include a fixed promotional video.
Write, test, then make the full reel
Use the prompts below in order. Approve the script first. Attach your photos and voice, then ask for a short test. If Claude opens a Higgsfield upload window, add the files there too.
Check the face, voice, pronunciation, and lip sync. Once the test looks right, approve the full 9:16 video. Save the returned voice and media IDs for your next reel.
Copy. Personalize.
Make it yours.
Replace the text in [brackets]. These three prompts take you from an idea to a reviewed video.
A suggested 9-beat structure is included in the research prompt.
Download all prompts (.txt)
You are an evidence-led topic researcher and short-form scriptwriter. Research [TOPIC] live for [AUDIENCE]. Search the web and accessible Reddit, YouTube, TikTok, Instagram and X sources; say which you could not access. Verify factual claims against primary sources. Never invent quotes, statistics, sentiment, results or personal experience. Pick one strong evidence-backed angle. Write exactly one roughly 45-second vertical-video script with 110–120 spoken words. Use this suggested 9-beat structure: hook → audience problem → surprising reframe → evidence → first action → second action → concrete example → takeaway → one-word comment CTA. End by asking viewers to comment [KEYWORD]. Keep the language conversational. Return the spoken script, its word count, and source links separately. Do not generate a video yet.
Use higgsfield-clone-reels. Make a 9:16 reel of me speaking in my own voice using the script below. I have three photos, a clean voice sample and, if available, a reference clip. Confirm which files Higgsfield can access; open its upload window if needed. Check the current models, input requirements, final assembly options and credit cost. Use my script exactly. Show the plan and cost before paid generation, then make a short test first. After I approve the test, build the full reel with readable captions and consistent face, voice, outfit and setting. Reuse successful assets for fixes. Return the final video link, actual duration and a short quality check. Script: [PASTE APPROVED SCRIPT]
At [TIMESTAMP], [DESCRIBE ISSUE]. Change only that section to [DESIRED RESULT]. Keep my approved face, voice, script and all other sections. Reuse existing assets. Tell me any extra cost before generating, then send the revised result and explain what changed.
Something
looks off?
Be specific about the problem. A timestamp is more useful than “make it better.”
The face or gestures don’t feel like me
Use sharper, evenly lit photos and keep the outfit and framing consistent. Add a reference clip for your mannerisms. Ask to repair the affected scene while preserving the approved voice and script.
The voice or lip sync sounds wrong
Check that the correct voice was selected. Re-record your sample without background noise, then test tricky names or words in a short line. For lip sync, give the exact timestamp and ask to align that segment with the approved audio.
Claude can’t see my files or connector
Enable Higgsfield for this chat and test the credit balance. Reconnect if needed. Files attached to Claude may still need to be added through Higgsfield’s upload window. Confirm all three photos and the voice sample were received before generating.
Can I use the original demo skill?
Yes. The original demo walkthrough explains its workflow, and you can download the original skill ZIP (38 MB). It recreates a fixed promotional reel. Use the smaller clone skill at the top for your own topic and script.
Before you post: watch the whole video with sound. Check every claim, caption, name, cut, and mouth movement. Disclose AI generation where your platform requires it.