How to make shorts without showing your face — connect Higgsfield to Claude or ChatGPT, then prompt

Many people want to start making shorts but don't want to show their face. One way to do it is with an AI avatar. The five people below all appeared in one of our short videos, and every one of them is a fictional person made with AI. The same video also features an AI avatar made from our CEO's registered photo.

We made the images, voice and video with Higgsfield, and instead of clicking around the Higgsfield site we asked Claude to do it in plain language. This guide follows that order.

1. Connect Higgsfield to Claude or ChatGPT
2. Prompt the AI (face → outfit → voice → video)
3. Keep the face consistent
4. Turn on AI labels before posting

1. Connect Higgsfield to Claude or ChatGPT

Higgsfield supports MCP, so you can create images and videos right inside a Claude or ChatGPT chat. According to Higgsfield's help center, you need an active paid Higgsfield subscription, and no API key is needed to connect Claude or ChatGPT.

Connect to Claude

1. In claude.ai (or the Claude desktop app), go to Settings → Connectors. In the current interface, Connectors sits under the "Customize" menu.
2. Click + Add → Add custom connector at the top right.
3. Enter Higgsfield as the name and https://mcp.higgsfield.ai/mcp as the URL, then click Continue.
4. Approve the Higgsfield sign-in and authorization window. If you're already logged in to Higgsfield, it goes through automatically.
5. When Higgsfield shows up in your connectors list, you're done.

To check the connection, start a new chat and ask "What is my Higgsfield credit balance?"

Connect to ChatGPT

ChatGPT uses the official Higgsfield plugin instead of a URL. On higgsfield.ai/mcp, click Add Higgsfield to ChatGPT, or find Higgsfield in ChatGPT's plugin directory, add it, then sign in and authorize. Note that, per the help center, audio generation isn't available in ChatGPT — for the voice step in this guide, use Claude or the Higgsfield site.

For developers: API and CLI

• In Claude Code, connect with npm i -g @higgsfield/cli → higgsfield auth login.
• To call it from your own service, create credentials in the Higgsfield Console (console.higgsfield.ai) and send requests to https://api.higgsfield.ai with the header Authorization: Key KEY_ID:KEY_SECRET. Use the keys only in server-side code, never in browser code.

2. Prompt the AI like this (copy-ready)

You don't have to build "elements" by hand on the Higgsfield site — just tell the connected Claude what you want. @name refers to an element (face, place, outfit) saved in Higgsfield.

① Pick a face — from your photo

Create a character element in Higgsfield from my attached photo. Name it my-face.

① Pick a face — a fictional person

Use Higgsfield to create a photo of a Korean woman in her 30s with a short bob and a black turtleneck.
Natural, photorealistic, facing the camera, upper body, sitting on a sofa, 9:16 vertical.

② Make the reference image (the most important step)

Using @my-face, create an upper-body photo of me sitting on a dark leather sofa at night, facing the camera.
A wooden table with a laptop in front, warm indirect lighting, camera at chest height straight on, 9:16 vertical.
When we get one I like, save the background as a place element named night-sofa-room.

③ Change only the outfit

In this reference image, change only the clothes to an olive cardigan.
Keep the face, pose, background, lighting and camera position the same.
Save the result as outfit-olive-cardigan.

④ Create your voice

Create my voice in Higgsfield from my attached voice recording.
Then read this line in that voice: "If you want to make shorts without showing your face, here's how."

⑤ Make the talking video

Use the image from ③ as the start image and make a 9:16 video with lip movements matched to the audio from ④.
Fixed camera, no close-up, keep the upper-body framing.

Repeat ③–⑤ for each scene, join the short clips, add captions and background music, and the video is done.

3. Keep the face from changing every time

The most common problem at first is "the face in the video doesn't match the photo." We had to redo clips ourselves because the face in the video differed from the photo we'd approved — especially the eyes. These are the rules we settled on.

1. Lock the face as an element. Don't describe "my face" in words each time; always call the same element name (@my-face).
2. Save the place as an element too. Same room, same lighting — and even different people feel like one video.
3. Always start the video from an image you've checked. If you just say "make a video with my face," the video AI redraws the face and it drifts. Make the photo first, check it, then use it as the start image.
4. Avoid close-ups. In our experience, an upper-body shot in the same seat with a fixed camera looked the most natural.
5. Say what changes and what stays. "Change only the clothes; keep the face and background."
6. Save good results as elements right away so you can reuse them in the next video.
7. Compare side by side with the original. Check the eyes, eyebrows and jawline — and count the fingers. In one scene where the avatar said "third", the hand showed four fingers.

Saved elements live on the Higgsfield site under the folder icon at the top right (Assets) → "Elements" in the left menu, split into Character, Place and Prop tabs. We keep four outfits and the sofa room there and pick from them for each video.

4. Before you post: turn on AI labels

When you post AI video that looks like real people, telling viewers it was made with AI is part of finishing the job.

Adding one line to the caption, such as "The people and voices in this video were made with AI", also helps avoid confusion. And never use someone else's face or voice without their consent — that rule stays the same whatever tool you use.

In short: connect Higgsfield, lock one reference image as an element, change outfits with words, and use your own voice. You can start making shorts without ever showing your face.

References

• Higgsfield Help Center, How do I connect Higgsfield to Claude, ChatGPT, or another AI agent?: https://higgsfield.ai/creator-hub/help-center/integrations/how-do-i-connect-higgsfield-to-ai-agent
• Higgsfield API docs, Authentication: https://docs.higgsfield.ai/docs/authentication.md
• Higgsfield Blog, How to Keep Characters and Locations Consistent Across AI Shots: https://higgsfield.ai/blog/consistent-characters-locations
• YouTube Help, Disclosing use of altered or synthetic content: https://support.google.com/youtube/answer/14328491
• Instagram Help Center, Label AI content on Instagram: https://help.instagram.com/761121959519495
• TikTok Support, About AI-generated content: https://support.tiktok.com/en/using-tiktok/creating-videos/ai-generated-content

SOL LABS AI

We build web apps, mobile apps and business systems, then add and run the AI they need. We build and operate our own meeting notes, contract and company analysis AI.

See our projects