How to Clone Your Avatar with Codex
Codex can turn local reference files into a reusable CreativeClaw Character. Give it several photos and a voice recording, approve the generated character sheet, and save the visual identity and cloned voice together.
The Character lives in your CreativeClaw workspace, so it can be reused beyond one repository or Codex task.
Step 1: connect the CreativeClaw MCP server
Add CreativeClaw to Codex as a remote MCP server and sign in. The connection guide has the current MCP details.
Confirm the connection by asking Codex to list your CreativeClaw Characters. An empty list is fine. It shows that Codex can reach the workspace.
Step 2: prepare the reference photos
Create a small folder with three to six clear images of yourself. Include front, three-quarter, side, and full-body views when possible. Avoid filters, covered facial features, other people, and major appearance changes between images.
Then give Codex explicit paths:
Use CreativeClaw to upload these images as references for a Character based on me:
- /absolute/path/avatar/front.jpg
- /absolute/path/avatar/three-quarter.jpg
- /absolute/path/avatar/profile.jpg
- /absolute/path/avatar/full-body.jpg
Import the files, but do not save the final Character until I approve the
generated character sheet.
Step 3: generate a character sheet
Ask Codex to generate one clean visual anchor from the uploaded references:
Generate a realistic character reference sheet from all uploaded images.
Show full-body front, three-quarter, side, and back views on a plain white
background. Keep the same face, hair, clothing, body proportions, pose, and
lighting. Use a clean studio style with no labels or text.
Open and inspect the result. Pay particular attention to facial proportions, hairline, skin tone, height, build, and distinctive features. Give Codex precise corrections and regenerate if necessary. Do not save a sheet that only vaguely resembles you.
Step 4: record and clone the voice
Record 1 to 2 minutes of clean speech in your primary language. Move naturally through calm, upbeat, serious, excited, and warm tones. Use a quiet room and one speaker.
Give Codex the local file path and explicit consent:
Upload /absolute/path/avatar/voice-sample.wav to CreativeClaw as a private
voice-cloning sample. Clone it with ElevenLabs and attach it to the Character
we are building. This is my voice, and I explicitly consent to its processing
and cloning for use in my CreativeClaw workspace.
You can select Cartesia instead. CreativeClaw supports both ElevenLabs and Cartesia. Keep the sample private, and never treat access to a file as proof that its speaker approved cloning.
Step 5: save the reusable Character
After approving the sheet and clone, ask:
Save a reusable CreativeClaw Character named "My Avatar." Use the approved
character sheet as its visual reference, write an accurate description, and
attach the cloned voice. Return the Character ID and summarize what was saved.
Codex should save the generated character sheet URL as the Character’s visual anchor and attach the cloned provider voice to that same Character.
Step 6: use the Character in future tasks
Use my CreativeClaw Character "My Avatar" to generate a clean founder portrait
for the product website. Preserve the approved identity and proportions.
Use my saved Character to create a talking-head release update. Generate the
speech with its cloned voice and use the Character reference for the video.
The script is in /absolute/path/release-script.md.
List my CreativeClaw Characters and return the ID for "My Avatar."
The saved ID is the most reliable way to select the Character when names are similar. Review the voice-cloning terms and privacy details before using anyone else’s recording.
Frequently asked questions
How do I clone my avatar with Codex?
Connect the CreativeClaw MCP server, give Codex paths to three to six reference photos, generate and approve a multi-view character sheet, privately upload a consented voice sample, and save the sheet and cloned voice on one reusable Character.
Can Codex create a reusable AI avatar from local files?
Yes, with the CreativeClaw MCP server connected. Give Codex the exact paths to several reference photos and a voice recording, then ask it to create and save a visual Character with a cloned voice.
Which photos should I provide?
Provide three to six clear and recent photos, ideally including front, three-quarter, side, and full-body views. Use consistent appearance and avoid filters or obstructed facial features.
What does Codex save in the Character?
CreativeClaw saves the approved character sheet as the visual reference, an appearance and role description, and the attached ElevenLabs or Cartesia cloned voice.
How should I record the voice sample?
Record 1 to 2 minutes of clean solo speech in your primary language. Include several natural emotional tones, keep the microphone distance steady, and avoid music or other speakers.
Can I use the Character outside the current repository?
Yes. The Character belongs to your CreativeClaw workspace rather than one repository. Connect the same account, list your Characters, and use its saved ID in another Codex task or supported client.
Does the Codex avatar workflow train a face model?
No. It creates a character sheet from your reference images and saves that sheet as the Character’s visual anchor. Supported generation models reuse the approved reference instead of relying on a separately trained face model.
Can Codex generate a talking-head video with my Character?
Yes. Give Codex a script and ask it to use the saved Character’s visual reference and cloned voice for a talking-head clip. You can also keep the tasks separate and generate only an image, video, or voiceover.
How do I update or delete the saved avatar?
Ask Codex to update the Character’s title, description, or reference image through CreativeClaw. To remove it completely, delete the Character, which also removes its associated provider voice copies and private voice samples.
Ready to try it?
Connect CreativeClaw to Codex and create your first result.
Connect CreativeClawContinue the workflow