# I Built a Realistic AI Influencer From Scratch (STEP-BY-STEP GUIDE)

Source: https://www.youtube.com/watch?v=SScQwX6sPAU
Recap page: https://rapidrecap.app/video/SScQwX6sPAU
Generated: 2026-01-12T21:33:14.862+00:00

---
## Quick Overview

The creator provides a step-by-step guide on building a realistic AI influencer by generating an initial image using ChatBT, extracting its details into JSON, generating a consistent voice prompt for 11 Labs via Gemini, creating a script, synthesizing audio, and finally using a lip-syncing tool to maintain facial and audio consistency across different generated scenes.

**Key Points:**
- The process requires generating an initial image, such as that of a monk, exporting it to JSON, and using that JSON data in Gemini to maintain character consistency for new image generations.
- The creator advises generating a voice prompt for 11 Labs by feeding the character description back into ChatBT, emphasizing the need to adjust prompts from 'studio quality' to include ambient noise for realism.
- The creator shares preliminary daily revenue numbers, reporting $830 total revenue yesterday, with Adore generating $650.
- The creator mentions that the tools used, including 11 Labs and ChatBT, cost money (e.g., 11 Labs is around $15), but believes the system is scalable by SOP-ing it out.
- For lip-syncing, the creator tested tools like Yapper and Infinite Talk, noting that the lip-syncing part is the last piece of the puzzle they need to fully perfect.
- The creator suggests that for purely image-based Instagram accounts, one can continuously generate different scenes using the initial character's JSON data without needing video every time.
- The creator expresses concern about the 'dead internet theory,' noting that if it is all AI interacting with AI online, 'it's Jover.'

**Context:** The video serves as a comprehensive tutorial detailing the technical workflow for creating a fully synthetic, consistent AI social media influencer from scratch. The speaker walks through several AI tools, including ChatBT, Gemini (formerly referencing it as nano banano), and 11 Labs, explaining how to use visual data (image JSON) and textual prompts to ensure the generated character maintains a consistent appearance and voice across various generated content, while also briefly sharing daily business revenue figures.

## Detailed Analysis

The tutorial outlines a multi-tool pipeline for AI influencer creation starting with generating a seed image using ChatBT and exporting its specifications into JSON format. This JSON is then fed into Gemini to create the base character visualization, ensuring consistency. Next, the creator uses ChatBT again, referencing the character description, to generate a specific voice prompt for 11 Labs; the creator stresses modifying the initial voice prompt to sound less 'studio like' and more authentic with ambient noise, demonstrated by an audio clip featuring wind whispers and echoing. The resulting audio is downloaded, and an image and audio are combined using a lip-syncing tool, where the creator expresses slight dissatisfaction with current tools like Yapper and Infinite Talk. The process is repeatable; to generate new scenes with the same face, users must input the previous photo along with the new scene prompt into Gemini, making sure to start a new chat session in Gemini each time to avoid history interference. The creator concludes by noting that this method is a powerful way to understand social media algorithms and build accounts rapidly, even suggesting Grok over ChatBT for more 'aggressive' or 'thirst trap' style photos if ChatBT becomes too restrictive.

### AI Generation Pipeline Steps

- Generate initial seed image with ChatBT
- Export image data to JSON
- Feed JSON into Gemini for character generation
- Use ChatBT to generate 11 Labs voice prompt
- Generate script and audio via 11 Labs
- Combine image and audio using lip-syncing tool

### Tool Usage and Cost Considerations

- ChatBT and 11 Labs cost money (e.g., 11 Labs is about $15)
- Gemini usage requires starting a new chat for each image generation to prevent history interference
- Recommended for SOP implementation to generate hundreds of videos

### Voice Prompt Refinement

- Initial voice prompt sounded too studio quality
- Creator adjusted prompt to include ambient noise for a more authentic, less highly produced sound
- Example audio clip included background echo that the creator liked

### Consistency Maintenance

- Use the initial image data exported as JSON as 'train data' when generating new scenes in Gemini
- This ensures the generated character remains the same person across different environments

### Alternative Image Generation Tactics

- For Instagram accounts focused only on images, the creator suggests creating different scenes non-stop without video
- For more aggressive or thirst trap content, the creator suggests using Grok instead of ChatBT due to ChatBT's limitations

### Creator Financial Update

- Yesterday's total revenue was $830, down $70 from the previous day
- Amazon contributed $15, Mediovina $66, and Adore revenue was $650