AI TikTok Visual Content System

A two-track image generation pipeline for producing branded AI persona carousels. Track 1 uses a Flux LoRA trained on combined character and style reference images to generate consistent character slides. Track 2 uses Flux Pro with style-locked prompts to generate atmospheric cinematic shots that carry the middle slides of a carousel without requiring the character to appear.

The standard carousel structure: character slide as the hook, atmospheric slides in the middle, character slide as the CTA.

Key settings: guidance_scale 3.5, LoRA weight 0.72, maturity negatives applied. This combination consistently produces semi-realistic painterly output and prevents the anime drift that is the default failure mode for Asian female character LoRAs. Candid and absorbed poses rather than portrait-facing also prevent the anime trigger.

Cost at current fal.ai pricing: approximately $2.50 per 6-image carousel.

What is included: the full pipeline documentation, LoRA training architecture and dataset setup, G3 formula settings, prompt templates for both tracks, the carousel structure formula, and cost breakdown.

Best for AI persona operators running TikTok or Instagram carousel accounts who need visually consistent output at scale.