Description
Summary We are building an automated pipeline that takes real reference videos and recreates them using AI video generation — specifically fal.ai Veo3.1. The pipeline is already built and running in n8n. We need someone who can dramatically improve the quality and consistency of what the video generator produces. This is NOT a chatbot project. This is NOT a general AI assistant project. If your experience is primarily with ChatGPT, Claude, or conversational AI — this role is not for you. The specific problem we are solving: We analyze a reference video using Gemini, extract detailed scene descriptions, and pass those descriptions as prompts to fal.ai Veo3.1 to generate new clips. The generated video needs to closely match the reference — same person appearance, same setting, same actions, same motion. Right now the generator is producing generic output that doesn't match the reference well enough. We need someone who understands exactly why this happens and knows how to fix it at the prompt level. You are the right person if: You have personally built prompts for text-to-video or image-to-video generation APIs — fal.ai, Runway, Kling, Pika, or similar You understand why a video generator produces a frozen static frame instead of motion and you know how to fix it in the prompt You know how content moderation works in video generation APIs and you know how to reframe a scene description to avoid flags without losing the meaning You have written scene description prompts that successfully reproduce a specific person, environment, and action in generated footage You understand the difference between a generate prompt and an extend prompt and why they require completely different writing approaches You can look at a generated video that doesn't match the reference and diagnose whether the problem is in the prompt, the model, or the pipeline What you will NOT be doing: Building chatbots Writing customer service AI General LLM prompt work Anything conversational AI related To apply you must answer the screening questions. Generic answers will be rejected