Generating advertisements with AI avatars has become a hot topic in digital marketing communities. A workflow circulating online describes how to create “UGC” (user-generated content) videos—where an AI avatar holds a product and promotes it while driving—using Gemini. The promise is simple: replace the cost of hiring talent with synthetic content. But what exactly happens in this workflow, and how accessible is it in reality?
The core capability isn’t new. Google announced video generation from images in July 2025, and the feature has been widely covered since then. The function lets you upload reference images to generate videos with visual continuity—a documented ability for over a year now. The workflow being promoted suggests the entire process lives within Gemini, yet the reality is more complex. The initial step actually uses ChatGPT to transform an avatar description into JSON structure; only then does the JSON move into Gemini for image generation. Google’s official documentation doesn’t mandate JSON-formatted prompts—its examples rely on plain language descriptions. There is no evidence that JSON structuring produces better results than straightforward descriptive text.
Gemini can generate audio alongside video, which explains how the avatar acquires a voice. The process carries significant constraints, however. The feature is not free; it requires a paid Google AI plan. Uploading images without proper rights or containing copyrighted material invites sanctions. Every image generated in Gemini includes an invisible SynthID watermark designating it as synthetic. Video duration and daily generation limits remain unspecified on the official support page, though informal sources suggest short clips, likely requiring assembly for longer content.
The ethical issues are real. An avatar that appears human, holds a product, and promotes it edges toward fraud if viewers mistake it for an authentic review. Google’s policies explicitly prohibit using generated video to deceive or manipulate. Labeling content as AI-generated is not enforced by the platform—it falls to the user. In jurisdictions with strict advertising standards, such as several European countries, correct labeling is a legal obligation, not a choice.
The AI avatar workflow functions, but it is more fragmented and constraint-riddled than the hype suggests. The cost, limited video duration, watermarks, and legal exposure from misrepresentation make this a partial tool, not a replacement for real talent. Its value lies in automation, not in flawless results.
