How to Create AI-Generated Videos Using Veo 3: A Step-by-Step Guide

This blog post outlines the process Justine Moore used to create an AI-generated video featuring a stormtrooper in the Andreessen Horowitz office, leveraging Veo 3's capabilities. The guide serves as a standard operating procedure (SOP) for similar projects.

Step 1: Accessing Veo 3

Description:

Veo 3, developed by Google, is an AI video generator that transforms text or image prompts into high-definition videos with native audio integration.

Access Requirements:

A Google AI Ultra plan, costing $250/month, is required. It is currently available only in the U.S. through Flow, Google's AI-powered filmmaking interface.

[Source: DataCamp, 2025-05-22]

Step 2: Preparing the Location Frame

Description:

Upload a frame of the desired location, in this case, the Andreessen Horowitz office, to serve as the base for the video.

Purpose:

This frame acts as the starting point for the AI to generate subsequent scenes.

[Source: Justine Moore's X post, 2025-07-17]

Step 3: Prompting Character Movement and Dialogue

Description:

Use a text prompt to describe the character's actions and dialogue. For example, "stormtrooper youtuber walks from the right into the frame past the chairs. he then pauses, turns back to the camera, and whispers 'It's time!'"

Purpose:

This prompt directs Veo 3 to animate the stormtrooper's movement and include specific dialogue.

[Source: Justine Moore's X post, 2025-07-17]

Step 4: Utilizing Fast Jump/Cut for Dynamic Framing

Description:

Employ the "fast jump/cut" prompt, inspired by Martin Nebelong, to change the camera angle within a scene.

Example:

Transition to a selfie-style vlog of the stormtrooper on a couch, saying, "False alarm, I got here early. I'm researching the investor I'm going to meet with. I think her name is Justine?"

Purpose:

This technique enhances the video's dynamic feel and realism.

[Source: Justine Moore's X post, 2025-07-17]

Step 5: Integrating Custom Characters with LoRA Training

Description:

For characters not easily generated by text prompts, use LoRA (Low-Rank Adaptation) training on Krea AI to create custom models.

Process:

Justine trained a Flux LoRA using photos of herself, then generated an image of her sitting with the stormtrooper.

Prompt:

"hdr @ justine sitting across a table in an office from a stormtrooper in white plastic suit."

[Source: Justine Moore's X post, 2025-07-17; Reddit, 2025-01-19]

Step 6: Ensuring Safety and Ethical Standards

Description:

Veo 3 incorporates safety measures, including red teaming and policy evaluations, to prevent inappropriate content generation.

Marking:

All videos are marked with a visible watermark and SynthID, indicating they are AI-generated.

[Source: Gemini AI video generator, accessed 2025-07-19]

Background Context

The rise of AI video generation tools like Veo 3 is part of a broader trend in 2025, where businesses and creators increasingly use AI to produce content efficiently. Andreessen Horowitz, a prominent venture capital firm, has been a focal point for tech innovations, making it a fitting backdrop for this demonstration.

[Source: Synthesia.io, 2025-06-11]
Mix Humans and AI Avatar Together