A Developer’s Guide to Nano Banana & Veo: Google for Startups

Summary

AI specialist Akanga Visasari walks viewers through building scalable video-generation workflows using Google’s Vertex AI. She explains prompt-engineering essentials for VO (8-second video) outputs—covering subject, action, scene/context, camera angle/movement, and audio—and emphasizes clarity, detail, and iterative testing. Visasari demos an end-to-end workflow in Vertex AI Studio: generating images with Gemini/Nano Banana, outpainting and upscaling in Imagine, then producing VO videos and audio. She also previews the open-source Vertex AI Creative Studio repo, showing how UI event handlers, generate_images logic, and Gemini-powered prompt rewrites enable repeatable, programmatic pipelines for consistent character, product, and story-based video production.

Takeaways

• 😀 Clear and concise prompt design is crucial for generating accurate AI video outputs, particularly when using tools like Vertex AI.
• 😀 Detail is important in prompts, but unnecessary fluff should be avoided. Be specific about the subject, action, scene, and camera angles to get the best results.
• 😀 AI-generated videos with Vertex AI (VO) are limited to eight-second clips, so it's recommended to focus on one scene at a time and stitch them together later.
• 😀 Iteration is key when generating AI content. The first video is unlikely to be perfect, and repeated adjustments based on the AI's decisions are necessary.
• 😀 A successful AI video prompt typically includes five main components: subject, action, scene/context, camera angles, and camera movement.
• 😀 Specificity in the subject matter (e.g., describing the exact person or object) greatly enhances the AI's ability to generate accurate visuals.
• 😀 The action within a video should be described in detail, including any subtle movements or expressions, to ensure the AI captures the desired effect.
• 😀 The environment or context in which a video takes place (e.g., setting a bird's flight in a beach scene or a dark forest) can significantly alter the mood of the video.
• 😀 Using detailed descriptions of camera angles (e.g., close-up or drone shots) and camera movements (e.g., pan, zoom) helps the AI generate more dynamic shots.
• 😀 The ability to generate both video and audio at the same time is a unique feature of Vertex AI, allowing for both environmental sounds and character dialogue in AI-generated content.

Q & A

What is the main focus of this presentation?

The main focus of the presentation is to demonstrate how to implement a scalable video generation workflow using Google Cloud's Vertex AI Studio and related tools, with a particular focus on the prompt engineering process for video generation using VO (video output) models.

What is the significance of prompt engineering in video generation?

Prompt engineering is crucial in video generation because it defines how well the AI can interpret and generate the desired content. Clear and detailed prompts ensure that the AI model understands the user's vision and produces more accurate results.

How long are the videos generated by the VO model?

The VO model currently generates videos that are 8 seconds long, so it's important to focus on one scene at a time and not try to generate an entire story in a single video generation.

What are the key components of a VO prompt?

The key components of a VO prompt include: subject, action, scene/context, camera angle, camera movement, and optionally, audio. These elements help guide the AI in generating accurate video content that aligns with the user's vision.

Why is iteration important in video generation with AI?

Iteration is important because the first video generated may not be exactly what the user expects. Fine-tuning the prompt based on the first output helps refine the video generation process and improves results through experimentation.

What role does camera angle play in video generation?

Camera angle is crucial in determining the composition and framing of the shot. Different angles (e.g., close-up, high angle, drone shot) can completely alter the look and feel of the video, helping to create dynamic and visually appealing content.

How does the Vertex AI Creative Studio enhance workflow?

Vertex AI Creative Studio allows users to deploy a UI to streamline and automate video generation workflows. It integrates multiple AI tools (such as Gemini for prompt assistance, Nano Banana for image generation, and VO for video creation) into a single, programmable workflow that can be customized for different use cases.

What is the purpose of using Nano Banana in the video generation process?

Nano Banana, also known as Gemini 2.5 Flash Image, is used to generate or edit images, which can serve as reference frames for video generation. It helps ensure that visual elements, like models or products, are consistent before being incorporated into video outputs.

What is the role of Imagine in the workflow?

Imagine is used to upscale and modify images (e.g., outpainting to change the background or adjust aspect ratios). It ensures that images generated by Nano Banana are in the right format and resolution for use in video generation.

How does the Vertex AI Creative Studio repository assist developers?

The Vertex AI Creative Studio repository provides a Python-based UI framework and sample code for developers to create customized applications that integrate Vertex AI models. It offers workflows like storyboard generation and character consistency, which can be useful for building scalable, automated video generation systems.