ANTIGRAVITY LABJP
Articles/Antigravity Basics
Antigravity Basics/2026-04-20Beginner

I Tried Google's Veo 3 — What It Can Do, What It Can't, and Whether It's Worth Your Time

An honest first look at Veo 3 from someone who builds apps and creates content. What actually works, what doesn't, and where indie developers and creators might find real value right now.

Veo 33video generation AIGoogle AI4Google Antigravity7indie development14creator tools

I wasn't immediately impressed when I first heard that text-to-video AI had arrived. Image generation becoming mainstream had already made that direction feel inevitable. But when I actually tried Veo 3, something shifted: this is moving faster than I expected.

Veo is Google's video generation model, evolved from Veo 2 (released late 2024) to Veo 3 in 2025. It's now accessible through the Antigravity API (Google's unified AI platform), which means indie developers and creators can actually experiment with it at scale.

This article is my honest take on what Veo 3 is like to use, and who might find it genuinely useful right now.

What Veo 3 Can Generate

The core specs worth knowing: up to 1080p resolution, up to about 60 seconds (configuration-dependent), and it supports both text-to-video and image-to-video (using a reference image as a starting point).

Generation time for an 8-second clip runs around 2–5 minutes. Not real-time, but within the space of a coffee break.

The quality leap from Veo 2 is noticeable — particularly in motion smoothness and lighting. The "obviously AI" look has diminished. That said, close-ups of hands, faces, and complex physical actions still produce inconsistencies. This is an active area of improvement, but it's real.

Prompts That Worked Well

One that produced results I'd actually consider usable:

"Slow cinematic shot of morning mist drifting across a quiet mountain lake. Camera gradually moves closer to the water surface. Desaturated color grade, peaceful atmosphere."

The camera movement instruction — "gradually moves closer" — was reflected with surprising accuracy. For a wallpaper app or ambient video context, the output quality was genuinely close to usable.

What didn't work well: scenes with multiple people interacting. Facial expressions and body movement frequently don't match in expected ways. I wouldn't try to use Veo 3 for anything requiring realistic human behavior yet.

Accessing Veo 3 via the Antigravity API

Veo 3 is available through Google AI Studio directly, or via the Antigravity API (Gemini API) using veo-003.

import google.generativeai as genai
import time
 
genai.configure(api_key="YOUR_API_KEY")
 
def generate_video(prompt: str, output_path: str = "output.mp4") -> str:
    """
    Generate a video using Veo 3.
    
    Args:
        prompt: Text description of the video
        output_path: Where to save the output file
    
    Returns:
        Path to the generated video file
    """
    client = genai.Client()
    
    operation = client.models.generate_video(
        model="veo-003",
        prompt=prompt,
        config={
            "duration_seconds": 8,    # 4–8 seconds
            "resolution": "1080p",
            "aspect_ratio": "16:9"
        }
    )
    
    # Poll until generation completes
    while not operation.done:
        print("Generating...")
        time.sleep(10)
        operation = operation.refresh()
    
    if operation.error:
        raise RuntimeError(f"Generation error: {operation.error}")
    
    video_data = operation.result.video
    with open(output_path, "wb") as f:
        f.write(video_data)
    
    return output_path
 
if __name__ == "__main__":
    prompt = """
    Slow cinematic shot of morning mist drifting across a quiet mountain lake.
    Camera gradually moves closer to the water surface.
    Desaturated color grade, peaceful atmosphere, no motion blur artifacts.
    """
    
    output = generate_video(prompt, "lake_morning.mp4")
    print(f"Done: {output}")

Note: Veo 3 API access currently requires Google AI Pro or API credits. Free tier access is limited.

My Honest Evaluation

What works well:

  • Natural landscape and ambient video quality is close to practically usable
  • Camera movement instructions (pan, zoom, dolly) are respected reasonably well
  • Real potential for pipelines combining Veo 3 with other Antigravity models

What still needs work:

  • Human subjects, especially hands and face close-ups, show typical AI inconsistencies
  • 2–5 minute generation time means it's not suited for quick iteration cycles
  • Credit costs accumulate faster than you might expect

My honest bottom line: If you create ambient content, wallpaper visuals, or environmental media, experimenting now is worthwhile. For content involving people or requiring precise scene direction, a few more model generations might be worth waiting for.

Who Should Try This Now

Veo 3 fits best for ambient and environmental video content creators, product designers who want rapid video prototypes, and app developers exploring animated wallpaper or Live Photo equivalents.

Using it as a primary content source for YouTube or brand advertising is probably premature — the quality threshold isn't quite there yet for those contexts. The realistic current use pattern is: generate, review, select the best takes, then edit or combine with other assets.

For combining video generation with other Antigravity models, the Google AI Studio agent design guide is worth reading. The pipeline possibilities between Veo 3 and other models in the Antigravity family are where things get interesting.

Share

Thank You for Reading

Antigravity Lab is ad-free, supported entirely by members like you. We publish practical guides daily with implementation code, benchmarks, and production-ready patterns. If you've found it useful, we'd love to have you on board.

  • Copy-paste ready implementation code
  • New advanced guides published daily
  • $5/mo or $10 for lifetime access
View Membership →

If you found this article helpful, a small tip ($1.50) would mean a lot to us. Your support helps keep this site ad-free and covers server and hosting costs.

Related Articles

Antigravity2026-06-23
When the Default Model Changes Underneath You: Pinning and Diff-Gating Scheduled Runs
Antigravity 2.0 promoted Gemini 3.5 Flash to the default fast model. It is a welcome upgrade, but any scheduled run that leaned on the default starts producing subtly different output one morning. Here is how I pin the model explicitly, fingerprint the output, and gate drift, sized for a solo developer's pipeline.
Antigravity2026-06-14
After Gemini 3.5 Flash Became the Default, Route Flash and Pro Per Task
Now that Antigravity's default Flash is Gemini 3.5 Flash, leaving everything on Flash wastes accuracy and forcing everything onto Pro wastes time. Here is a two-axis decision table for splitting work between Flash and Pro, plus the routing setup to wire it into your agents.
Antigravity2026-06-12
Measuring the Break-Even Point Between Google AI Pro and Ultra — 14 Days of Quota Data from Parallel Agent Runs
Is AI Ultra ($100/month, 5x the Pro limits) actually worth it? A Python harness that aggregates daily quota consumption from agent logs, 14 days of real measurements, and a formula that converts wait time into money to settle the question.
📚RECOMMENDED BOOKS
Build a Large Language Model (From Scratch)
Sebastian Raschka
LLM Dev
Prompt Engineering for LLMs
Berryman & Ziegler
Prompting
AI Engineering
Chip Huyen
AI Eng
* Contains affiliate links
See all →