video

Antigravity

Antigravity Review: AI Animation Project Generator Tested (2026)

Visit Antigravity

Tested Hands-OnAI Video GeneratorCode-based Animation

Testing History

May 2026Generate Animation Videos Programmatically Using AI#2 Ranked

Our take

In-Depth Review

Our detailed analysis of Antigravity — features, performance, and real-world testing.

Mahreen Fathima

AI Demos Team

Verified Review

Feature-by-Feature Breakdown

We tested each feature individually. Click any card to see inputs, outputs, and our observations.

Implementation Plan

Strong — adds transparency and reduces surprises in generated output.

9/10

▾

Test Summary

Feature tested: Implementation Plan

Result: Passed (9/10) — Strong — adds transparency and reduces surprises in generated output.

Before generating code, Antigravity provides feature to plan the project by creating a structured implementation markdown for the animation sequence. Review and approve before execution.

TEXT

Search Engine Prompt : Using Remotion library, Create an animation video explaining how search engines work. Render the video as output.mp4

↓→

SCREENSHOT

Bottom Line

We gave it a vague search engine prompt. Scan the implementation document above, specifically the goal description section. How it structured the stages without us specifying them is worth noticing.

Automatic MP4 Rendering

Strong — renders without CLI commands, external tools, or render-pipeline debugging.

8/10

▾

Test Summary

Feature tested: Automatic MP4 Rendering

Result: Passed (8/10) — Strong — renders without CLI commands, external tools, or render-pipeline debugging.

Feature tested: Automatic MP4 Rendering

Result: Passed (8/10)

Verdict: Strong — renders without CLI commands, external tools, or render-pipeline debugging.

Expected behavior: Antigravity automatically creates the full Remotion project and renders it to MP4 without requiring a manual render pipeline, through its agent.

Test case: Artifact → Video file

Input type: Artifact

Input used: Input artifact (Artifact): Using Remotion library, Create an animation video explaining how search engines work. Render the video as output.mp4

Observed output: Output artifact (Video file): Rendered output — output-1.mp4

Input artifact: Input artifact (Artifact): Using Remotion library, Create an animation video explaining how search engines work. Render the video as output.mp4

Output artifact: Output artifact (Video file): Rendered output — output-1.mp4

What changed: Artifact transformed into Video file

Test case: Artifact → Video file

Input type: Artifact

Input used: Input artifact (Artifact): Using Remotion library, Create an animation video that explains how Retrieval-Augmented Generation works. A user query gets embedded into a vector. That vector searches a vector database to retrieve similar text chunks. If no relevant chunks are found above a confidence threshold, show the system returning a "no relevant context found" message. If relevant chunks are found, those chunks are combined with the original query into a prompt. That prompt is sent to an LLM which generates a response. The LLM response is shown to the user. Additionally, show a feedback loop where the user can rate the response as helpful or unhelpful — if unhelpful, the system re-runs the vector search with a modified query to try retrieving different chunks, then sends the new chunks to the LLM for a second attempt. Label each component clearly — user query, embedding, vector database, confidence threshold, retrieved chunks, prompt assembly, LLM, response, feedback, and query refinement. Show how data flows between each component, including the retry loop when initial retrieval fails. Render the video as output2.mp4

Observed output: Output artifact (Video file): output2.mp4

Input artifact: Input artifact (Artifact): Using Remotion library, Create an animation video that explains how Retrieval-Augmented Generation works. A user query gets embedded into a vector. That vector searches a vector database to retrieve similar text chunks. If no relevant chunks are found above a confidence threshold, show the system returning a "no relevant context found" message. If relevant chunks are found, those chunks are combined with the original query into a prompt. That prompt is sent to an LLM which generates a response. The LLM response is shown to the user. Additionally, show a feedback loop where the user can rate the response as helpful or unhelpful — if unhelpful, the system re-runs the vector search with a modified query to try retrieving different chunks, then sends the new chunks to the LLM for a second attempt. Label each component clearly — user query, embedding, vector database, confidence threshold, retrieved chunks, prompt assembly, LLM, response, feedback, and query refinement. Show how data flows between each component, including the retry loop when initial retrieval fails. Render the video as output2.mp4

Output artifact: Output artifact (Video file): output2.mp4

What changed: Artifact transformed into Video file

Test case: Artifact → Video file

Input type: Artifact

Input used: Input artifact (Artifact): Using Remotion library, Create an animation video that explains how cloud storage services like Google Drive or Dropbox work. Show a title card with the text "How Cloud Storage Works" at the start. Explain the following steps: when a user saves a file, it gets broken into chunks and encrypted. Those encrypted chunks are distributed across multiple servers in different geographic locations for redundancy. When the user accesses the file from another device, the app detects which chunks are already on that device and only downloads the new or modified ones. Show what happens when the same file is edited on two different devices at the same time — both devices make conflicting edits and save them. The system detects the conflict, preserves both versions, and lets the user choose which one to keep. Label each component clearly — user device 1, user device 2, file chunks, encryption, chunk servers (distributed), sync service, conflict detection, version history. Show how data flows between devices and servers, and how the system handles the sync and conflict scenarios. Render the video as output3.mp4

Observed output: Output artifact (Video file): output4.mp4

Input artifact: Input artifact (Artifact): Using Remotion library, Create an animation video that explains how cloud storage services like Google Drive or Dropbox work. Show a title card with the text "How Cloud Storage Works" at the start. Explain the following steps: when a user saves a file, it gets broken into chunks and encrypted. Those encrypted chunks are distributed across multiple servers in different geographic locations for redundancy. When the user accesses the file from another device, the app detects which chunks are already on that device and only downloads the new or modified ones. Show what happens when the same file is edited on two different devices at the same time — both devices make conflicting edits and save them. The system detects the conflict, preserves both versions, and lets the user choose which one to keep. Label each component clearly — user device 1, user device 2, file chunks, encryption, chunk servers (distributed), sync service, conflict detection, version history. Show how data flows between devices and servers, and how the system handles the sync and conflict scenarios. Render the video as output3.mp4

Output artifact: Output artifact (Video file): output4.mp4

What changed: Artifact transformed into Video file

Why it matters / Conclusion: All three outputs rendered smoothly — but jump to 0:10 in the first render that shows the ranking visualization. The sequence is rendered with a sophistication and detail that goes beyond what we expected. Compare our original prompt against what rendered above.

Antigravity automatically creates the full Remotion project and renders it to MP4 without requiring a manual render pipeline, through its agent.

TEXT

Using Remotion library, Create an animation video explaining how search engines work. Render the video as output.mp4

↓→

VIDEO

TEXT

Using Remotion library, Create an animation video that explains how Retrieval-Augmented Generation works. A user query gets embedded into a vector. That vector searches a vector database to retrieve similar text chunks. If no relevant chunks are found above a confidence threshold, show the system returning a "no relevant context found" message. If relevant chunks are found, those chunks are combined with the original query into a prompt. That prompt is sent to an LLM which generates a response. The LLM response is shown to the user. Additionally, show a feedback loop where the user can rate the response as helpful or unhelpful — if unhelpful, the system re-runs the vector search with a modified query to try retrieving different chunks, then sends the new chunks to the LLM for a second attempt. Label each component clearly — user query, embedding, vector database, confidence threshold, retrieved chunks, prompt assembly, LLM, response, feedback, and query refinement. Show how data flows between each component, including the retry loop when initial retrieval fails. Render the video as output2.mp4

↓→

VIDEO

TEXT

Using Remotion library, Create an animation video that explains how cloud storage services like Google Drive or Dropbox work. Show a title card with the text "How Cloud Storage Works" at the start. Explain the following steps: when a user saves a file, it gets broken into chunks and encrypted. Those encrypted chunks are distributed across multiple servers in different geographic locations for redundancy. When the user accesses the file from another device, the app detects which chunks are already on that device and only downloads the new or modified ones. Show what happens when the same file is edited on two different devices at the same time — both devices make conflicting edits and save them. The system detects the conflict, preserves both versions, and lets the user choose which one to keep. Label each component clearly — user device 1, user device 2, file chunks, encryption, chunk servers (distributed), sync service, conflict detection, version history. Show how data flows between devices and servers, and how the system handles the sync and conflict scenarios. Render the video as output3.mp4

↓→

VIDEO

Bottom Line

All three outputs rendered smoothly — but jump to 0:10 in the first render that shows the ranking visualization. The sequence is rendered with a sophistication and detail that goes beyond what we expected. Compare our original prompt against what rendered above.

Full Remotion Project Code

Strong — full code transparency and customizability without platform lock-in.

8/10

▾

Test Summary

Feature tested: Full Remotion Project Code

Result: Passed (8/10) — Strong — full code transparency and customizability without platform lock-in.

Generated output is a complete, editable Remotion React project with full source code accessible. You own the codebase, can modify it at any level, version-control it, and integrate it into other projects.

TEXT

Input 1 (Search Engines): Using Remotion library, Create an animation video explaining how search engines work. Render the video as output.mp4

↓→

SCREENSHOT

Output artifact for "Full Remotion Project Code" test, image-166.png

Bottom Line

The generated code could have been one big file — but scan the component structure in the code output above to see how the agent broke it down. The separation reveals the agent's architectural thinking about organizing animation sequences.

Use Case Track Record

Generate code-based animations from text inputs

Pricing & Access

Plans as of May 2026

TESTED

Individual

Agent model: access to Gemini 3.1 Pro, Gemini 3 Flash, Claude Sonnet & Opus 4.6, gpt-oss-120b Unlimited Tab completions Unlimited Command requests Generous weekly rate limits

Developer Plan

via Google One

Everything on Individual and More generous rate limit Flexible AI credit pool

Pricing as of May 2026

Is This Right For You?

A side-by-side guide based on our hands-on testing.

✓ Use This If

●You want automatic MP4 rendering without manual pipeline setup

●You value full code ownership and want to modify generated projects

●You're comfortable with upfront framework setup (or agent-assisted setup)

●You're willing to iterate on first-pass output for visual polish

●You want structured generation with implementation review step before execution

✕ Skip This If

●You need one-click, zero-iteration output

●You want to avoid any setup (local or agent-assisted)

●You need publication-grade visual polish immediately without iteration

●You're non-technical and prefer GUI-only workflows without code visibility

Frequently Asked Questions

Antigravity's agent can handle installation during the workflow. Either way, the setup is straightforward.

Yes — complete code ownership. Full codebase is editable and can be customized, extended, and enhanced with agent.

Antigravity provides the full project code, which you can render yourself locally using npm start for remotion projects, ffmpeg+puppeteer or screen recording for other frameworks.