Google Veo 3 just wrecked the video industry

Key Concepts

VO3, Google Flow, Gemini, text-to-video, image-to-video, frames-to-video, ingredients-to-video (reference to video), AI video generation, prompt engineering, limitations of AI video generation, censorship, pricing plans (Pro, Ultra), free access methods, lip-sync, human anatomy, jiggle physics, fight scenes, camera movements, video resolution, aspect ratio (16:9), voice control, music synchronization, sports game generation, ELO score.

VO3 Overview and Access

Google's VO3 is a god-tier AI video generator capable of creating audio and video from text prompts. It's accessible primarily through Google's Flow platform and, with limitations, through Gemini for pro users or above. Flow offers more options than Gemini, including image-to-video and reference-to-video features.

Prompt Engineering and Capabilities

• Text-to-Video: VO3 excels at understanding detailed prompts, allowing for step-by-step scene descriptions and dialogue specification using quotation marks.
  • Example: A Zoom call between four colleagues, one farts loudly, others react, with dialogue like "Who was that?".
• Settings and Character Specification: Detailing settings (e.g., "cozy, dimly lit studio") and characters (e.g., "Indian man with a thick Indian accent") improves results.
• Gameplay Generation: VO3 can generate video gameplay, such as GTA 6 or Starcraft, with varying degrees of accuracy.
  • Example: "Gameplay of GTA 6" or "Live gameplay of Starcraft with the gamer's webcam in the bottom left corner."
• Emotion and Tone Control: Specifying emotions (e.g., "a woman looks extremely sad") and tones (e.g., whispering) influences the generated dialogue.
  • Example: A woman whispers ASMR-style, or a man screams excitedly.
• Singing and Different Languages: VO3 can generate singing in different languages, as demonstrated with a K-pop girl group example.

Limitations and Flaws

• Text Accuracy: VO3 often generates gibberish or misspelled text in videos.
• Accents: It struggles to accurately reproduce accents in speech.
• Full Body Generation: Generating full-body videos, especially with fast movements, often results in warping and inconsistencies.
• Vertical Videos: VO3 cannot generate vertical videos, limiting its use for platforms like TikTok and YouTube Shorts.
• Voice Control: Users cannot control the voice of the characters, making it difficult to create consistent characters across multiple videos.
• Music Synchronization: VO3 struggles to synchronize music with instrument playing, as seen in the piano and violin examples.
• Sports Game Generation: It performs poorly when generating sports games like soccer or tennis, with inaccurate player positioning and ball movement.
• Censorship: VO3 has strict censorship policies, preventing the generation of most celebrities or existing people directly.

Image-to-Video (Frames to Video)

• Functionality: Allows users to upload an image as a start or end frame for video generation.
• Workarounds: Can bypass some censorship by using images of multiple people (e.g., Will Smith and Queen Elizabeth).
• Limitations: Dialogue cannot be added to characters in image-to-video. It can only generate horizontal 16:9 videos.

Ingredients-to-Video (Reference to Video)

• Functionality: Allows users to upload reference characters or objects to include in the video.
• Limitation: Currently only works with V2, not VO3.

Human Anatomy and Physics

• Jiggle Physics: VO3 can generate realistic jiggle physics, as demonstrated in the bikini-clad women example.
• Anatomical Accuracy: Struggles with complex movements like gymnastics flips, often resulting in anatomically incorrect poses.

Fight Scenes and Camera Movements

• Fight Scenes: VO3 can generate realistic fight scenes with accurate punches and kicks, minimizing warping and inconsistencies.
• Camera Movements: It can handle complex camera movements, such as tracking a race car and zooming into a blimp window.

Pricing and Free Access

• Pro Plan: $20/month, limited access to VO3 (around 20 videos per month).
• Ultra Plan: $250/month, more credits for VO3 (over 80 videos per month).
• Free Access:
  • One-month free trial of the Pro plan (requires payment details and cancellation).
  • Free Pro plan for students for 15 months (requires student email verification).

Monica AI Assistant (Sponsored Segment)

Monica is an AI assistant that provides access to various AI tools, including GPT, DeepSeek, Gemini, Flux, Stable Diffusion, Cling, and High Law. It offers features like summarizing web pages and YouTube videos, generating mind maps, and is available as a browser extension. A discount code "AIARCH10" is provided.

Conclusion

VO3 is a powerful AI video generator with impressive capabilities, particularly in generating realistic interviews, fight scenes, and understanding complex prompts. However, it has significant limitations, including censorship, lack of voice control, struggles with full-body movements, and inability to generate vertical videos. Despite these flaws, it remains the top-rated video generator based on the Artificial Analysis leaderboard, with an ELO score of 1247.