Veo

Original Article ↗ Hacker News Discussion ↗

Google’s unveiling of Veo, a high-end text-to-video model intended to rival OpenAI’s Sora, is met with mixed reactions: many find the demos technically impressive but visually underwhelming or overly safe, especially compared to Sora’s longer, more dynamic clips. Commenters criticize Google’s muddled branding (Veo vs VideoFX), georestricted waitlists, and reluctance to show realistic humans, seeing this as part product immaturity and part PR/safety overcorrection after past Gemini missteps. Beneath the model-vs-model rivalry, there’s a broader anxiety about AI-generated video flooding platforms like YouTube and TikTok, the erosion of human creative workflows, and whether Google’s management and incentives allow it to capitalize on its vast data and research lead.

Overall reception and comparisons

  • Many find Veo’s demos less impressive than OpenAI’s Sora: clips are shorter, often slow‑motion or simple pans, with limited complex motion and few humans.
  • Some argue expectations shifted unrealistically fast; a few months ago this would have seemed astonishing, and it’s still a major technical leap.
  • Others note Sora’s best-known short was heavily edited with VFX, so direct demo-to-demo comparison is misleading; both Veo and Sora remain unreleased.

Product access, branding, and UX

  • Veo is only accessible via a VideoFX waitlist; many complain about:
    • Region blocking (especially EU).
    • Multiple sign‑ins, broken forms, and needing to re-enter email.
    • Confusing naming: Veo (model) vs VideoFX (tool) vs other “FX” products.
  • Some see this as emblematic of recent Google I/O: lots of demos and waitlists, little immediately usable product.
  • Parallel discussion of GPT‑4o: text model is widely available to paid users, but voice/video features are not; roll‑out is uneven and confusing.

Capabilities and limitations

  • Strengths: highly polished “stock-footage” style shots, timelapses, scenic B‑roll, depth‑aware camera moves, masked edits, and image‑to‑video.
  • Weaknesses:
    • Poor continuity across shots and limited control over exact actions or camera coverage, reducing usefulness for serious filmmaking.
    • Artifacts and uncanny motion (e.g., horse/camel gait, cars merging into ground, surreal Northern Lights).
    • Some prompts not fully followed; Google is at least transparent that outputs aren’t perfectly prompt-faithful.

Safety, humans, and censorship

  • Notable lack of human-heavy clips; commenters speculate about:
    • Ongoing Gemini image controversies (race, WW2 depictions).
    • Nudity/objectification concerns and PR risk.
  • Some argue safety filters often degrade quality or block benign content.

Watermarking and misuse

  • Veo videos are watermarked with SynthID; it also extends to images, text, and audio.
  • Commenters question:
    • Whether text watermarking will be noticeable or harm quality.
    • How any watermark meaningfully prevents deepfake propaganda, since powerful actors can run unwatermarked models.

Broader impact and Google’s strategy

  • Fears of AI‑generated video spam, TikTok/Shorts auto‑content, and “infinite jest”‑style ultra-personalized distraction; some note this is already emerging, especially for porn and low-effort monetized clips.
  • Mixed views on Google:
    • Critics: squandered AI lead, over‑cautious, ad‑driven, confusing product strategy, history of killing products.
    • Defenders: research strength, huge context windows, longstanding core products, and meaningful if imperfect catch‑up with OpenAI.