Ovi: Twin backbone cross-modal fusion for audio-video generation
Visual quality, uncanny valley, and aesthetics
- Many find Ovi “mindblowing” yet firmly in the uncanny valley: odd facial expressions, anatomical glitches (e.g., extra limbs), and inconsistent scenes.
- Comparisons are made to CGI in mainstream films: when done well you don’t notice; when you notice, it’s usually budget or execution.
- Some argue AI video should lean into stylization/lo‑fi aesthetics (anime like Dandadan, Chainsaw Man) rather than chase realism that amplifies uncanny artifacts.
Models, open source, and technical details
- Ovi’s video component is reported as being based on Wan 2.2, with audio from MMaudio.
- Commenters highlight growing strength of open Chinese video models as credible alternatives to big proprietary systems.
- Some see no strong “moat” in the core tech; moats are expected to come from distribution, tooling, integrations, and IP deals, not the base models.
Performance, hardware, and hosting
- Ovi can run on a single high-end GPU (32GB VRAM), making realistic fakes broadly accessible.
- People discuss cheap cloud access (sub‑$0.50/hr shared GPUs) and splitting large accelerators (e.g., MI300X) into smaller slices for hobbyists.
- Experiences with generation time vary widely (minutes vs. hours) and may depend on settings; some note better results with newer CUDA/Torch and specific attention kernels.
Use cases, potential, and limits
- Enthusiasts already use Ovi locally to produce surprisingly real-looking clips, but describe it as a “slot machine” requiring many runs.
- Hopes include bringing manga/anime to life, fan-made adaptations of novels, and party-style collaborative movie generation.
- Others struggle to see positive use cases, viewing current output as more nuisance than benefit.
Ethics, misuse, and company criticism
- Significant concern about realistic deepfakes: fake videos to harm reputations are seen as imminent.
- Criticism of the hosting company’s broader business model: accusations of exploiting lonely and underage users with AI “relationships” and harvesting intimate data.
Future of movies and cultural acceptance
- Debate centers on whether we’ll see a <$1000 “blockbuster” from one person.
- Skeptics emphasize missing pieces: coherent character continuity, fine-grained control, good writing, and acting.
- Some argue audiences inherently dislike AI-made art; others think resistance will fade once content is indistinguishable and personalized media becomes normal.
- There’s disagreement over inevitability: some foresee AI-driven hits or new formats (daily AI soaps), others doubt AI movies will ever be widely embraced as “art.”
Scams and name confusion
- Commenters note opportunistic domain squatters/SEO sites popping up around new open models, often without real services.
- Side thread on the older Nokia “Ovi” brand, with nostalgic commentary and confusion over the shared name.