| Spec | Value |
|---|---|
| Maker | Google DeepMind (Google) |
| Released | 16 Dec 2024, VideoFX waitlist (Google Labs) |
| Max resolution (Google claim) | Up to 4K (4096 x 2160) (Google DeepMind; TechCrunch) |
| Max resolution in VideoFX | 720p cap (TechCrunch) |
| Clip length (Google claim) | Extended to minutes in length (Google blog); two minutes plus (TechCrunch) |
| Clip length in VideoFX | 8 seconds cap (TechCrunch) |
| Input types | Text prompt; text plus a reference image (TechCrunch) |
| Access | VideoFX waitlist; Vertex AI later, 'as the model becomes ready for use at scale' (TechCrunch) |
Google's headline 4K and minutes-long figures were not what VideoFX delivered at launch; both readings are shown with their source. · source
| Spec | Value |
|---|---|
| Model code | veo-2.0-generate-001 |
| Resolution | 720p |
| Frame rate | 24 fps |
| Clip length | 5 to 8 seconds |
| Aspect ratios | 16:9 and 9:16 |
| Videos per request | 1 or 2 |
| Native audio | No (silent only) |
| Input types | Text-to-video; image-to-video |
| Request latency | Min 11 seconds; max 6 minutes at peak |
| API price (paid tier) | $0.35 USD per second of video |
Gemini API video docs as archived 30 May 2025 (Veo 2 GA in the API from April 2025). Price is from the Gemini API pricing page archived 31 May 2025 (web.archive.org/web/20250531063345/https://ai.google.dev/gemini-api/docs/pricing). ·
Estimate from the version's published specs and reviews, 0-5; scores are only comparable within one criteria year. Scored as shipped (16 Dec 2024): Google claims 'up to 4K' and 'minutes in length', but TechCrunch and VideoProc found VideoFX and AI Studio output capped at 720p, 24 fps and 8 seconds, so resolution is scored at 720p and duration at 8 s (disputed: 4K would score 5, minutes-long 5). Silent; text or text plus one reference image. No speed figure is published. Realism: Beebom and VideoProc praise visuals and faces but flag physics errors; Tom's Guide's praise is from Google's own samples ('I haven't tried Veo 2 myself'); TechCrunch says it can't clear the uncanny valley.[1]
Veo 2 is the state-of-the-art video generation model and outperforms all competing AI models, including OpenAI’s Sora.
Veo 2 excels in producing visually stunning and detailed videos and offers impressive control over camera work through prompting.
Veo 2 can’t quite clear the uncanny valley.
a new version of the Veo artificial intelligence video model that has the most accurate understanding of physics I've seen from any video tool so far.
| Google DeepMind unveils a new video model to rival Sora · 16 déc. 2024 | -0.40 |
|---|---|
| Google Veo 2 is one of the best AI video models I've ever seen — here's 5 examples of what it can do · 17 déc. 2024 | 0.80 |
| Google Veo 2 Hands-On: Stunning AI Generated Video Visuals But Weak Physics · 27 avr. 2025 | 0.70 |
| Google Veo 2 Review (Tested): Is It Truly Worth the Hype · 16 mai 2025 | 0.70 |
Photos de lecteurs, distinctes des médias sourcés du pin. Notez ce que vous avez mangé ou vu et, sur un pin de restaurant, identifiez les plats directement sur votre photo.
La première entrée est toujours la source de l’épingle. La confiance globale est une moyenne pondérée de la fermeté avec laquelle chaque référence étaye les heures de début et de fin retenues ci-dessus ; une référence compte moitié moins pour chaque tranche de 180 jours d’ancienneté par rapport à la plus récente.
Google's post 'Updates to Veo, Imagen and VideoFX, plus introducing Whisk in Google Labs' is dated 16 Dec 2024 and says 'Today we're introducing a new video model, Veo 2'; TechCrunch's[5] report of the same day says 'On Monday, DeepMind announced Veo 2'.
Free AI Studio output limited to 720p, 24 fps and 8-second clips, against Google's 4K and minutes-long claims.[1]
Hands-on test finding believable motion and faces but a glass that hangs in the air, and trouble with multiple subjects.[1]
Praise for physics based on Google's own samples; the writer says he had not tried Veo 2 himself.[1]
Says Veo 2 can make two-minute-plus clips up to 4K in theory, but VideoFX caps videos at 720p and eight seconds, and that it takes text or text plus a reference image.[1]
Il manque quelque chose ou une erreur s'est glissée ? Dites-le avec vos mots : un lien qui étaye cette épingle, une autre date de début ou de fin et pourquoi, ou une information absente ou erronée. L'IA la vérifie au regard des sources de l'épingle, en cherche de meilleures et ajoute toute page qui vous donne raison. Les sources de l'épingle restent celles qui comptent le plus. L'IA regarde aussi les images : une image qui montre autre chose, ou qui le montre mal, passe en dernier ou est remplacée.