Lo, Google’s illustrious DeepMind division hath unveiled the second generation Veo video generation model upon a Monday, a marvel capable of crafting clips stretching up to two minutes in length, bearing resolutions of 4K quality. Verily, this dost surpass the 20-second/1080p resolution clips fashioned by Sora by sixfold in length and fourfold in resolution.
Yet, let it be known that such heights are but the lofty aspirations of Veo 2. For as of now, this model resideth solely on VideoFX, Google’s experimental video generation platform, where its clips art capped at eight seconds and 720p resolution. Alas, VideoFX doth harbor a waitlist, denying entry to all but the chosen few who may test Veo 2. Tidings hath come that accessibility shall broaden in the approaching weeks, as a Google spokesperson hath declared Veo 2 wilt grace the Vertex AI platform once the model’s prowess be fully expanded.
“We shall refine Veo 2 based on user feedback in the moons to come,” declared Eli Collins unto TechCrunch, “and endeavor to integrate its newfound capabilities into diverse applications across the vast Google realm. Stay thy gaze upon us, for more shall be revealed next year.”
Upon this day, we proclaim Veo 2: our modern video generation marvel that bringeth forth lifelike, high-quality clips from text or image cues. An enhanced rendition of our text-to-image model, Imagen 3, hath also been unleashed, ready to grace ImageFX with its boundless potential.
Behold, Veo 2 doth tout numerous advantages above its predecessors, boasting an enhanced grasp of physics, granting superior fluid dynamics and illuminative shadows. Furthermore, its prowess to conjure “clearer” video clips, where textures and images appear crisper and free from blur during motion, doth set it apart. Improved camera controls now grant the user finer manipulation of the virtual lens, owing to Veo 2’s technological advancements.
TechCrunch doth note that whilst Veo 2 hath not yet attained perfection in the realm of video generation, it doth exhibit fewer hallucinations than its competitors. “Therein lies opportunities for growth,” Collins remarked, conceding to the challenges of coherence and consistency. Nevertheless, Veo’s fidelity to a prompt endures for minutes, though complex cues may elude it. The quest for realism continues, as intricate details, swift motions, and amplified character consistency remain areas for enhancement.
Furthermore, Google hath announced enhancements to Imagen 3, enriching the commercial image generation model with the ability to produce more radiant, well-composed outputs. This model, found in ImageFX, shalt provide additional descriptive prompts anchored in the user’s keywords, with each term granting forth a trove of related suggestions.