Veo 3.1 Is A Way Bigger Upgrade Than We Thought

Google's Veo 3.1 introduces significant new video generation and editing features, offering more creative control, though it still lags behind Sora 2 in realism and physics but surprisingly allows generation of trademarked IP.

• Veo 3.1 significantly enhances video generation with multi-image referencing and frame-to-frame animation.

• Veo 3.1 offers more editing control and is surprisingly permissive with trademarked content, unlike Sora 2.

• Sora 2 remains superior in realism, physics, and initial video quality, despite its stricter content guardrails.

Google has launched Veo 3.1, a major update to its video generation model, introducing 'ingredients to video' for multi-image referencing, 'frames to video' for animating between start and end images, and enhanced video extension and editing capabilities. While Veo 3.1 offers greater customizability and is more permissive with trademarked content than Sora 2, it still falls short of Sora's realism, physics accuracy, and initial video quality, requiring more editing iterations to achieve desired results.

Key Veo 3.1 Features

• 00:01:33 Veo 3.1 introduces several new features, including 'ingredients to video,' which allows users to control characters, objects, and style using multiple reference images. The 'frames to video' feature enables animation between a specified first and last image, while enhanced video extension capabilities allow for longer, stitched-together clips. Additionally, users can now add new elements to generated scenes, though the ability to remove unwanted objects is 'coming soon'.

Access and Model Options

• 00:03:19 Veo 3.1 is accessible through Google's Flow platform (labs.google.com/fx/tools/flow) and is available in over 150 countries, requiring a Google paid plan. Users can choose between 'Veo 3.1 Fast' for quicker, potentially lower-quality generations and 'Veo 3.1 Quality' for higher quality but longer processing times. Notably, the 'ingredients to video' feature currently only works with 'Veo 3.1 Fast'.

Comparison with Sora 2 (Physics & Realism)

• 00:08:24 When compared to Sora 2, Veo 3.1 struggles with realistic physics, such as accurate backflips, where Sora demonstrates superior performance. While Veo's initial video generations are often not as satisfactory, requiring further editing, Sora generally produces more realistic videos, excels at generating people, and offers better initial one-shot prompt results. However, Sora has recently implemented stricter guardrails, refusing to generate content with trademarked IP that Veo 3.1 currently allows.

Trademarked Content and Guardrails

• 00:19:06 Surprisingly, Veo 3.1 allows the generation of videos featuring trademarked characters like Mickey Mouse, Super Mario, Batman, and Spongebob, which was not expected from a Google product. In contrast, Sora 2, which previously had lax guardrails, now frequently refuses to generate such content due to violations. This difference in content moderation gives Veo a temporary advantage for 'fanfiction' type creations, though it's anticipated that Google will eventually tighten its guardrails.