Google announced that it will be releasing Google Chrome. Veo 3.1 LiteThe new tier of models within the generative video product portfolio is designed to solve the main bottleneck in production scale deployments: Pricing. The generative video market has made rapid improvements in terms of visual quality, but the cost per second generated content is still high and prohibitive to developers who build high volume applications.
Veo 3.1 is available now via the Gemini API The following are some examples of how to get started: Google AI Studio For users on the paid level. Google has positioned this model to be the industry standard by offering the Veo 3.1.1 Fast model with the same production speed at around half the cost.
Diffusion Transformator (DiT), a Technical Architecture
Veo 3.1 is a family of products that has a number of unique features. Diffusion Transformer architecture. The traditional generative video model relies on U-Net diffusion. This can be problematic with large data sets and time-dependent dependencies.
Veo 3.1 Lite is a lightweight version of Veo 3.1 that uses a transformer as underlying backbone. The use of spatiotemporal patterns. This architecture does not process video frames as 2D static images, but rather as tokens that are continuously arranged in latent space. This model can maintain its stability by using self-attention on these patches. Temporal consistency. This reduces the number of artifacts, such as pixelation, that are common in older models.
Model performs computations in compressed format latent space instead of pixels. It can then handle video creation’s high computing demands while maintaining its smaller memory footprint. Developers can use this model to generate high-definition material without having their compute times increase exponentially.
Performance and output specifications
Veo 3.1 Lite has specific parameters that allow AI developers to easily integrate the product into their workflows. The Lite Veo 3.1 version, unlike the flagship Veo 3.1 that supports 4K, is optimized for HD outputs.
- Supported Resolutions Both 720p and a 1080p version are available.
- Aspect Ratios: Both landscape (16:9) orientations and portrait (9/16) orientations are supported natively.
- Video Clips The developer can choose between 4, 6 or 8 second generation.
- Prompt Adherence The model is optimized for ‘Cinematic Control,’ recognizing technical directives such as ‘pan,’ ’tilt,’ and specific lighting instructions.
The ‘Lite’ tag does not refer to a reduction in generation speed compared to the ‘Fast’ tier. The ‘Lite’ tag refers to a set of optimized parameters which allows Google to sell the model for a much lower price, while retaining the low-latency characteristics.
The Price Shift: Demonizing Video Inference
Veo 3.1’s cost structure is the core of its value proposition. The current market is characterized by high-quality videos inference that can cost several dollars for a minute. This makes it very difficult to justify applications such as dynamic ad creation or social media automation.
Veo 3.0 Lite is priced as follows:
- 720p: Not more than $0.05 per minute.
- 1080p: The cost of a second is $0.08
Gemini API and AI Studio for deployment
Accessibility is managed through Gemini API. It allows the generation of videos to be integrated into Python and Node.js apps using REST calls or gRPC.
The inclusion of a scalable, dependable and secure. SynthID. SynthID was developed by Google DeepMind as a way to watermark and identify AI-generated material. This embeds an invisible digital watermark into each pixel of the video. However, specialized software can detect it. It is mandatory for all developers who are concerned about safety, compliance and separating synthetic media from real footage.
What you need to know
- Get the same quality at Half the Price: Offers the same low-latency performance as the ‘Fast’ tier at less than 50% of the price ($0.05/sec for 720p).
- Scalable High Definition OutputSupports native aspect ratios of 16:9 or 9:16 for 720p, 1080p in clips lasting 4, 6 and 8 seconds.
- ArchitectureDesigned on a Diffusion Transformer Using spatiotemporal patches to improve motion and consistency.
- Developer ReadyAvailabile now Gemini API Google AI Studio is a paid-tier product that includes Google AI Studio. SynthID Digital watermarking.
Check out the Technical details. Access to the model can be gained via the paid tier of the Gemini API The following are some examples of how to get started: Google AI Studio. Also, feel free to follow us on Twitter Don’t forget about our 120k+ ML SubReddit Subscribe Now our Newsletter. Wait! Are you using Telegram? now you can join us on telegram as well.
Michal is a professional in data science with a Masters of Science degree from the University of Padova. Michal Sutter excels in transforming large datasets to actionable insight. He has a strong foundation in statistics, machine learning and data engineering.
