Below is a submission for a short film competition (Runway Studios Gen:48) using Runway Gen-2. It seems that over 75% of the video must be made using Gen-2, but the quality is extremely high.

Link: https://gen48.runwayml.com/winners

https://twitter.com/aoyama_pokare/status/1728015809350664291/video/1


• Storyline
• Video effects
• Composition
• Tone and style
• Sound effects

The attention to detail in each of these elements is extraordinary.

Although that the film was made by a professional team, the level at which professionals utilize AI is on a different scale.

However, the charm of AI lies in its ability to enhance the abilities of even amateurs, making it possible for them to create something like this if they put in the effort.

AI is the strongest partner, so centering on upgrading oneself using AI seems like a good approach.

Regarding the specific details of the competition, the rules require participants to create a 1-4 minute video primarily using Runway tools, with at least 75% of the total video content generated using Runway's generative tools, such as Gen-1, Gen-2, or any Text-to-Image tool.

A High-Quality Short Film Utilizing AI

Inflection 2: Redefining AI Performance and Efficiency
Inflection just finished making Inflection-2. It’s one of the best big computer models in the world, better than their earlier model. It knows more facts, writes better, and thinks more clearly.
https://inflection.ai/inflection-2

Tech Specs: Inflection-2 was made using 5,000 special NVIDIA H100 computers. It’s as powerful as Google’s big PaLM 2 model, but even better at many computer tests. It also works faster and costs less to run than Inflection-1.

Safety & Security: When making Inflection-2, the team paid a lot of attention to making it safe and trustworthy. They follow strict rules to make sure the model is good and safe for everyone.

Performance Benchmarking: Before Inflection-2 is shared with others, it is compared with other big models like LLaMA-2 and GPT-4. It does really well, especially in a test called MMLU, where it’s almost as good as GPT-4.

Comparison with GPT-4: In some tests, like HellaSwag, Inflection-2 got a score of 89.0, which is close to GPT-4’s score of 95.3. But we can’t fully compare them because they were tested differently.

Coding & Math Capability: Even though Inflection-2 wasn’t mainly made for coding or math, it still does really well in those areas. This means it could be made even better for coding in the future.