How much will it cost to make a feature film using Sora? 

OpenAI recently released more experiments created with Sora, tapping 7 leading creatives to create short films

While OpenAI has not given any indication on costs or pricing for its text-to-video model, it was fun diving into some projections based on current hardware assumptions. Factorial Funds recently released a blogpost diving into how the inference compute needed if Sora made say, 50% of Tiktok and Youtube videos released every day — and they landed on an estimate of a whopping 720,000 H100 Nvidia GPUs. For context, each H100 chip retails for $40,000 today. 

I was curious to put it into context of a feature film. Assuming a shooting ratio of 100:1 minutes, a H100 chip running continuously for 24 hours would produce 120 mins of “raw footage”. Following similar assumptions around fail rate and peak demand, we’d need 20 H100 chips running continuously for 20 days to create enough footage for a two hour film. 

And that’s just for inference compute. Throw in initial research and development costs, maintenance and troubleshooting, scaling, and Open AI’s pricing — and the costs start looking prohibitive to the indie creators and studios. 

This is where the game gets interesting. Will compute costs go down over time? Will free and open source alternatives to Sora —like Open-Sora, Mora, and Stable Video Diffusion — ever become comparable in quality? 

🌊 The Long Reads

With layoffs ramping up across the media industry, reporters are trying a new model: co-operative newsrooms of their own

🔥 Latest News

1.

OpenAI sets meetings with Hollywood in a bid to break into the movie business: OpenAI has announced plans to break into the Hollywood film industry, alarming some writers. The first official meetings to discuss OpenAI’s partnership with the film industry are poised to occur in Los Angeles this coming week, with executives hoping to utilize the artificial intelligence software to generate effects and video files.

2.

Intel is planning a sprawling U.S. expansion by building the largest AI chip manufacturing site in the world, in a new five-year, $100 billion plan. Under Biden’s CHIPS act, Intel received nearly $20 billion in federal funding and loans from the U.S. government,in a bid to increase US prominence in the AI arms race. 

3.

Adobe’s new GenStudio platform is an AI factory for advertisers: Adobe has announced a new AI-powered ad creation platform that aims to make it easier to use the company’s generative AI tools for building marketing campaigns. GenStudio provides access to brand kits, copy guidance, and preapproved assets alongside a bunch of generative AI-powered tools that can generate backgrounds and ensure the overall tone remains on-brand. 

Quick Hits

• Sakana AI released a new method to combine the 500,000 open-source models to build new ones

• Infinity launches at YCombinator. They make a script-to-movie foundation model that allows people to generate talking-head videos just by training it on a few minutes of an existing video. 

• Meet Hume’s Empathic Voice Interface (EVI), the first conversational AI with emotional intelligence.

• Character AI launches a new feature that adds voices to characters with just 10 seconds of audio

• Blackbird adds indexing and AI auto sequencing to its video editing platform enabling editors to locate specific storytelling moments within their video content using natural language. 

• Benchmark invests $60M in HeyGen, an AI video startup with a valuation of $440 valuation