Sora and Generative Video: Which AI Announcements Companies Can Watch Calmly

Sora is a model from OpenAI that creates videos of up to one minute from a text description. For most companies the first task is to place the announcement: is the tool available, has it been proven in operations, and does it touch a process that matters for the business result? Five checking questions and a maturity model let you judge any AI announcement within a few hours.

OpenAI presented Sora in mid-February 2024. The sample videos show scenes with people and landscapes as well as camera movements, at a quality above what earlier video models showed. Within days, executive teams ask their marketing and communications heads whether the company has to be part of this. The pressure is real, and it deserves an answer with substance. This article explains what Sora can do and what remains open, and offers a method with which decision-makers can assess every AI announcement of the coming months.

What is Sora, and what can the model do?

Sora is an AI model from OpenAI that creates realistic-looking videos of up to one minute from text instructions. In its announcement of 15 February 2024, OpenAI describes a model that can render complex scenes with several characters and specific types of motion as well as accurate background detail. It can also develop a video from a still image or extend an existing video. In the accompanying technical report, OpenAI presents the model as a step towards systems that simulate processes of the physical world.

OpenAI names the limits itself. The model struggles to depict the physics of complex scenes accurately and does not always understand cause and effect: a person bites a cookie, and afterwards the cookie shows no bite mark. It also sometimes confuses spatial details such as left and right. Sora is not yet generally available. Access lies with safety researchers who probe misuse risks and with a selected group from art, design and film who give feedback. OpenAI has named neither prices nor licence terms, nor an interface for companies.

Other providers already offer generative video tools that create clips a few seconds long. Sora raises the bar for length and consistency but does not yet replace a process in product communication or training, because nobody outside the test group can work with it.

Does a company have to react to Sora now?

A company does not have to react to Sora now as long as the tool is unavailable and affects no key process. You should act where moving images form part of the value creation, for example at agencies and media houses or at manufacturers with products that need explaining. There it is worth naming a person who follows the development and prepares a test for when the tool becomes available.

In all other cases a simple monitoring process is enough. The communications department or IT names a responsible person who collects announcements, assesses them with the checklist and reports to the executive team at a fixed rhythm, for example once a quarter. The effort is a few hours a month. How competitive pressure can be placed in this situation is described in the article on AI competitive pressure in mid-sized companies.

How do you assess an AI announcement? Five checking questions

You assess an AI announcement with five questions that any manager can answer without understanding the technology. Together they produce a verdict on whether to act, observe or discard.

1. Availability: can an employee use the tool today, with a contract and support, or are there only demonstrations?
2. Maturity: are there reports from users who have run it in operations for weeks, and what error rates do they name?
3. Relevance: does the tool change a process that moves a metric the executive team cares about, such as cost per transaction or lead time?
4. Integration: can the tool be connected to existing systems, or does it become an isolated island?
5. Alternatives: does a tool that is available today solve the same problem well enough?

Whoever answers questions 1 and 2 with no, observes. Whoever answers them with yes and can name a metric for question 3 starts a limited test with a clear stop criterion. Question 5 deserves particular attention: a company that waits for the better tool forgoes the good one that already exists.

On this scale Sora sits at the start of the “limited preview” level. The verdict for most companies is therefore to observe and to note a metric against which a later test is measured. That includes the question of which videos the company produces today, what they cost and how much time they take.

Which applications are conceivable for companies?

Applications are conceivable wherever videos are short-lived or expensive to produce. They include product clips for online shops and explainer videos for training, as well as visualisations for sales conversations. In every case the model delivers raw material that people check, edit and approve.

For corporate use, properties count that rarely show in demonstrations. A brand needs recognisable characters and products across several videos, and tools have to prove that consistency first. Products must appear in correct form, since an invented detail on a device damages trust. And every published video passes a review by the company's specialists. The time saving lies where drafts and variants arise rather than in the final approval.

How do you calculate the benefit?

You calculate the benefit from the videos a company produces today and the share a tool can take over. The calculation has four steps. First record the number of videos per year and the cost per video including internal working time. Then determine the share for which generated clips would suit as raw material, for instance background scenes or product variants. Third, estimate the saving per video after a test with real tasks. Fourth, deduct the cost of the tool, of review and of training.

Whoever lacks these figures cannot yet assess a test. Recording the baseline is therefore the most sensible preparation a company can make at the moment. It costs little and makes every later test meaningful.

How do you stay able to act amid AI announcements?

You stay able to act when you turn the frenzy of announcement cycles into a fixed rhythm. New models appear at short intervals, and any week can bring a headline to the executive team. A company that reacts to each one ties up capacity in initiatives without results. A company that follows none misses the moment at which a tool becomes ready for operations. The article on the model race between Gemini and GPT-4 describes the same logic for large language models, and the article on AI activism shows how boards turn pressure into a plan that holds.

Success stories from other companies belong in this check as well. The questions about maturity and the metric apply to every case study, as the article on checking AI success stories explains.

Generative video with Niehues Consulting

Niehues Consulting is a consultancy for AI management and organisation design from a delivery perspective, based in Düsseldorf. Founder Marin Niehues brings AI into operations, in projects ranging from VC-funded scale-ups to DAX 40 corporations and in more than 20 countries on six continents. In the service area of AI management, the work runs from business analysis and requirements through the business case to the selection and sequencing of applications.

The approach follows the Niehues Delivery Model with its three phases. For assessing an announcement, three of its checking questions count: Which business result should change measurably? What is the baseline, and how do we measure success? Does the prototype prove the benefit in the real workflow with real data? The entry point is a free initial consultation of 30 minutes, which you can arrange via the contact page.