No ratings yet.

Filmmaking has always been a time-consuming, expensive business. Until now. Find out how our start-up is rewriting the rules of the film industry with generative AI (GenAI). This start-up focuses on automating different facets of a film’s production process. The goal is to streamline film production, reduce costs and allow new creative opportunities into the production process. These goals will be realised through five core activities: idea generation, scriptwriting, character design, video creation and voice generation. These core activities and the market approach will be further explained in this blog.

Having laid the foundation for how GenAI can transform the film industry, we now dive deeper into the ways our start-up is using this technology to innovate every aspect of the production process. The first application, ‘idea generation,’ speeds up idea generation using GenAI. This includes ‘scriptwriting,’ where models develop a script based on the generated idea. Then, based on the script, a text-to-image model develops characters, which designers further refine. In video creation, GenAI generates complex visual scenes, which speeds up production. Voice generation creates realistic voices based on text, which simplifies the voice-over process and saves time. Thus, many manual steps are automated and overall production time is reduced.

Our start-up targets a wide range of customers, ranging from large film studios to independent filmmakers. To best serve these different groups, we have developed a multi-channel strategy. Licences will be sold to large studios, letting them use the product freely. To smaller independent filmmakers, Saas subscriptions will be sold. This will ensure that we can attract the top and bottom end of the market, allowing this technology to gain wider adoption within the film industry.

To analyse our differentiation in the film industry, we used the VRIO model. This analysis shows that our start-up has a unique competitive advantage thanks to the early integration of GenAI. Indeed, Generative AI has not yet been widely adopted in film production. Therefore, our advantage lies in the first mover position we aim to achieve, also partly due to offering two types of our product. In addition, the VRIO model also reflects that there are indeed challenges. Namely, there are high costs associated with developing generative AI. Also, the film industry is a market dominated by established players with high budgets which can make it difficult to get a foothold.

In conclusion, this start-up could as of now transform the film industry using GenAI by deploying a more efficient production process. Various aspects of the production process can be developed with lower costs through automation and creative capabilities are expanded. The focus on major film studios and independent filmmakers ensures broad market access. Although there are challenges, our first mover strategy gives us a competitive advantage to transform this industry.

No ratings yet.

I’ve recently experimented with video-generative AI, trying various models and tools. I started my journey by trying out KlingAI and Pikalabs. Out of these two, I preferred KlingAI as it delivers more realistic videos. However, it is rather slow, often taking 15 minutes or more to produce a 5-second video. So after trying these two, I was not completely satisfied so I did some more research and came across Runway.

I first tried out their free Gen-2 model, while this model can produce decent videos using image-to-video or text-to-video techniques, their Gen-3 model is miles ahead. To illustrate this difference I used the same prompt in the Gen-2 and Gen-3 models.

As you can see the Gen-3 model produced a video which adheres better to the prompt and is far more realistic than the Gen-2 output. Whilst you could go and make yourself a cup of coffee and some lunch while KlingAI was generating your video, you don’t have to with Runway. A 5-second video will take about 60 seconds to generate and a 10-second video will take about 90 seconds. The only big downside of Runway Gen-3 is the cost, a 10-second video will set you back about a dollar in credits. If every prompt results in the desired outcome, which does not always happen, a minute of video will cost you at least 6 dollars. I made a 2-minute video which cost me probably five times that amount. This is only partly due to failed generations. The other side of the story is that it takes a bit of time to figure out the model and get to know the prompt structure that works best. To maximize the effectiveness of my prompts I used a GPT called  “Runway Prompt Pro”. This GPT will convert your input into a prompt specifically tailored to Runway’s model. I’ll show you an example:

My input: drone shot flying over an active volcano

Runway Prompt Pro: [drone shot, smooth movement]: The drone flies over an active volcano, capturing the molten lava bubbling inside the crater. Ash clouds rise from the fiery pit, glowing red and orange in the twilight. Lava streams flow down the dark mountainside. The surrounding rugged landscape contrasts the intense heat and vibrant colours. [wide-angle, steady]

I used this prompt as part of the video I made, resulting in the following video:

Using this prompt GPT significantly boosted my first generation’s success rate. The video above was the result of the first generation. I experienced quite a good first-generation success rate when generating landscape videos. I however had some more difficulties when I tried to make videos of people and animals. While Runway is still very good at this, it might take a few more tries. Because generations are not cheap this can be quite frustrating. For this reason, Runway is often combined with Midjourney. Midjourney allows you to make hyper-realistic imagery which Runway can then transform into videos. If you are serious about video making with AI, I would recommend this combo.

No ratings yet.

On the 3rd of September, the Dutch Data Protection Authority (Autoriteit Persoonsgegevens) fined the American Clearview AI 30,5 million euros, the second highest fine ever given by the DDPA (NOS, 2024). The fine was imposed on Clearview AI, a facial recognition software company, for building an illegal database containing over fifty billion pictures. The pictures were scrapped from the web, meaning everyone with an online presence could be in the database, even you. The record for the highest fine went to Uber only a week earlier. Uber shared data it had collected on its drivers with its headquarters in the United States, without taking proper safety measures. Uber is set to pay the DDPA a fine of 290 million euros (NOS, 2024a). These fines might pale in comparison to the ones imposed by the EU on tech giants like Google and Facebook. However, they contribute to an increasing pushback from regulators on tech companies and their usage of data. Besides regulators, consumers are also increasingly more concerned with their (online) privacy. An interesting point to mention here is that while their privacy concern has grown in the past two decades so has the voluntary sharing of data (Bartneck et al., 2020). 

As seen in the first example, AI companies collect vast amounts of data to operate and improve their AI models. While on the one hand, it does seem necessary if we want to be able to apply AI more and more in our daily lives, it does also come with notable risks. One of the primary issues identified by Bartneck et al. (2020) is the possibility that the gathered data is not used for its intended purpose. The surge of AI has also introduced new ways in which our data can be used. As the general usage of AI is still in its early days most of us are not fully aware of the issues these new uses might pose. These might include more obvious examples like impersonations and fake news, but also less obvious ones like predicting mortality. Bartneck et al. (2020) mention that AI has the potential to predict someone’s mortality by analyzing their movement. The resulting analysis could be used by or sold to undertakers or insurance companies. It is therefore important that users become increasingly aware of how their data is used. In order to realize this further transparency on data collection, safety and usage is required. 

This post is not meant as a plea against AI or the collection of user data. I would however like to make you think about what you want to share online and with whom. So maybe next time you will look into your cookie allowances instead of hitting the accept all button. 

References 

Bartneck, C., Lütge, C., Wagner, A., & Welsh, S. (2020). Privacy Issues of AI. In SpringerBriefs in ethics (pp. 61–70). https://doi.org/10.1007/978-3-030-51110-4_8

NOS. (2024a, augustus 26). Privacywaakhond legt hoogste straf ooit op: 290 miljoen euro boete voor Uber. https://nos.nl/artikel/2534629-privacywaakhond-legt-hoogste-straf-ooit-op-290-miljoen-euro-boete-voor-uber

NOS. (2024b, september 3). Boete van privacywaakhond voor verzamelaar van miljarden foto’s van gezichten. https://nos.nl/artikel/2535633-boete-van-privacywaakhond-voor-verzamelaar-van-miljarden-foto-s-van-gezichten