SORA Uses 10k GPUs This AI Uses Just 32 (Fast, Cheap, AI Video For ALL)
TLDRHicks Field, a startup company, is making waves in the AI video generation industry with their innovative approach. They've developed a foundational AI video generation model that, unlike Sora, was trained using only 32 GPUs, which significantly reduces costs and speeds up the process. This model is capable of producing highly realistic and coherent videos, as demonstrated by their previews. The company is also working on an app called Diffuse, available on iOS, which allows users to create animated dancing videos from a single selfie. With a focus on social media creation and a commitment to democratizing media creation for everyone, Hicks Field is poised to make AI video generation more accessible and affordable. Their technology has the potential to revolutionize content creation for social media, offering unparalleled personalization and control over the generated content.
Takeaways
- ๐ Hicks Field, a new startup, has created highly realistic AI videos with significantly less computational resources than Sora, making it potentially faster and cheaper.
- ๐ค Hicks Field aims to democratize social media creation by providing a pre-trained model that can be fine-tuned for specific tasks like video generation.
- ๐ฑ They are developing an app called Diffuse, available on iOS, which allows users to create animated dancing videos from a single selfie.
- ๐จ The AI video generation model from Hicks Field is designed to produce lifelike human faces and movements, with a focus on realism and social media content creation.
- ๐ Despite the high quality, the model was trained on only 32 GPUs, a fraction of what Sora from OpenAI used, which could mean lower costs and faster rendering times for users.
- ๐ฐ The cost of GPUs, like Nvidia's which can cost up to $40,000 each, is a significant factor in the expense of training AI models, which can run into the hundreds of millions of dollars.
- ๐ Hicks Field's approach is targeting social media and events, with plans for a wide range of products beyond the Diffuse app.
- ๐ The company is focusing on personalization and control in their video model, allowing users to make specific changes to characters and scenes.
- ๐ The source of Hicks Field's training data is not revealed but is said to come from multiple publicly available sources, which could raise copyright concerns.
- ๐ The video generation model is currently available by invitation only, with a waitlist on their website for early access.
- ๐ OpenAI's Sora has released an official music video demonstrating the artistic potential of AI video generation, showcasing a consistent style and immersive experience.
Q & A
What is the name of the startup that generated the AI videos discussed in the transcript?
-The startup's name is Hicks Field.
What is remarkable about the AI video generator from Hicks Field in comparison to Sora?
-The AI video generator from Hicks Field was trained using 100 times fewer GPUs than Sora, which suggests it will be significantly cheaper, faster, and more accessible.
What are the two products that Hicks Field is currently working on?
-Hicks Field is working on an app called Diffuse, which is available on iOS, and a foundational AI video generation model.
How does the AI video generation model from Hicks Field render human faces?
-The model renders human faces with realistic and natural proportions, making it difficult to distinguish from real photos or videos.
What is one of the challenges AI often faces when generating videos?
-AI often struggles with rendering teeth in a coherent and believable way.
How many people were on the team that developed the generative models for Hicks Field's platform?
-The generative models were developed by a team of 16 people.
What is the estimated GPU usage for training Sora from Open AI compared to Hicks Field?
-Sora is estimated to have used between 4,200 to 10,500 GPUs for training, which is 130 to 328 times more than what Hicks Field used.
Why is the cost of Nvidia GPUs significant in the context of AI video generation?
-The cost of Nvidia GPUs is significant because they are expensive, with a single one costing around $40,000, and are required in large quantities for training AI models like Sora.
What is the name of the mobile app offering from Hicks Field?
-The mobile app offering from Hicks Field is called Diffuse.
What is the primary focus of Hicks Field in terms of their video model's capabilities?
-Hicks Field is focusing on personalization and control, as well as generating realistic looking humans and environments.
What is the current availability of Hicks Field's video generation model?
-The video generation model is currently available by invitation only, with interested users able to join a waitlist on their website.
How does the AI video generation technology represent a new artistic medium?
-The technology allows for the creation of immersive, dreamlike, and psychedelic experiences, offering artists new ways to express themselves and turn their ideas into reality.
Outlines
๐ Introduction to Hicksfield's AI Video Technology
The video introduces Hicksfield, a new startup that has developed highly realistic AI videos using significantly fewer GPUs than competitors like Sora. The company's approach to democratizing social media creation is highlighted, emphasizing their foundational model's ability to be fine-tuned for specific tasks. The presenter, Samson, expresses excitement about the potential of AI video generation and invites viewers to explore the technology's capabilities, promising a deep dive into a 2-minute AI music video created by Sora.
๐ Hicksfield's Efficiency and Impact on AI Video Generation
The video compares Hicksfield's AI video generator to Sora's, noting that Sora requires a substantial amount of computational power, estimated to be between 130 to 328 times more GPUs. Despite this, Hicksfield's smaller team has achieved impressive results with a fraction of the resources, which is exciting because it suggests that their technology will be more affordable and faster. The video also discusses the high costs associated with GPUs, the potential for public access to AI video generation, and showcases more examples of Hicksfield's output, including a makeup ad and a detailed critique of the rendering quality.
๐ฑ Hicksfield's Mobile App and Broader AI Video Ambitions
The video discusses Hicksfield's mobile app offering called Diffuse, which allows users to create short animated dancing videos from a single selfie. The app's preview is shown, highlighting the lifelike motion and stylized aesthetic. The presenter also talks about Hicksfield's broader plans for AI video generation, including personalization and control over video content, focusing on realism. The company's target audience is social media users, and they are currently rolling out their app in select regions with plans for global availability. The video generation model is available by invitation only, with a waitlist on their website.
๐จ Sora's Artistic AI Video and the Future of Creative Expression
The final paragraph showcases an official music video made with Sora, emphasizing the beautiful and consistent style, color palette, and narrative coherence. The video demonstrates Sora's ability to generate a range of different scenes and adjust camera movements effectively. The presenter reflects on the new artistic medium that AI video generation represents and the opportunities it provides for creative expression. The video concludes with an invitation for viewers to subscribe to the channel for more insights on AI technologies and a wish for the viewers to have a delightful day.
Mindmap
Keywords
AI Video Generation
GPUs
Hicks Field
Diffuse App
Foundational Model
Realism
Social Media Creation
Rendering
Sora
Nvidia GPUs
Artistic Medium
Highlights
A new startup called Hicks Field is creating realistic and beautiful AI videos with significantly less computational resources than Sora.
Hicks Field's AI video generator was trained using only 32 GPUs, compared to the 10,000 GPUs Sora reportedly used.
The reduced GPU usage implies faster and cheaper AI video generation, making it more accessible.
Hicks Field aims to democratize social media creation with their foundational model, which can be fine-tuned for specific tasks.
The company is developing two products: an app called Diffuse and a foundational AI video generation model.
The AI generated videos from Hicks Field are highly coherent, with realistic human features and natural movement.
The platform has evolved rapidly, with significant improvements in image quality and movement rendering in just a few months.
Hicks Field's generative models were developed by a small team of 16 people in under 9 months.
The cost efficiency of Hicks Field's model is a key advantage over Sora, which requires substantial computational resources and funding.
The AI video generation model can produce clips of around 7 seconds, which is longer than most current video generators.
Hicks Field's training data is sourced from multiple publicly available places, though the specifics are not disclosed.
The mobile app Diffuse allows users to create short animated dancing videos from a single selfie image.
The app is initially available in select regions and will gradually roll out globally.
Hicks Field is focusing on personalization, control, and realism in their approach to video generation.
The company is targeting social media and aims to create a wide range of products for different use cases.
Hicks Field is led by the former Snap AI Chief, indicating a strong background in competitive social media companies.
The potential of AI video generation as a new artistic medium is highlighted through the creation of a music video by August Camp with Sora.
Sora's ability to generate different scenes with high resolution and understand complex visual effects like parallax is showcased.
The advancements in AI video generation are seen as an opportunity for creative expression in new ways.