Infrastructure, talent, models: 3 lessons for gen AI startups from Higgsfield.ai

Adults in the U.S. spend about three hours a day watching some form of video online, confirming that mobile video is an important channel for companies to get their unique products, services, and ideas in front of tuned-in audiences.

Like any good movie, though, there’s a twist: Video is expensive. The time and investment can be quite substantial, with the back and forth of ideation, shooting, editing, and costs (sometimes thousands of dollars per minute). That’s a lot to stake on any single asset for even the biggest corporations with large marketing budgets. For small or mid-sized companies, it’s often untenable.

We knew there had to be a way to democratize video creation, so companies of any size could deliver compelling content. That’s why we launched Higgsfield.ai a year ago.

We’re a startup made for marketers. We offer a lightweight, easy-to-use video creation tool that uses gen AI to create video content for a fraction of the price and half the time. And since people are often the star of any story, we also prioritize the ability to generate videos with realistic people and movements.

Want to see for yourself? Just today, we launched Diffuse 2.0, a multimodal AI platform that combines your photos, video, and text with your personality and creativity to create fun and meaningful stories — all starring you. It’s like combining the self-discovery features of TikTok with the storytelling content in Choices, all powered by our foundational AI video model.

I’ve been on a long journey with AI, going back to my time as a co-founder of another creativity-focused startup, AI Factory, which developed filters and tools for — and was later acquired by — Snap for use in Snapchat. I then worked as head of AI at the social media company before leaving to start Higgsfield.ai.

In that time, I learned some important lessons about what makes working with AI a unique challenge for startups. AI has tremendous potential, and it’s evolving quickly. Companies need to invest their resources intelligently if they want to continue pushing the limits of their models and stay ahead of competitors.

There are three main choices that your business can make to address the challenges and set your startup on the path to success.

1. Build on fast, powerful infrastructure

When I worked for Snapchat, I saw countless brands join the platform to reach our growing audiences. But I also saw how few of those brands considered making unique content that was optimized for the mobile experience native to Snapchat.

As channels and outlets continue to grow in the digital age, having content optimized for where it will exist is essential. Our goal at Higgsfield is to help users open up new possibilities by getting creative for a fraction of the time and resources. That means that we need to train models that can handle a wide range of visuals — anything our users can imagine. And our platform needs to generate video at a speed that encourages people to experiment with personalized variations. Any lag, and they’re clicking away or switching apps.

The key to both training and generating content with AI is infrastructure. So far, Google Cloud has offered the best foundation for our generative-AI business. With its decades of experience in AI and purpose-built platforms like Vertex AI and the AI Hypercomputer, it’s provided us with the ideal tools for training inferences and processing petabytes of data quickly. We’ve been able to constantly iterate on our models, develop our orchestration layer, and set up experiments with ease.

The result is a video platform that is always reaching for the next level of style and performance to help our users create highly personalized videos that speak to different audiences. Instead of just creating one scene with a young woman drinking coffee outside of a café, why not also show an older man, or even an astronaut drinking their coffee at night, in a bar, or even on a space station? With the right infrastructure, we can give our users the power to try anything quickly and creatively.

2. Look for support from talented partners

The AI market is changing fast, and so is Higgsfield.ai. We have a fantastic team that works hard to make our video models more realistic every day — more life-like faces and expressions, dynamic camera movements and lighting, and visuals that can meet any type of prompt.

When you partner with a cloud provider you trust, you can give your team the support they need to scale and focus on development that improves your AI rapidly. We all know that debugging, fixing errors, and catching exceptions represents about 90% of engineering pain. With a highly reliable cloud provider like Google Cloud, we have more time to focus on our core business. Now our models improve constantly, getting us closer to delivering any video that marketers can imagine.

Another essential ingredient is the level of support that you get from your cloud provider. Especially as a startup, we rely on training and easy-to-use toolsets to configure and maintain our infrastructure. As part of the Google for Startups program, we received credits for additional training and toolsets that helped us get more from the cloud and scale quickly. Leaning on expert advice and battle-tested tooling, we can build with confidence.

3. Take from the best of available models

Across the gen AI space, new models are improving every day. Startups have limited resources in time, money, and personnel.

Spend your resources wisely and don’t feel as if you need to build everything from scratch. Utilize the tools available to you, built by some of the smartest minds in the industry. Identify the elements of your business that will be best developed in-house, as well as those that can be covered by existing gen-AI tools.

For us, that means focusing on video that shows people being people: cheering, dancing, playing music, laughing, crying. For other aspects of our videos, we take advantage of existing gen AI tools where we can.

Take Gemini, for instance. It pushes the boundaries of video understanding, helping us gain more control over video generation. It’s also great for things like video captioning, delivering highly accurate captions that deliver 50% better matching than competitors. That helps marketers make social media videos more accessible to wider audiences.

Gen AI is evolving every day, and we’re excited to see how it can level the playing field for social video. For us to do that for our customers and partners in marketing, we need an AI partner like Google Cloud that can level the playing field for AI startups.

Launching an AI business requires a strong commitment to constant research, experimentation, and improvement. You need to always strive for the next level of performance. Between my time at AI Factory, Snap, and Higgsfield.ai, I’ve seen first-hand how important the right infrastructure, tools, and partners can be to startup success. You want to focus on your unique business, not solving problems others have already mastered.

Going it alone means more work and more risk. Working together, you can go far.