Creative Workflow Roundup: Pika’s AI Video Editing Upgrades, xAI’s Entertainment Play, Veo 2's Expansion, Kaiber’s ModelMaker, and AI’s Next Creative Shifts

In This Week’s Roundup: Pika adds an intuitive AI replacement feature for AI video editing and xAI’s Grok 3 launch raises new questions about AI’s future in entertainment production. Google’s Veo 2 is now available on Freepik and Fal, making high-end AI video more accessible. Kaiber’s new ModelMaker enables custom AI model training (characters, styles, objects, places) integrated with Superstudio Pro.

Other big updates: Mosaic launches AI-powered video editing automation, MidJourney adds folders and works on a faster image model, Alibaba unveils WanX 2.1 as an open-source video generator, and new research accelerates HunyuanVideo’s AI video model rendering speeds by up to 10x. Meanwhile, Voices App removes limits, Suno improves AI music editing features, and OminiControl enables object placement in scenes without training.

Pika Brings Intuitive Replacement Feature to AI Video

The News: Pika, an AI video generation platform, has introduced a new feature called Pikaswaps, allowing users to replace objects (or really anything) in videos using uploaded photos or text descriptions. This builds on their earlier Pikadditions feature, which let users add anything into a scene. Pika has also launched an iPhone app, making its AI video tools more accessible on mobile.

Lab Notes: Pika keeps proving itself as a serious competitor, not just with its video model, which is solid, but with how they roll out their tools. When Pika first launched, its model wasn’t quite at the level of the best in the industry, and I honestly didn’t expect it to last. But they’ve more than made up for that with updates to the core model and smart, creative features that make AI video intuitive and fun to use.

The way they integrate AI into storytelling stands out. Pikaswaps makes AI replacement feel effortless, and it fits right in with their other tools, like Pikadditions and Pikaffects, which offer easy-to-use scene modifications. They’ve also had features for maintaining consistent characters and other storytelling elements for a while now. It’s one of the most enjoyable AI platforms to experiment with, but more importantly, it’s actually useful for producers looking to enhance their videos and tell cohesive stories.

xAI, Grok 3, and the Future of AI Entertainment Production

The News (And a Note): Avi Schiffmann, founder of AI hardware/software startup Friend, recently posted an interesting take on X about the future of AI, specifically where foundational model companies might be headed next:

"The real consumer business model for foundational model companies is entertainment production."

This was posted during last Sunday’s livestream release of Grok 3, the latest AI model from Elon Musk’s xAI. AI models are expensive to run, and these companies need sustainable revenue streams. Foundational AI companies may not just power entertainment production but eventually compete in entertainment themselves.

The Grok 3 announcement reinforced this idea by revealing that xAI is launching a game studio. This could be an early sign of a shift toward AI-generated media that moves beyond traditional formats. Rather than just AI-generated films or TV shows, we may see more interactive, real-time, and experiential content emerge. Midjourney, for example, has been internally working on real-time storytelling worlds, signaling how generative media could evolve into something far more dynamic than static videos.

Lab Notes: As a creative producer, it’s time to think seriously about where this is heading. If foundational model companies turn to content production as a major revenue stream, it means that, at some level, we’ll be competing with them. That might not be an easy thought to accept, but it’s something that’s becoming clearer over time.

That doesn’t mean all AI-generated content will be generic, low-effort material. There will still be plenty of human-driven creativity, and AI-assisted storytelling will likely define the future. Some artists will lean into AI, while others will push for purely human-made work. But the key takeaway is that producers and creatives need to start developing a strong, unique voice, something recognizable that stands out in an era where AI-generated content will only become more widespread.

It’s becoming clear that the AI shift isn’t just about making media easier to produce; it’s about entirely new forms of entertainment emerging. Foundational AI models will find ways to stay financially viable, and entertainment could be one of their most sustainable paths. For creatives, I think that means it’s important to stay ahead, "be different," and carve out a distinct identity while preparing for a rapidly evolving industry.

Veo 2 Expands to Freepik and Fal.ai, Bringing Wider AI Video Access

The News: Google’s Veo 2, a state-of-the-art AI video generation model, has officially launched on Freepik following its recent debut in the YouTube app for Shorts creation. The model is also now available on Fal.ai, a platform that provides API access for developers looking to integrate AI video generation into their own applications. Fal.ai also offers a playground interface, allowing users to experiment with various models through pay-as-you-go access. Veo 2 is designed for high-quality, realistic outputs and smooth motion. It’s one of the most powerful AI video models currently available.

Lab Notes: Last week, Veo 2’s rollout in the YouTube app was a big moment, opening up AI-powered video creation to a much wider audience. But the big question was when Google would expand access beyond the constraints of YouTube’s mobile editing interface. Now, with Veo 2 available on Freepik and Fal.ai, we’re starting to see that next step. Veo 2 access is suddenly expanding fast, and I wouldn’t be surprised to see it pop up in more services soon. Fal.ai’s API access means other platforms can start building on top of it, and a broader web rollout from Google itself seems inevitable.

Having tested Veo 2 on Fal.ai, I can say it’s still one of the most capable models out there, but cost quickly becomes a factor. At $2.50 for five seconds a currently discounted rate of $1.25 for five seconds on Fal, it makes experimentation expensive, especially compared to other models. For example, Kling 1.6 is a high-quality video model and only costs $0.15 for five seconds. There are plenty of other strong models competing in this space. Luma’s Ray 2 is very capable, and companies like Tencent and Alibaba have impressive alternatives. Veo 2 still has a lead in overall output quality, but AI video generation is evolving as a whole, with all these models advancing in parallel. Fal.ai has said they’re working on lowering prices, so we’ll see how that affects adoption moving forward.

Kaiber's ModelMaker Brings Custom AI Training to Superstudios’s Pro Tier

The News: Kaiber has introduced ModelMaker, a new feature coming to Superstudio Pro, its high-end $150 per month tier. ModelMaker allows users to train their own AI models by uploading high-resolution images and selecting a model type: Character, Object, Style, or Place. Once trained, these models can generate outputs based on the user’s vision, offering more control and personalization.

Lab Notes: Kaiber Superstudio is a really interesting platform. Competing in this space is tough, as I’ve mentioned with Pika, there are so many strong AI tools out there, and standing out requires real innovation. Superstudio’s ModelMaker seems like a powerful step in that direction, unlocking a lot of potential use cases.

The biggest draw here is the flexibility it could bring to storytelling. The ability to train a model around specific characters, styles, or objects could be a big unlock for creative workflows. I still need to test it more to see how well it performs, but on paper, this is a strong move from Kaiber.

Additional Findings: Quick Updates on Important News and Workflows

Mosaic Canvas Launches – A new AI-powered video editing tool enables autonomous editing agents that can generate multiple variations from a single video.