Picture a marketing team trying to cut a 30-second product spot. They're bouncing between RunwayML for generation, CapCut for editing, maybe a third platform for storyboards, a fourth for upscaling. Each one wants a subscription. Each has its own credit system, its own upload queue. By the time the final render finishes, they've logged into four different tools and waited through three separate processing gauntlets.
Josephine Lee and Serena Pei think that's absurd.
Their answer is Palette Technology, a San Francisco startup that emerged from Y Combinator's Summer 2026 batch with an audacious pitch: collapse the entire workflow into a single canvas. Route creative tasks across multiple state-of-the-art video generation models—Google's Veo 3.1, ByteDance's Seedance 2.0, Kuaishou's Kling 3.0 Pro—without ever leaving one unified interface.
No tool-hopping. No context loss. Just point, prompt, iterate.
Whether the bet pays off depends on a lot of moving parts, not least whether Palette's routing logic can actually pick the right model for the job—and whether users trust a two-person team to outmaneuver Adobe, Runway, and the rest of the creative AI establishment.
The Routing Play
Palette isn't building its own models. Instead, it's building the layer between you and the models—a kind of intelligent switchboard. The platform currently taps five major video generation engines: Veo 3.1, Seedance 2.0, Kling 3.0 Pro, MiniMax's Hailuo 02, and Lightricks' LTX Video. Users can manually toggle between them, or let Palette's "Best Fit" auto-routing decide which model suits the task.
The company claims this approach delivers 35% faster generation and cuts costs by half compared to juggling standalone subscriptions. Those figures come from Palette's own internal benchmarks—a caveat worth noting as Palette tries to win over skeptical enterprise buyers.
According to the company's site, the platform also handles failed job refunds automatically and returns credit differences when a cheaper fallback model gets the work done. Nice touches, if they hold up under real-world load.
Lee, the CEO, and Pei, the CTO, are both MIT computer science grads with backgrounds in multimodal AI research. They frame Palette as a response to what they call "tool and context fragmentation"—the productivity bleed that happens when creative direction gets carved up across disconnected platforms. Their pitch centers on a conversational interface that remembers everything: initial prompt, storyboard tweaks, generation attempts, iterative edits. One session, one thread.
The YC company page lists the team size as two, though that figure may already be stale. Most startups start hiring in earnest after launch.
Pricing: A Penny Per Credit
Palette runs on a credit system priced at one cent per credit. As a benchmark, the company says 1,000 credits can produce a full character reel—character sheet, storyboard, 15 seconds of character-consistent video. That's $10 for what might otherwise require separate subscriptions to Runway, Midjourney, and an upscaling service.
Enterprise customers get custom quotes and access to what Palette calls "end-to-end services": white-glove ad generation, brand-safe influencer content, interactive training modules. There's a live API at studio.palettelabs.com/api/v1, with endpoints for images, videos, music, upscaling, and job status checks.
The pricing model assumes creators value simplicity over specialist tools. Whether that trade-off works depends on how often teams actually need to switch models mid-project—and whether Palette's routing logic consistently picks the right one. If it gets that wrong too often, the value proposition crumbles.
Timing Is Everything
Palette's launch comes as the creative AI landscape undergoes something like rapid tectonic shift. Adobe announced a major expansion of its "creative agent" capabilities across Firefly and Creative Cloud in June, positioning an end-to-end agent-guided workflow. Runway introduced Agent 2.0 shortly after, folding autonomous task execution into its platform. Even Descript, traditionally a video editor, began integrating external generation models like Kling O1 around the same time.
The video generation models Palette routes to have themselves been iterating at breakneck speed. Google made Veo 3.1 generally available through its Gemini API in January, with expanded creative controls. ByteDance launched Seedance 2.0 in February and integrated it into CapCut by late March, though TechCrunch reported in mid-March that the company paused broader global launch plans. Kling 3.0 debuted in February and added native 4K support in May.
That velocity creates opportunity—and risk. Palette's entire value proposition hinges on staying current with model releases and maintaining API access across providers. That's a moving target when some platforms have uncertain distribution roadmaps, or when a provider decides to pull access or change terms.
It's a dependency chain that could snap.
The OpenAI Footnote
Palette displays both Y Combinator and OpenAI logos on its site. The OpenAI reference likely stems from the $2 million token credit offer Sam Altman extended to every company in that YC batch earlier in the year—a program-level perk that TechCrunch and other outlets covered. Whether Palette accepted the SAFE arrangement hasn't been publicly disclosed, and Palette hasn't published any standalone funding announcements beyond its YC participation.
The lack of a traditional seed round press release isn't unusual for a freshly launched YC company. But it does leave open questions about capitalization and runway as the team tries to scale, especially if they need to hire engineers, support staff, and salespeople to chase enterprise deals.
Who's This For, Really?
Palette targets three tiers: individual creators, startups, and enterprise teams. The YC Launch post singles out marketing leaders, learning and development heads, and operations managers as core buyers. Use cases span entertainment, product marketing, safety training, field service documentation, healthcare education.
Whether those verticals actually consolidate their creative workflows into a single platform remains to be seen. Enterprises often have entrenched vendor relationships and compliance requirements that favor specialized, audited tools over newer aggregators—even if the aggregator promises cost savings. Palette's roadmap toward SOC 2 Type II and ISO 27001 certifications suggests Lee and Pei recognize that gap, but those audits take time and money.
The company updated its security and privacy policies in early August, laying out its stance: Palette says it doesn't use customer prompts or outputs to train any models, its own or third-party, and that output ownership stays with the user. Standard language for a 2025 AI company, but table stakes for enterprise buyers.
What Comes Next
The product is live. The pricing is public. The demo video is up. What Palette doesn't yet have: customer traction data, third-party validation of its performance claims.
For now, the company is asking marketing teams and creators to try the platform and decide whether routing between five video models in one interface beats managing five separate subscriptions. The penny-per-credit entry point lowers the barrier to experimentation. But sustained adoption will hinge on whether Palette's auto-routing actually works in the wild—and whether the creative outputs justify trading specialist tools for a generalist layer.
In a market where Adobe, Runway, and Descript are all racing toward unified creative agents, Palette is making a bet that MIT-trained routing logic and a clean API can carve out space before the incumbents close the gap.
The next few months will reveal whether two-person teams can still outmaneuver platform giants in the generative AI era. Or whether the window's already shut.
