Palette

Quick Take

Palette is an AI creative studio that generates images, video, and music from a single workspace, built by a team with MIT and multimodal-AI research backgrounds and backed by Y Combinator and OpenAI. What sets it apart from a typical single-model generator is routing: Palette integrates multiple third-party models, including Seedance 2.0, Google’s Veo 3.1, and Kling 3.0 Pro, and automatically picks between them based on what a given request actually needs.

Why Model Routing Matters for Creative Work

Picking the right generative model for a specific shot is its own skill. Some models render text and fine detail better. Some are faster but lower fidelity. Some are meaningfully cheaper for similar quality on simpler shots. Manually testing and choosing between providers for every single generation is slow and easy to get wrong, especially across a project with dozens or hundreds of shots.

Palette’s pitch is handling that decision automatically: speed versus quality, routing simple or draft shots to faster, cheaper models and reserving stronger models for shots that need them; cost control, since failed generations refund credits and a cheaper fallback model returns the price difference; and one canvas, keeping character references, storyboard structure, and generated assets together in a single project instead of juggling separate tools per model.

What Palette Is Best For

• Multi-shot projects that need visual consistency: define a character or style once, reuse it across many shots and formats
• Teams that don’t want to manually pick a model per shot: let routing handle capability and cost tradeoffs
• Storyboarding before generating: setting audience, format, pacing, and intent up front for repeatable direction at scale
• Mixed media projects: producing image, video, and music from the same creative brief in one place
• Ad and marketing creative, UGC-style content, and structured content pipelines: Palette offers API access and enterprise workflows for these use cases

Key Features

• Model routing across providers (Seedance, Veo, Kling, and others), chosen automatically per request
• Character consistency across multiple shots and formats from a single defined character
• Storyboarding tools to set pacing, format, and creative intent before generation
• Natural-language editing, revising scenes by describing the change rather than manual re-editing
• Automatic credit refunds on failed generations, and partial refunds when routed to a cheaper fallback model
• Synchronous image generation and asynchronous video job queuing via API

Honest Limitations

• You don’t fully control which model handles a given shot. The routing is largely automatic, which is convenient but means less manual control than picking a model directly for creators who want that precision.
• Underlying model quality still varies. Routing to a cheaper model for cost savings can mean a visible quality tradeoff on specific shots.
• Newer, smaller company. As a younger, VC-backed platform, feature set and pricing are more likely to change than with longer-established tools.
• Commercial rights depend on the underlying model used. Since Palette routes across multiple third-party models, confirm commercial-use terms for your specific output rather than assuming one blanket policy.

Alternatives Worth Knowing

• Seedance, one of the video models Palette routes to directly, also usable on its own via API or through Dreamina
• Veo, Google’s video model, also available directly
• Kling AI, another model Palette routes to, also usable standalone
• Runway, a single-platform alternative with deep manual editing control
• Higgsfield AI, focused on character-consistent stylized video

For the broader idea of routing work to the right model instead of one default choice, see Model Routing and How to Make Cheaper AI Models Work Smarter.