Lovable Ships New Creative Engine With Multi-Model Routing and Design Previews.

Lovable has launched a new creative engine designed to give users more control over AI-generated applications while avoiding template-based shortcuts. The first pieces of the system shipped today, according to a LinkedIn post from Lovable's co-founder.

The update represents a shift in how the platform approaches AI-driven software creation. Rather than relying on predefined templates disguised as creative output, Lovable's new engine routes different steps of the generation process to whichever frontier model performs best for that specific task. The company describes this as an "agentic chain" that composes multiple models without requiring users to manually select or switch between them.

The engine is built to support full-stack applications from the start, not just static artifacts like landing pages or slide decks. Lovable's team argued that creative projects often evolve: what begins as a static site may later need backend logic, user authentication, or agentic behavior. The new system aims to avoid technical constraints that would require rebuilding when those moments arrive.

This aligns with Lovable's recent feature launches, including native SEO capabilities and integrations with services like Stripe for payments. The platform has been positioning itself as a tool for building production-ready web applications, not just prototypes.

One of the first visible features is the ability to preview multiple design directions before committing to a build. Instead of generating a single output and iterating from there, users can now review variations upfront and choose which path to pursue. This is meant to preserve creative agency while letting the AI handle execution.

The post emphasized that "creative agency doesn't mean AI on autopilot." Lovable's bet is that the best results still require a clear point of view from the person building, with the engine amplifying that vision rather than replacing it.

The engine's routing logic picks the best available model for each step, whether that's generating code, designing interfaces, or producing images. The LinkedIn post noted that even the images in the announcement were generated by the new system. By remaining model-agnostic, Lovable aims to stay current as new frontier models are released, rather than locking into a single provider.

This approach mirrors trends across the vibe coding space, where platforms like Cursor, v0, and Bolt are experimenting with multi-model workflows and agentic systems. The difference, according to Lovable, is that the routing happens invisibly to the user.

The company indicated that more components of the creative engine are in development, but it chose to release the first pieces now rather than wait for a complete rollout. The initial features are available to all Lovable users starting today.

Lovable has been iterating rapidly since launching its mobile app in April and adding Wiz security scanning earlier this month. The platform competes in a crowded market of AI development tools, but its focus on full-stack output and model composition sets it apart from code-assistant tools that augment traditional workflows.

The creative engine is now live at lovable.dev.