Lucy 2.5 Is Here: Decart's Real-Time AI Video Editing Hits 30 FPS 1080p

Decart launched Lucy 2.5 on July 16, 2026 — a major upgrade to its real-time AI video editing platform that pushes live generative video into production territory. The new model runs at 30 FPS in 1080p, a leap from the previous generation's lower resolution and frame rate, making it viable for live broadcast, real-time VFX, and interactive video workflows.

The release hit 2.5 million tweet views within its first 48 hours and reinforces Decart's position as the company to watch in real-time generative video — backed by $300 million in funding from Radical Ventures, Sequoia Capital, and NVIDIA.

What Lucy 2.5 Does

Unlike traditional AI video generators that take minutes to render a clip from a text prompt, Lucy 2.5 operates in real time. You edit video frames as they are generated — changing prompts mid-stream, applying VFX effects, and seeing the output update at 30 frames per second without a processing delay.

The model is designed for three core workflows:

Workflow	What it replaces	Lucy 2.5 approach
Live VFX	Offline compositing and post-production	Real-time style transfer, object insertion, and environment changes applied on the fly
Interactive video	Pre-rendered branching narratives	Dynamic video generation that responds to viewer input or changing prompts
Rapid prototyping	Iterative render-review cycles	See the effect of prompt changes immediately at 1080p 30 FPS

Self-Anchoring: The Key Technical Innovation

The headline feature in Lucy 2.5 is Self-Anchoring, a technique that maintains visual consistency across rapidly changing prompts and edits. In earlier real-time video models, changing the prompt mid-generation often caused the subject to drift, warp, or change identity. Self-Anchoring locks key visual features (subject identity, composition structure, color palette) so that edits apply on top of a stable base rather than destabilizing the entire frame.

Rule of thumb: Self-Anchoring is most useful when you need to iterate on a live scene — swap a background, change lighting, or add an object — without the subject shifting between frames. For full scene transitions where you want a complete visual reset, you can disable anchoring to allow the model to generate freely.

Coarse-to-Fine Prompts

Lucy 2.5 introduces a two-stage prompting system called coarse-to-fine. Instead of writing a single prompt and hoping the model interprets it correctly, you write:

A coarse prompt — describes the broad scene, composition, and mood
A fine prompt — adds specific details, VFX effects, or real-time adjustments

The model uses the coarse prompt as a stable scaffold and applies the fine prompt as a conditional overlay. This separation makes real-time editing more predictable because changes to the fine prompt do not destabilize the base scene.

Pricing

Tier	Price	What you get
Realtime generation	$0.02 per second	Live 30 FPS 1080p editing and streaming
Video output	$0.04 per second	Rendered video export at full resolution

At $0.02 per second of realtime generation, a 60-second live editing session costs $1.20. Rendered exports are $2.40 for the same duration. This is competitive with GPU rental costs for equivalent real-time inference workloads and significantly cheaper than traditional VFX pipelines for live production.

$300M in Funding and What It Means

Decart has raised $300 million to date from Radical Ventures, Sequoia Capital, and NVIDIA. The NVIDIA investment is particularly telling — it suggests deep infrastructure collaboration, likely giving Decart early access to next-generation hardware optimized for real-time diffusion inference.

The funding positions Decart as a serious contender in the generative video space alongside OpenAI (Sora), Google (Veo), and xAI (Grok Imagine Video). While those models focus on offline generation quality, Decart is the only company betting entirely on real-time interactivity as the primary use case.

Availability

Lucy 2.5 is available now through Decart's API and web platform. API access provides programmatic control over real-time video generation, including prompt streaming, Self-Anchoring toggles, and coarse-to-fine prompt mode selection.

The web interface at decart.ai/lucy provides a direct visual editor for live experimentation — no code required.

The Bottom Line

Lucy 2.5 moves real-time AI video editing from "interesting demo" to "production tool." The combination of 30 FPS 1080p output, Self-Anchoring for stable mid-stream edits, and coarse-to-fine prompt control addresses the three biggest friction points that kept real-time generative video out of professional workflows: resolution, consistency, and control.

At $0.02 per second for live generation, the pricing is accessible enough for individual creators while scaling to production budgets for live events and broadcast.

For the latest updates and hands-on testing, visit Decart's Lucy page — the model is live and available today.