Decart Launches Lucy 2.5: Real-Time AI Video Editing at 30 FPS 1080p
Decart's Lucy 2.5 brings real-time 1080p video editing at 30 FPS with Self-Anchoring temporal consistency and physically-aware VFX. Pricing starts at $0.02/sec. The $300M-funded model ships July 16, 2026.
You scrub to the 14-second mark of a 4K clip, paint a fireball into the frame with a brush stroke, and the flame propagates across the next three seconds with believable heat shimmer and physics-driven smoke — without waiting for a render queue. That is what Decart's Lucy 2.5 does, and it does it today.
On July 16, 2026, Decart shipped Lucy 2.5, a real-time AI video editing model that brings interactive, frame-accurate editing to a browser at 30 FPS in 1080p. Unlike every video generation model before it — which requires you to type a prompt, wait 30 to 90 seconds, and then decide if you like the result — Lucy 2.5 treats video as a live canvas. You paint, composite, and animate edits that render instantly as you work.
What Lucy 2.5 Actually Changes
The headline capability is real-time 1080p video editing at 30 frames per second, but the architecture underneath is what makes it possible. Decart built three technical pillars into this release:
Self-Anchoring temporal consistency. Previous video models lose coherence after a few frames — subjects warp, backgrounds flicker, lighting jumps. Lucy 2.5 anchors each edit to a learned spatiotemporal signature that persists across the entire clip, so a mask painted on frame 1 stays locked to the same object on frame 90. The result is consistent video that does not degrade as you extend or modify it.
Physically-aware VFX compositing. Add fire, water, smoke, debris, or lighting changes, and the model calculates how those elements interact with the existing scene — not as a static overlay but as a physical simulation. Drop a virtual light source into a dark room, and Lucy 2.5 recalculates shadows, reflections, and ambient occlusion across affected frames in real time.
Coarse-to-fine prompt layering. You can start with a rough directional prompt ("make this scene stormy"), refine with a targeted edit ("add lightning strike on the left tower at 0:12"), and then micro-adjust with a paint-like mask ("reduce rain density over the foreground character"). Each layer of the prompt stack respects the ones beneath it, so refinements compound without breaking earlier choices.
Pricing and Access
Decart priced Lucy 2.5 to compete with per-second compute costs rather than per-generation credits:
| Mode | Price | What You Get |
|---|---|---|
| Realtime editing | $0.02 per second | Interactive canvas, live preview, brush-and-composite workflow |
| Video output | $0.04 per second | Final rendered 1080p export with full temporal processing |
At those rates, a 30-second edit session costs roughly $0.60, and a finished 15-second export runs $0.60 — significantly below the effective per-second cost of competing models that charge per generation and discard failed attempts.
The model is available today through Decart's web platform at lucy.decart.ai. An API is expected to follow, though Decart has not announced a timeline.
The $300M Bet
Lucy 2.5 arrives with considerable weight behind it. Decart has raised $300M to date, with its most recent round led by a16z at a valuation that sources place well north of $1B. The funding reflects a broader thesis among investors that the next inflection point in generative AI is interactivity — moving from "type and wait" to "point, click, and edit" — and that the teams that crack real-time rendering will own the professional video editing market as it shifts from deterministic tools (Premiere, DaVinci Resolve) to AI-native workflows.
The competitive landscape is already forming. Runway's Gen-4 series supports real-time preview at lower resolutions, and Pika's 2.5 model offers real-time stylization but not physics-aware VFX. Lucy 2.5's differentiation is the combination of temporal stability, physical simulation, and the coarse-to-fine prompt stack — features that target professional compositors and VFX artists rather than casual users.
What This Means for AI Video Editing
Lucy 2.5 pushes the category from "generation" into "editing" — a distinction that matters for the market. Generation tools produce a clip from scratch; editing tools let you modify an existing clip interactively. The former is useful for prototyping and social content. The latter is what post-production workflows actually look like.
If Lucy 2.5 delivers on its stability claims in production use, it effectively eliminates the iteration tax that currently makes AI video editing a multi-hour process of generate → review → discard → regenerate. A VFX artist who previously needed 12 generations to land a single composite shot can now paint it in seconds and export the final frame on the first pass.
Decart's next move — API availability, higher resolutions, and multi-track support — will determine whether Lucy 2.5 becomes a professional tool or remains a compelling demo. For now, the real-time editing category has a lead horse, and it runs at 30 FPS.
More Resources
Grok Imagine 1.5 Video Prompt Guide: Write Better Prompts (2026)
Learn how to write better Grok Imagine 1.5 video prompts with practical formulas, reusable templates, and example prompts for motion, camera moves, lighting, and scene consistency.

How to Use the Grok Imagine 1.5 API: A Complete Guide to Endpoints, Pricing & Code Examples
Complete guide to the Grok Imagine 1.5 API. Learn the endpoint, authentication, async workflow, pricing per second, and get working code examples in Python, cURL, and Node.js for video generation.

How to Use Grok Imagine 1.5 Video Generator: A Complete Workflow Guide for Better AI Clips
Learn how to use Grok Imagine 1.5 video generator like a pro. From prompt structure and source asset prep to video extension and longform assembly — a complete workflow guide.
Grok Imagine 1.5 AI Updates
Join the Grok Imagine 1.5 AI community
Get updates about AI video generation features, prompts, and pricing.
