Decart Launches Lucy 2.5: Real-Time AI Video Editing at 1080p 30 FPS

Video editing has always been a waiting game — render, preview, tweak, render again. Decart's Lucy 2.5, released July 16, 2026, removes that loop entirely.

The model generates and edits video in real time at 30 FPS in 1080p. You type a prompt or adjust a scene, and the result appears frame by frame with no rendering delay. For editors, that shifts the workflow from batch processing to live steering.

What Lucy 2.5 Does Differently

Self-Anchoring Temporal Consistency

The biggest problem in AI video editing is flicker — objects shift, textures warp, lighting changes between frames. Lucy 2.5 introduces Self-Anchoring, a mechanism that locks key visual elements across the timeline. Once a character, object, or background region is established, the model preserves its appearance across subsequent frames without requiring per-frame re-prompting.

This means you can edit frame 50 and frame 500 of the same scene without the character's face or clothing drifting between them.

Physically-Aware VFX

Lucy 2.5 understands basic physics in the context of a scene. When you add an effect — fire, water, smoke, gravity, cloth motion — the model applies it with awareness of the existing environment. A flame added to a windy outdoor shot behaves differently from one added to an indoor still. This removes the manual rotoscoping and keyframe adjustment that typically follows AI-generated effects.

Coarse-to-Fine Prompts

Instead of requiring a detailed prompt for every edit, Lucy 2.5 accepts a rough directional instruction and refines it. You can start with "make this scene darker and add rain," see the output, then layer in more specific instructions like "increase rain density and shift lighting to cold blue." The model iterates on its own output rather than re-generating from scratch each time.

Pricing and Access

Lucy 2.5 is available now through Decart's API. Pricing is usage-based:

• Realtime editing (interactive, frame-by-frame): $0.02 per second
• Full video generation (rendered output): $0.04 per second

At $0.02/sec, one minute of realtime editing costs roughly $1.20. Full renders run $2.40 per minute of output.

The Company Behind It

Decart has raised approximately $300 million to date. Lucy 2.5 represents the company's push from experimental AI video toward production-grade editing tools. The model's focus on temporal consistency and real-time interaction targets professional post-production workflows rather than casual novelty generation.

What This Means for the Industry

Real-time AI video editing at 1080p 30 FPS changes the cost structure of post-production.

Tasks that previously required multiple software passes — color grading, object removal, background replacement, physics simulation — can now be handled in a single live session. For independent creators, the per-second pricing removes the upfront software licensing cost. For studios, it compresses iteration time from hours to minutes.

The caveat is consistency over longer sequences. Self-Anchoring reduces drift substantially but does not eliminate it across very long clips. Editors working on multi-minute segments should still expect to review outputs in chunks.

What Comes Next

Decart has not announced a specific timeline for the next major release, but the improvements in Lucy 2.5 point toward three likely directions: higher resolutions (4K support), longer temporal windows (beyond Self-Anchoring's current range), and multi-modal input (video reference clips alongside text prompts).

For now, Lucy 2.5 is the closest thing to a real-time video editing assistant that works at production resolution.

FAQ

Is Lucy 2.5 available as a web app or only through API?

Currently available through Decart's API. There is no standalone desktop or web application at launch.

What hardware does Lucy 2.5 require?

Decart handles inference server-side. Users only need a stable internet connection and a compatible API client.

Can Lucy 2.5 edit existing video footage?

Yes. You can upload existing footage and apply edits, effects, or style changes through the real-time interface.

How does Self-Anchoring compare to traditional keyframing?

Self-Anchoring is automatic — it locks visual elements without manual keyframe placement. Traditional keyframing gives artists per-frame control but requires significant manual effort. The two approaches can be complementary for complex projects.