Home2.5FastShowcasePricing

LTX 2.5 AI Video Generator

LTX 2 now runs LTX 2.5 in this workspace — text-to-video and image-to-video with native multishot character lock. Prompt a scene or drop a still, then render a clip you can download.

Built for Speed

Three vertical samples on deliberately hard motion: glass shatter, dense rain crowds, layered hair and silk.

Crystal shatter

Hard specular highlights and fast debris in one 720p pass.

Neon rain crowd

Dense crowd motion with layered reflections and spray.

Silk and hair

Overlapping soft-body movement that usually smears.

Native Multishot Consistency

Three hard cuts of one person — wide, medium, close — is where most generators recast the lead. Need to generate right now? LTX 2.3 Fast is still live.

Barista · three cuts

Same face, apron, gold hoops and wrist tattoo across wide → medium → close.

Courier · three locations

Identity holds across crosswalk, elevator and rooftop.

Painter · atelier sequence

One painter through wide study, profile stroke and face return.

What's New in LTX 2.5

What actually changed from 2.3, in three lines.

LTX 2.3LTX 2.5
BackboneEarlier LTX stack22B asymmetric dual-stream transformer
Multishot identityStitch separate clips and hope they matchCharacter lock inside a single generation
Published benchmark—~6.8s for a 10s 720p image-to-video job on dual GB200

How LTX 2.5 Works

Think of the 22B dual-stream transformer as two readers: one locked to your prompt, one locked to how light, cloth and faces move. That split buys speed without physics turning to sludge.

  • Dual-stream backbone — Prompt intent and visual motion stay aligned through the render.

  • Distilled paths — Fewer steps when turnaround matters, full fidelity when the frame needs it.

  • Open-weight family — The community self-hosts the weights; here it runs in the browser.

Back to generator
LTX 2.5 multishot sample frame — painter in an atelier

How to Use LTX 2.5

Three steps, no node graph.

01

Prompt or upload

Brief the scene the way you would brief an editor, or drop a still for image-to-video.

02

Set options

Pick Fast or Pro, then aspect ratio, resolution, duration and fps in the panel.

03

Generate and download

Preview the clip, re-roll only the cut that missed, then export what you need.

Writing a Multishot Prompt

Three habits that decide whether the character survives the cuts.

  1. 01

    Name the character once

    Fix hair, wardrobe and one signature prop in the first sentence. LTX 2.5 anchors identity on those nouns across every cut.

  2. 02

    Spell the cuts in order

    Write wide, then medium, then close as separate beats. The model reads that order as the edit, not as a suggestion.

  3. 03

    Keep the camera honest

    Ask for one move per shot. Stacking a pan, a zoom and a rack focus into one beat is what makes motion drift.

Where LTX 2.5 Fits

Product motion tests

Animate a still of glass, liquid or fabric to see how the physics reads before booking a shoot.

Short-form story beats

Build a three-cut vertical scene with one recurring character for social placements.

Ad concept boards

Show a client the cut rhythm in motion instead of a static frame deck.

Continuity drafts

Draft a sequence where wardrobe and face have to survive a location change.

Pricing

Credits never expire and finished generations stay in your library for six months. The panel shows the exact credit cost before you generate.

Basic

$19.90

400 credits

  • 400 Credits - $0.050/Credit
  • 480p / 720p / 1080p output
  • Up to 133s of 1080P video
  • Commercial License
Start Creating

Pro

Popular

$49.90

1150 credits

  • 1150 Credits - $0.043/Credit
  • 480p / 720p / 1080p output
  • Up to 383s of 1080P video
  • Priority Generation
Start Creating

Elite

$99.90

2500 credits

  • 2500 Credits - $0.040/Credit
  • 480p / 720p / 1080p output
  • Up to 833s of 1080P video
  • Maximum Render Priority
Get Elite Access

See full plans

FAQs: LTX 2.5

Straight answers about the model, this workspace and how credits work.

The August 2026 release in the LTX open world-model line: a 22B dual-stream diffusion transformer with native multishot consistency. On this site it runs as a browser workspace for text-to-video and image-to-video.

2.5 adds the heavier dual-stream backbone and native multishot identity lock. The earlier 2.3 workspaces stay available if you prefer that stack.

No. ltx2ai.net is a third-party tool that runs LTX-family models. We are not Lightricks, and nothing here implies partnership, endorsement or an authorized console.

Public launch figures cite ~6.8 seconds for a 10s 720p image-to-video job on dual GB200 hardware, and ~23.7 seconds on the hosted API. Those are model benchmarks, not a promise for every job here — queue depth, resolution and duration all move it.

New accounts get a starting allowance; larger volumes use credit plans. Credits never expire and outputs stay in your library for six months. The panel shows the exact cost before you submit.

The model keeps one character stable across intentional hard cuts inside a single generation, instead of hoping three separate clips match. Watch the barista, courier and painter clips above.

The panel exposes the aspect ratios, resolutions, durations and frame rates that the LTX 2.5 tier you selected supports. Fast and Pro do not expose an identical set, so the options refresh when you switch tier.

No. Generation runs server-side and the finished clip comes back to your browser. A laptop and a stable connection are enough — open weights only matter if you want to self-host separately.

Paid plans on this site include a commercial license for what you generate here. The underlying open model carries its own terms from its owner, so review those separately if you plan to self-host the weights.

Start Creating with LTX 2.5

Open the workspace above, drop a prompt or a still, and render the cut you actually need.

Generate with LTX 2.5

Don't just watch the AI revolution—direct it.