Text → video · Image → video · Audio → video
VideoPopularFast

LTX-2.3 22B

ltx23-22b-fp8_t2v_distilled

Lightricks' audio-driven video model — fast Distilled or full-quality Dev, with native dialogue and sound.

No GPU. No setup. Cancel anytime.

About

LTX-2.3 22B is Lightricks' audio-driven video model: production-ready text-to-video with native dialogue and sound generated in the same pass.

Two builds share the same LoRA ecosystem and control workflows: Distilled runs an eight-step schedule for fast, cheap generation (the default), and Dev runs the full schedule for maximum coherence and quality at higher step counts.

Variants

LTX-2.3 22B ships in 2 builds — same family, different speed/quality trade-offs. Pick the variant that fits, or start with the default.

Variant Steps From
Distilled Fast
ltx23-22b-fp8_t2v_distilled
20 default · 2–60 13.8 Spark · $0.069 Create →
Dev Quality
ltx23-22b-fp8_t2v_dev
20 default · 2–60 34.6 Spark · $0.17 Create →

Prompting tips

LTX-2.3 is audio-driven and cinematic — the strongest prompts describe four things. Use the prompt field’s wand button to auto-expand a short idea into this shape.

  • What — the subject, action, and setting.
  • Feel — style, mood, and genre.
  • Camera — lens, angle, and how it moves.
  • Time — motion type, speed, and frame-rate intent.

LoRAs

LTX-2.3 has grown a real LoRA ecosystem, and Sogni ships with it wired in — Lightricks’ official adapters drive the control workflows, a research ID-LoRA handles voice identity, and community LoRAs can be attached to a render straight from the app or API. Nothing to download, no graphs to build.

LoRA What it does How you use it
Union Control
Lightricks IC-LoRA
Canny edge, depth-map, and pose-skeleton control in a single adapter — steer structure, camera moves, and body motion from a reference video. Powers the video → video canny, depth, pose, and detailer modes.
In/Outpainting
Lightricks IC-LoRA
Masked video editing — remove or replace anything in the shot, or extend the frame beyond its original borders. Powers the video → video inpaint and outpaint modes.
TalkVid-3K
ID-LoRA · research
Identity-preserving speech — a reference image and voice clip drive likeness and vocal identity together in a single generative pass. Applied automatically when you add reference audio to a text → video or image → video render.
Transition
ValiantCat · community
First-frame → last-frame morph transitions with unusually good motion continuity. Trigger word: zhuanchang. Attach it to an image → video render in the app, or pass loras: ['transition'] in the API.

The attachable catalog is delivered server-side, so when a new LTX-2.3 community LoRA is worth running, Sogni can light it up across the whole Supernet without an app update — expect this list to keep growing.

Pricing

On the Sogni API you pay per clip in Spark (1 Spark = $0.005). Prefer a flat rate? The Sogni app runs LTX-2.3 22B and 100+ other models credit-free under fair use.

Configuration Spark USD
5 s · 1280 × 720 · 8 steps 13.8 Spark $0.069
10 s · 1280 × 720 · 8 steps 27.6 Spark $0.14
5 s · 1920 × 1088 · 8 steps (scales with pixels) 31.4 Spark $0.16

API

One Sogni API key reaches every model on the Supernet — call LTX-2.3 22B by id.

import { SogniClient } from '@sogni-ai/sogni-client';

const client = await SogniClient.createInstance({
  appId: crypto.randomUUID(),
  apiKey: process.env.SOGNI_API_KEY,
  network: 'fast',
});

const project = await client.projects.create({
  type: 'video',
  modelId: 'ltx23-22b-fp8_t2v_distilled',
  positivePrompt: 'a slothicorn surfing a wave of liquid paint, slow push-in, cinematic',
  numberOfMedia: 1,
  duration: 5,
});

const [url] = await project.waitForCompletion();
console.log(url); // result link — download within 24h
curl https://api.sogni.ai/v1/creative-agent/workflows \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SOGNI_API_KEY" \
  -d '{
    "input": {
      "title": "First render",
      "steps": [{
        "id": "step1",
        "toolName": "generate_video",
        "arguments": { "prompt": "a slothicorn surfing a wave of liquid paint, slow push-in, cinematic", "videoModel": "ltx23", "duration": 5 }
      }]
    },
    "confirm_cost": true
  }'

REST workflow alias: ltx23. Pin the exact id with the SDK. Full reference at docs.sogni.ai.

Why run it on Sogni

Unlimited, not metered

Most platforms ration credits or rent a cloud GPU by the hour. Sogni gives you a flat monthly price for credit-free fair-use generation, powered by a creator-owned GPU network — or per-render Spark pricing if you'd rather pay as you go via the API.

No credits, ever

One flat price in the app. Generate under fair use without a per-image meter.

🧩

100+ models

Image, video, music, and language models in one workspace and one API key.

Per-render API

Prefer pay-as-you-go? Call LTX-2.3 22B by id and pay in Spark.

🌐

Powered by people

Runs on a decentralized GPU network where workers share subscription revenue.

FAQ

LTX-2.3 22B on Sogni

Which LTX-2.3 variant should I use?

Start with Distilled — eight steps, fast and cheap, and the default. Switch to Dev when you want maximum coherence and quality and can spend more steps and time. Both share the same LoRA control workflows.

Start with LTX-2.3 22B today

Create in the app, or build with the API. Your call.