LTX-2.5 22B

About

LTX-2.5 22B generates video and synchronized sound together. Create from text, animate one image, guide a transition with opening and closing frames, generate from audio alone, or combine an image with an audio track.

The video-to-video family supports canny, depth, pose, detailer, inpaint, and outpaint workflows. Pose uses a source video for motion plus a still image for the subject's appearance; inpaint uses a mask to identify the region to replace.

Fast, HQ, and Pro currently use the same release-validated official Distilled/Turbo path. Dev checkpoints are not publicly routed until Lightricks publishes an official ComfyUI Dev recipe and Sogni validates it.

Choose an LTX-2.5 workflow

Five public model IDs cover generation and editing. First-and-last-frame video uses the I2V ID with its dedicated template; Fast, HQ, and Pro all route to these release-validated Distilled/Turbo workflows.

Every mode is measured side by side in the Sogni Engineering field guide, H3 directs. LTX-2.5 brings the camera rig.

What LTX-2.5 changes

Output is native 1920×1088 at 24 fps by default, and the same paths accept resolutions up to 4K and frame rates up to 60 fps at 1080p. Audio is generated jointly with the picture in 48 kHz stereo, and every public launch workflow renders at a fixed 8 steps — roughly a third of the price of the 30-step Dev tier.

The gain that matters most is continuity across a cut. Given prose with a timestamped hard cut, LTX-2.5 treats the shots as one scene and holds character, wardrobe, environment and lighting across the edit, where LTX-2.3 re-dressed the same character at every cut. A custom Gemma 4 12B text encoder holds longer captions, and the newer decoder keeps skin texture, typography, product detail and fast motion cleaner.

Measured render times

One render each on Sogni Supernet fast-network workers, 2026-08-14. Fast-network GPUs differ, so treat every row as one honest sample rather than a benchmark.

Prices are live network estimates. Every LTX-2.5 render is also covered credit-free under fair use on Sogni Unlimited plans.

Prompting notes

• Write one flowing prose paragraph. Shot scale, scene, character, action, camera and audio in present tense. Name every cut in natural language rather than relying on [Shot N] markup alone.
• Direct the audio explicitly. Omitting sound design is not a mute switch — an underspecified caption can still produce generic ambience or a score. Write “Natural diegetic sound only. No music, no score, no soundtrack.” and ask for actual silence when that is what you need.
• Timestamp every cut. “At exactly 3.5 seconds, a hard cut switches to…”, re-identify the character on each shot, and say what the audio does across the cut. Expect cuts to land within about a second of the timestamp, keep the caption under roughly 140 words, and make the two shots genuinely different places — a close-up in the same room usually comes back as a push-in instead of an edit.
• Quote dialogue verbatim with the speaker and the delivery, and budget about three spoken words per second.
• For image-to-video dialogue, audition the still, not the caption. Across 28 takes of the same subject one portrait landed clean lip sync every time while a tighter macro crop landed about one in four, and it stayed there through eight caption strategies. If the mouth is not tracking the words after two or three takes, change the source image.
• State the camera on every shot — including “the camera remains static” when you want no move, or the model tends to invent a push-in that drifts off your subject.
• Video-to-video output dimensions must be multiples of 128, not the usual 64. Canny, depth and pose read structure straight from the source video and share one strength control (0.3–1.0, default 0.85) with the detailer; inpaint adds a mask and outpaint a canvas position.

Compatibility boundaries

LTX-2.3 remains available as an explicit rollback family. Voice ID-LoRA, the community transition LoRA, and 10Eros remain LTX-2.3-only and are not attached to LTX-2.5 requests.

Pricing

LTX-2.5 22B is included on every Sogni plan — generate credit-free under fair use, alongside 200+ other models. Pay-as-you-go Spark packs are also available for eligible Spark renders.

API

One Sogni API key reaches every model on the Supernet — call LTX-2.5 22B with the exact model id for the variant you want.

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: 'ltx25-22b-int8_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": "ltx25", "duration": 5 }
      }]
    },
    "confirm_cost": true
  }'

Example uses ltx25-22b-int8_t2v_distilled. Use any exact model id from the Variants table above. Full reference at docs.sogni.ai.