Introducing LTX 2.3 Image-to-Video on WaveSpeedAI

Bring Your Images to Life with LTX-2.3 Image-to-Video on WaveSpeedAI

Static images tell a story. Moving images with sound make audiences feel it. With LTX-2.3 Image-to-Video now available on WaveSpeedAI, you can transform any still image into a high-fidelity video — complete with synchronized audio — in a single generation pass. No post-production. No separate audio tools. Just upload, prompt, and play.

Built by Lightricks on the Diffusion Transformer (DiT) architecture, LTX-2.3 represents a leap forward in unified audio-video generation. Where most image-to-video models produce silent clips that require separate sound design, LTX-2.3 generates motion and audio together as one coherent output. The result is animated content that feels whole from the first frame.

What Is LTX-2.3?

LTX-2.3 is the latest iteration of the LTX-2 model family — a 19-billion-parameter foundation model split roughly into 14 billion parameters for video processing and 5 billion for audio. It is one of the first open-source models capable of generating synchronized audio and video within a single unified architecture, using cross-attention mechanisms to keep sound and motion perfectly aligned.

The “2.3” release introduces meaningful improvements over its predecessor: a rebuilt VAE (Variational Autoencoder) trained on higher-quality data, an upgraded HiFi-GAN vocoder for cleaner audio output, stronger image-to-video consistency, and better prompt adherence throughout the generation pipeline.

Key Features

• Synchronized Audio-Video Generation: Sound isn’t bolted on as an afterthought. Ambient noise, music, dialogue cues, and sound effects are generated alongside visual motion in a single pass, eliminating the need for separate audio workflows.

• New VAE for Sharper Details: The rebuilt latent space in LTX-2.3 preserves fine textures, facial features, hair, text, and edge detail across the full frame. Outputs are visibly sharper than previous versions.

• Cleaner Audio Output: An improved HiFi-GAN vocoder reduces noise artifacts and silence gaps. Dialogue, ambient sound, and music come through with noticeably greater clarity.

• Faithful Image Preservation: The model maintains the subject, composition, framing, and lighting of your reference image while adding natural, coherent motion — no identity drift or visual degradation.

• Flexible Resolution and Duration: Generate video at 480p, 720p, or 1080p, with durations ranging from 5 to 20 seconds, letting you balance quality, cost, and creative needs.

• Portrait and Landscape Support: Native 9:16 portrait mode makes it easy to produce content optimized for social platforms like Instagram Reels, TikTok, and YouTube Shorts.

• 24/48 FPS Options: Choose the frame rate that matches your output requirements, from standard playback to smoother high-frame-rate delivery.

Real-World Use Cases

Product Marketing

Turn product photography into dynamic showcase videos. Upload a hero shot of a sneaker, a skincare bottle, or a piece of furniture, and LTX-2.3 animates it with subtle motion — a rotating view, shifting lighting, environmental atmosphere — while generating matching ambient audio. What once required a videographer and sound designer can now be drafted in seconds.

Social Media Content

The demand for short-form video is relentless. LTX-2.3 lets creators convert their strongest still images into scroll-stopping animated posts with built-in sound. A landscape photograph becomes a cinematic moment with wind and birdsong. A food photo becomes a sizzling, steaming clip ready to post.

Portrait and Character Animation

Animate headshots, portraits, and character artwork with natural movement. The model excels at preserving facial identity while adding lifelike motion — subtle head turns, blinking, expression changes — making it valuable for digital avatars, creative projects, and personalized content.

Storyboarding and Pre-Visualization

For filmmakers and creative directors, LTX-2.3 transforms static storyboard frames and concept art into animated sequences with synchronized audio. This accelerates pre-production by giving stakeholders a tangible feel for pacing, mood, and sound design before a single frame is shot.

E-Commerce and Advertising

Static product listings lose attention. Animated product videos with ambient sound increase engagement and conversion rates. LTX-2.3 makes it practical to generate video assets at scale — iterate quickly at 480p, then render final assets at 1080p.

Getting Started on WaveSpeedAI

Running LTX-2.3 Image-to-Video on WaveSpeedAI is straightforward. With no cold starts and fast inference, you get results in seconds rather than minutes.

import json
import os
import time
from urllib.request import Request, urlopen

api_key = os.environ["WAVESPEED_API_KEY"]
headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
payload = {
    "prompt": "A cinematic ocean wave at sunrise, highly detailed",
    "image": "https://interactive-examples.mdn.mozilla.net/media/cc0-images/painted-hand-298-332.jpg",
    "resolution": "720p",
    "duration": 5,
    "seed": -1
}

def request_json(url, data=None):
    request = Request(url, data=data, headers=headers, method="POST" if data else "GET")
    with urlopen(request) as response:
        return json.load(response)

# 1. Submit the prediction.
submit_body = request_json("https://api.wavespeed.ai/api/v3/wavespeed-ai/ltx-2.3/image-to-video", json.dumps(payload).encode())
task = submit_body.get("data", submit_body)
prediction_id = task.get("id")
if not prediction_id:
    raise RuntimeError("Submission response did not contain a prediction id")
result_url = task.get("urls", {}).get("get") or f"https://api.wavespeed.ai/api/v3/predictions/{prediction_id}/result"

# 2. Poll until the prediction finishes.
while True:
    body = request_json(result_url)
    result = body.get("data", body)
    status = result.get("status")
    if status == "completed":
        print(result.get("outputs", []))
        break
    if status in {"failed", "cancelled", "timeout"}:
        raise RuntimeError(result)
    if status not in {"created", "processing"}:
        raise RuntimeError(f"Unexpected status: {status}")
    time.sleep(2)

You can also specify resolution and duration:

import json
import os
import time
from urllib.request import Request, urlopen

api_key = os.environ["WAVESPEED_API_KEY"]
headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
payload = {
    "prompt": "A futuristic city at blue hour with soft volumetric light",
    "image": "https://interactive-examples.mdn.mozilla.net/media/cc0-images/painted-hand-298-332.jpg",
    "resolution": "720p",
    "duration": 5,
    "seed": -1
}

def request_json(url, data=None):
    request = Request(url, data=data, headers=headers, method="POST" if data else "GET")
    with urlopen(request) as response:
        return json.load(response)

# 1. Submit the prediction.
submit_body = request_json("https://api.wavespeed.ai/api/v3/wavespeed-ai/ltx-2.3/image-to-video", json.dumps(payload).encode())
task = submit_body.get("data", submit_body)
prediction_id = task.get("id")
if not prediction_id:
    raise RuntimeError("Submission response did not contain a prediction id")
result_url = task.get("urls", {}).get("get") or f"https://api.wavespeed.ai/api/v3/predictions/{prediction_id}/result"

# 2. Poll until the prediction finishes.
while True:
    body = request_json(result_url)
    result = body.get("data", body)
    status = result.get("status")
    if status == "completed":
        print(result.get("outputs", []))
        break
    if status in {"failed", "cancelled", "timeout"}:
        raise RuntimeError(result)
    if status not in {"created", "processing"}:
        raise RuntimeError(f"Unexpected status: {status}")
    time.sleep(2)

Pro tip: Start with 480p and short durations to dial in your prompt and motion direction. Once you have the result you want, scale up to 1080p for final delivery. Use a fixed seed when comparing prompt variations so you can isolate exactly what changed.

Pricing

LTX-2.3 on WaveSpeedAI starts at just $0.10 for a 5-second clip at 480p, scaling up to $0.80 for a 20-second 1080p video. No subscriptions required — pay only for what you generate.

Resolution	5s	10s	15s	20s
480p	$0.10	$0.20	$0.30	$0.40
720p	$0.15	$0.30	$0.45	$0.60
1080p	$0.20	$0.40	$0.60	$0.80

Why WaveSpeedAI?

In a landscape where synchronized audio-video generation is rapidly becoming the standard — with models like Veo 3.1, Kling 3.0, and Sora 2 all pushing the boundaries — LTX-2.3 stands out as a powerful open-source option with production-grade quality. And running it on WaveSpeedAI gives you the infrastructure to match: fast inference with no cold starts, simple API integration, and pricing that makes experimentation affordable.

Whether you’re a solo creator animating social content or a team generating video assets at scale, the combination of LTX-2.3’s unified audio-video generation and WaveSpeedAI’s optimized infrastructure means less time waiting and more time creating.

Start Creating

The gap between a still image and a complete video with sound has never been smaller. Try LTX-2.3 Image-to-Video on WaveSpeedAI today and see what your images sound like in motion.