LTX 2.5 Pro and Fast Guide: Features, Examples & How It Compares

Lightricks put two versions of LTX 2.5 on Segmind, and the naming is misleading in a way that costs money. I assumed Pro was the capable one and Fast was the budget trim. It is not that simple. Fast is cheaper on every shared configuration, and it also reaches resolutions and durations that Pro cannot touch at any price.

So I ran both. Seven generations, every parameter confirmed against the API spec before firing, every cost read from the billing header rather than estimated. This guide covers what LTX 2.5 Pro and Fast actually do, what each one bills, the four constraints that are not in the docs, and how the pair compares to Seedance 2.5, Grok Imagine 1.5 and Veo 3.1.

What LTX 2.5 is

LTX 2.5 is Lightricks' video generation family, exposed on Segmind as two separate endpoints: ltx-2.5-pro and ltx-2.5-fast. Both take a text prompt or a single input image, both generate synchronized audio in the same pass, and both return the MP4 directly in the response body with no polling step.

The shared parameter surface is small and easy to reason about: prompt, an optional image, aspect_ratio at 16:9 or 9:16, resolution, duration, fps, generate_audio, a camera_motion enum, and an optional last_frame_uri. That is the whole API. There is no seed parameter on either model, which matters more than it sounds and I come back to it later.

Where they split is the ceiling. Pro tops out at 1080p and 10 seconds. Fast reaches 4K and 20 seconds. Here is the full capability split as the two specs define it:

Read that last row again. Fast is not a cut-down Pro. On the spec sheet it is a superset, and it undercuts Pro by 30.8% at the one resolution they share at the top end.

How LTX 2.5 pricing works

Both models bill a flat rate per second of output, fixed by resolution. Duration is perfectly linear: a 20 second clip costs exactly twice a 10 second clip. Aspect ratio does not change the price, and neither does frame rate, which I confirmed by accident and describe in the gotchas.

Every one of my seven fires billed the published rate to the cent. No surcharges, no rounding drift, no surprise line items. That is not something I can say about every video model I have tested this month.

Pro versus Fast, same prompt, same settings

The only honest way to price the Pro premium is to change nothing except the endpoint. Same prompt, same resolution, same duration, same camera motion, same frame rate, audio on for both.

Both came back at exactly 1920x1080 with a real audio track. Pro cost $0.30 more and took 14.8 seconds longer. Pro's file was 7.0 MB against Fast's 4.2 MB, which usually means more encoded detail or more motion.

If you are producing at volume, that $0.30 is the whole decision. An agency cutting 200 six second social clips a month pays $195 on Fast and $255 on Pro. Run the head to head on your own prompt before you commit to the premium, because it is the only test that reflects your material.

4K, which Pro cannot do at all

Fast's headline capability is 4K, and it is genuinely 4K. The container reported 3840x2160, not an upscaled 1080p and not a pixel budget dressed up as a tier.

For a VFX or previsualization team, 4K matters because you can crop, stabilize and reframe without giving up delivery resolution. At $0.375 per second it is the most expensive thing either model does, and it was also the slowest generation in my set at 60.1 seconds. Both of those are still cheap and fast by video model standards.

A 20 second multi-shot sequence

Both specs advertise consistent multi-shot scenes. Fast is the only one of the pair that can hold 20 seconds, which is where a multi-shot prompt has room to breathe. I gave it three shots and asked for the same subject and wardrobe throughout.

Twenty seconds of finished video with sound for $2.25, generated in 37 seconds, is a strong number for a production house cutting social edits or animatics. Note the timing: this 20 second clip rendered faster than the 6 second 4K one. Generation time tracks pixel count far more than it tracks duration, so long and small is the cheapest fast path.

Vertical for social

Both models take 9:16 natively, and the aspect ratio does not change the bill. The vertical clip cost exactly the same $0.975 as the 16:9 clip at the same resolution and duration.

At 1080p vertical, 9:16 gives you a true 1080x1920 frame. For an MCN or a brand social team, that is the native delivery spec for Reels, Shorts and TikTok with no reframing step.

Image to video, and what it costs

Both endpoints accept an image as the opening frame. I generated a product still with Nano Banana 2, uploaded it, and passed it to Pro.

Input  ·  image

Output  ·  $1.275

Image to video on LTX 2.5 Pro. Billed $1.275, identical to the text to video rate at the same settings.

The important finding is the bill: $1.275, exactly the same as text to video at 1080p for 6 seconds. There is no per image surcharge on LTX 2.5. Some competing endpoints do add one, so it is worth stating plainly that this one does not.

For e-commerce, that makes an existing product photo library the cheapest possible input. You already have the stills, and animating one costs no more than generating a scene from scratch.

Four things the docs do not tell you

1. The 20 second ceiling collapses to 10 at high resolution or high frame rate. Fast advertises 20 seconds and 4K, but never together. Ask for 20 seconds at 2K or 4K and it refuses. Ask for 20 seconds at 48 or 50fps and it also refuses, even at 720p. The API is explicit about it, which I appreciated:

{"error": "duration must be one of [6, 8, 10] at 4k/24fps. Got 20"}
{"error": "duration must be one of [6, 8, 10] at 720p/50fps. Got 20"}

The duration enum is a function of resolution and frame rate together. Long clips are a 720p and 1080p feature at 24 or 25fps only.

2. Invalid parameters are rejected, not silently clamped. Every bad enum I sent came back as a 400 with the valid set spelled out, and billed nothing. This sounds unremarkable until you have used a model that accepts an out of range duration, quietly clamps it, and charges you the full rate anyway. LTX 2.5 tells you what is wrong for free, so iterating on parameters costs only latency.

3. Audio and frame rate are both free. Setting generate_audio: false billed $0.975, exactly the same as leaving it on. The container confirmed the silent clip genuinely had no audio track while the others carried AAC. Frame rate is free too: a 10 second 1080p clip at 50fps billed $2.125, the same as 24fps. You get double the temporal resolution at no cost, so there is little reason to leave fps at the default for motion heavy work.

4. Every clip runs one frame long, and there is no seed. A 6 second request returns 145 frames at 24fps, which is 6.04 seconds. A 20 second request returns 481. Harmless for most uses, worth knowing if you are cutting to a hard timecode. More significant: neither model exposes a seed, so you cannot reproduce a generation exactly or hold a look steady across a batch. For any workflow that needs consistency across shots, that is the real limitation, not resolution.

How LTX 2.5 compares

Priced at a common configuration: 8 seconds, highest resolution each model reliably delivers, audio included where the model supports it.

Three things stand out. Veo 3.1 charges for audio as a multiplier: 8 seconds costs $1.60 silent and $3.20 with sound, which makes LTX 2.5 Fast less than half the price for a comparable 1080p clip with audio. Seedance 2.5 goes furthest on duration at 30 seconds and prices per token rather than per second, working out to roughly $1.91 for 8 seconds at 720p.

If you want 1080p with sound at the lowest price, LTX 2.5 Fast currently wins that specific bracket. If you want 4K at all, it is the only one on this list that delivers it.

Calling it from Python

Both endpoints are synchronous. You POST, you wait, you get an MP4 in the response body. No job IDs, no polling loop.

import requests

resp = requests.post(
    "https://api.segmind.com/v1/ltx-2.5-fast",
    headers={"x-api-key": "YOUR_API_KEY"},
    json={
        "prompt": "A barista pulls an espresso shot as steam curls through "
                  "warm morning light. Diegetic sound only: the hiss of the "
                  "steam wand, quiet cafe murmur. No music.",
        "resolution": "1080p",
        "duration": 8,
        "aspect_ratio": "16:9",
        "fps": 24,
        "generate_audio": True,
        "camera_motion": "dolly_in",
    },
    timeout=600,
)

print(resp.status_code, resp.headers.get("x-cost"))
with open("out.mp4", "wb") as f:
    f.write(resp.content)

Two practical notes. Set a generous timeout: my slowest fire took 60.1 seconds, but a client timeout does not cancel the generation, so a short timeout means you pay for a clip you never receive. And read the x-cost response header, which reports exactly what the call billed. It is the only reliable per call spend figure, so log it if you are tracking budget.

For image to video, add "image": "https://your-cdn/frame.jpg". The URL has to be publicly reachable.

Honest assessment

What LTX 2.5 does well: pricing is flat, linear and completely predictable, and all seven of my fires matched the published rate exactly. Validation is strict and free, so you find out about a bad parameter in under a second instead of after a charge. Audio, frame rate and aspect ratio are all included rather than upsold. Generation is fast, between 32 and 60 seconds across every configuration I tried.

Where it falls short: the absent seed parameter is the big one, because it rules out reproducible output and makes consistency across a batch a matter of luck. The 16:9 and 9:16 only aspect list is narrow next to models offering 1:1 and 21:9.

FAQ

What is LTX 2.5 Pro used for?

LTX 2.5 Pro generates up to 10 seconds of 1080p video with synchronized audio from a text prompt or a single image. It suits short cinematic shots, product films and previsualization where you want Lightricks' higher quality tier.

What is the difference between LTX 2.5 Pro and Fast?

Fast reaches 4K and 20 seconds and costs $0.1625 per second at 1080p. Pro stops at 1080p and 10 seconds and costs $0.2125 per second. Fast is cheaper and has the higher ceiling, so Pro is only worth it if its output quality justifies the 30.8% premium on your material.

How much does LTX 2.5 Pro cost?

LTX 2.5 Pro bills $0.15 per second at 720p and $0.2125 per second at 1080p. A 6 second 1080p clip with audio costs $1.275 and a 10 second one costs $2.125. Audio is included at no extra charge.

Does LTX 2.5 generate audio?

Yes. Both models produce synchronized audio in the same pass, on by default, at no extra cost. Setting generate_audio to false removes the audio track and bills exactly the same amount.

Can LTX 2.5 generate 4K video?

LTX 2.5 Fast generates true 3840x2160 at $0.375 per second, for 6, 8 or 10 seconds. LTX 2.5 Pro cannot generate 4K at all. At 4K the 20 second maximum drops to 10 seconds.

How do I use the LTX 2.5 Pro API?

POST to https://api.segmind.com/v1/ltx-2.5-pro with your x-api-key header and a JSON body containing at minimum a prompt. The response is the MP4 binary itself, returned synchronously with no polling required.

The takeaway

LTX 2.5 is one of the more honest video endpoints I have tested. The pricing is flat per second and it billed exactly as published on all seven runs, invalid parameters fail fast and free, and audio, frame rate and aspect ratio are all bundled rather than metered. The missing seed parameter is the one real gap.

Start with Fast. It is cheaper than Pro on every shared configuration and it reaches 4K and 20 seconds, neither of which Pro can do. Move to Pro only after a same prompt comparison on your own material convinces you the premium is buying something you can see.

Try LTX 2.5 Fast and LTX 2.5 Pro on Segmind. Both are one POST request, no setup required.