LTX 2.5: Fast, Pro or Audio to Video, and What Each Mode Costs
LTX 2.5 is Lightricks' open-weights video model, with sound generated in the same pass. Fast runs to 20 seconds at 1080p and to 4K at 10 seconds, Pro stops at 10 seconds and 1080p, and Audio to Video times the picture to your own track. Checked against Lightricks and the API schemas on 30 September 2026.

LTX 2.5 is Lightricks' current video model, and the Studio runs it in three modes. Fast generates 6 to 20 seconds, with optional sound, at up to 4K. Pro stops at 10 seconds and 1080p. Audio to Video takes a track you supply and generates picture timed to it.
A 6-second Fast clip at 720p costs 42 credits. The same shot on Pro costs 54. Audio to Video bills per second of your track: 9 credits a second on the standard mode, 12 on Pro.
The thing most worth knowing is that Pro is not an upgrade from Fast. Its ceilings are lower on both axes, length and resolution. The API's endpoint descriptions put Fast in "a speed-optimized mode built for rapid iteration and previews" and Pro in "a quality-optimized mode for final, high-fidelity output". Pro buys finish per frame. Fast buys length and pixels.
What LTX 2.5 Is
Lightricks released LTX-2.5 on 11 August 2026. Its own page calls it "an open-weights world model for video generation", and lists what is new: "a better video decoder for cleaner motion, native multi-shot scenes that stay consistent across cuts, stronger prompt understanding, automatic clip duration, beta precise video editing, native 4K HDR and RAW support, and a faster, higher-quality distilled model."
Two things set it apart from most of the roster.
Sound comes from the same pass. The API documentation says LTX 2.5 generates "synchronized video and audio from a text prompt in a single pass", and that native audio "is included at every resolution".
You can own the weights. Lightricks publishes LTX-2.5 as open weights "with a permissive license, no mandatory branding, and free use under $10M ARR". You can download it and run it on your own hardware, or skip the hardware and run it here on credits.
Not everything on that list reaches the Studio, and it is better to know which parts before you plan around them.
- Automatic duration is not used. The endpoints accept "auto" for length, but the Studio always sends an explicit length. Credits are charged before the job starts, and a length the model picks for itself cannot be priced in advance.
- HDR and RAW are not offered. The endpoints the Studio uses return MP4 and expose no HDR or RAW option.
- Native multi-shot is untested by us. Lightricks leads with it. We have not yet checked how it behaves through these endpoints.
The Three Modes, Checked Against the Schema
Verified against the API schemas for the LTX 2.5 endpoints, the API documentation and Lightricks' model page on 30 September 2026:
| Fast | Pro | Audio to Video | |
|---|---|---|---|
| Length | 6 to 20 seconds, in 2-second steps | 6, 8 or 10 seconds | Set by your track |
| Resolution | 720p, 1080p, 1440p, 4K | 720p, 1080p | 1080p |
| Aspect ratios | 16:9, 9:16 | 16:9, 9:16 | Auto, 16:9, 9:16 |
| Frame rate | 24, 25, 48 or 50 | 24, 25 or 50 | Not selectable |
| Sound | Optional, same price | Optional, same price | Your own track |
| Start and end frame | Yes | Yes | Optional first-frame image |
| Camera move presets | 8 | 8 | None |
Two limits in the schema matter more than anything in that table.
Long clips need 720p or 1080p. The schema's duration field is specific: "At 720p and 1080p, 24 or 25 FPS supports up to 20 seconds, while 48 or 50 FPS supports up to 10 seconds. At 1440p and 2160p, all frame rates support up to 10 seconds." Treat 20 seconds as a 1080p-and-below length, and keep 1440p and 4K shots to 10 seconds.
Audio to Video has a length window. The schema asks for a track "between 2 and 20 seconds; pro models support a maximum of 10 seconds." Trim the track to the section you want picture for before you upload it.
The Studio applies both limits before you pay. At 1440p and 4K it caps a take at 10 seconds and charges only for the seconds it renders, and Audio to Video turns away a track outside the window, with a message, before any credits are taken.
Three smaller points:
- Sound costs nothing extra, but starts off. The price is identical with audio on or off. The Studio's console starts LTX with audio switched off, so turn it on if you want the soundtrack.
- Frame rate is free too. Billing is per second of output, and 24fps or 50fps does not change the second count.
- A start frame sets the shape. With an image to video shot, the Studio lets the image decide the aspect ratio, which the schema calls "auto". Prepare start frames at 16:9 or 9:16, the two shapes LTX 2.5 generates.
What It Costs
Every figure is the Studio price in credits, shown before you press Generate. A "10s maximum" cell marks a length the model does not support at that setting.
LTX 2.5 Fast
| Length | 720p | 1080p | 1440p | 4K |
|---|---|---|---|---|
| 6 seconds | 42 | 61 | 89 | 140 |
| 10 seconds | 70 | 101 | 148 | 233 |
| 20 seconds | 140 | 202 | 10s maximum | 10s maximum |
LTX 2.5 Pro
| Length | 720p | 1080p |
|---|---|---|
| 6 seconds | 54 | 77 |
| 8 seconds | 72 | 102 |
| 10 seconds | 90 | 128 |
LTX 2.5 Audio to Video (billed on the length of your track)
| Track length | Audio to Video | Audio to Video Pro |
|---|---|---|
| 5 seconds | 45 | 60 |
| 10 seconds | 90 | 120 |
| 20 seconds | 180 | 10s maximum |
Pro runs about 29% dearer than Fast at 720p (54 against 42 credits for 6 seconds) and about 26% dearer at 1080p (77 against 61).
For the provider cost behind every model on one page, see the Render Ledger. For how LTX sits against the rest of the video roster per second, see what AI video costs per second.
How to Prompt It
We have no observed prompting notes of our own for LTX 2.5 yet. For Fast and Pro, the Studio's Prompt Director (1 credit a pass) carries the published LTX guidance, labelled as provider guidance rather than something we have confirmed. The examples below apply that guidance. They are examples of structure, not results we are showing you.
The published guidance asks for cinematic, single-subject scenes with clear camera language, consistent lighting and described audio. Put spoken lines in quotation marks, and name the language or accent when it matters. Say how the camera moves and how the subject looks after the move, so the motion resolves. Show emotion through physical cues rather than labels. And do not leave the audio unspecified, because it is generated in the same pass.
Fast, text to video, one shot with sound.
Medium close-up of a fisherman in a yellow oilskin coiling rope on
a wet harbour wall at dusk. Sodium lamps behind him, cold blue sky.
The camera pushes in slowly until his weathered hands fill the
lower third of the frame. He pauses, looks out to sea, and his jaw
tightens. Audio: gulls, rope creaking, a distant diesel engine
idling, water slapping against the wall.
What we would pick: Fast, 1080p, 10 seconds, 24fps, audio on, camera move left on "Let model decide" because the prompt already describes the move. That costs 101 credits.
The console also offers eight camera move presets: static, dolly in, dolly out, dolly left, dolly right, jib up, jib down and focus shift. We have not tested how a preset interacts with a move written into the prompt, so our own rule is one or the other per shot, never both.
Pro, a line of dialogue.
Two-shot in a small bookshop, afternoon light through the window.
An older woman behind the counter slides a wrapped book across to
a young man. She says, in a soft Irish accent, "Mind the last
chapter." He smiles and tucks it under his arm. Audio: the shop
bell, rain on the glass, her line clear over the room tone.
What we would pick: Pro, 1080p, 8 seconds, audio on. That costs 102 credits. Pro here follows the provider's own positioning for final, face-led shots. It is not the result of a comparison we have run.
Image to video, with a start and an end frame. Upload a still of a closed watch box as the start frame and the same box open as the end frame. The schema says an end frame "generates a transition video between start and end frames".
The lid lifts slowly and the watch inside catches the light as the
camera jibs down to eye level. Audio: a soft click, then a low
synth swell.
What we would pick: Fast, 720p, 6 seconds. That costs 42 credits.
Audio to Video, a music bed. Upload the 12 seconds of track you want picture for.
A night drive through a city in the rain, neon reflections sliding
across the windscreen, the driver's eyes in the rear-view mirror,
the road ahead, all in time with the beat.
What we would pick: Audio to Video (standard), 16:9. Twelve seconds of track costs 108 credits. The schema requires a prompt, an image or both alongside the track. With an image, it becomes "the first frame of the video" and the prompt "describes how the image should be animated".
Fast, Pro, or Another Model
Our working plan for the pair follows the provider's positioning, and it is a plan rather than a verdict. Block the shot on Fast at 720p, where a 6-second try costs 42 credits. Move to Pro at 1080p for the final when the shot fits inside 10 seconds and carries a face. Stay on Fast and go to 1440p or 4K when you need more pixels rather than more finish per frame.
Outside the LTX line, four comparisons come up most:
- Veo 3.1 Lite is the cheaper short clip, at 20 credits for 8 seconds at 720p with sound. It stops at 1080p and 8 seconds, so length and resolution are where LTX takes over.
- Wan 3.0 runs to 30 seconds and accepts far more references (10 images, 5 clips and 5 audio files). At 1080p the two cost almost the same: 20 seconds is 200 credits on Wan 3.0 and 202 on LTX Fast. At 720p Wan is cheaper, at 5 credits a second against Fast's 7.
- MiniMax H3 answers a question people genuinely search for: what can LTX 2.5 do that H3 cannot? Going by the two models' schemas, LTX Fast runs longer (20 seconds against H3's 15), lets you switch sound off (H3 always generates it), offers 48 and 50fps and camera presets, and can drive picture from your own audio track. H3 goes the other way on references (up to 9 images, 3 clips and 3 audio files), offers 1:1, 4:3, 3:4 and 21:9, and is far cheaper at 4K: 80 credits for 10 seconds against Fast's 233.
- Existing footage. To re-crop a clip you already have to a new aspect ratio, use LTX 2.3 Reframe. To make a speaker in existing footage match a new audio track, use VEED Lipsync v2. Audio to Video generates new picture; neither of those does.
Start frames can come from the Studio's image models. SeeDream 5 and Nano Banana both render 16:9 and 9:16.
What We Have Not Tested
We have verified the schemas, the length and resolution limits, the list prices and the Studio's credit ladder for all four LTX 2.5 modes. We have not run a controlled comparison of Fast against Pro, and we have no measured first-try hit rate for any of them. Nothing in this post is a quality verdict.
Still open:
- Whether Pro's per-frame fidelity is visible at 1080p against Fast on the same prompt.
- How native multi-shot behaves through these endpoints, and how well the generated audio and lip movement line up with dialogue.
- Whether 48 or 50fps earns its 10-second limit for fast action.
- How the camera presets interact with a camera move written into the prompt.
When we have results, they go here with a date on them.
LTX 2.5 Questions
What is the difference between LTX 2.5 Fast and Pro? Fast reaches further and Pro is tuned for finish. Fast generates 6 to 20 seconds at up to 4K, although 1440p and 4K stop at 10 seconds, while Pro offers 6, 8 or 10 seconds at 720p or 1080p. The endpoint descriptions position Fast for rapid iteration and previews and Pro for final, high-fidelity output. In the AIVideos Studio a 6-second clip at 720p costs 42 credits on Fast and 54 on Pro.
How much does LTX 2.5 cost? In the AIVideos Studio, LTX 2.5 Fast costs 42 credits for 6 seconds at 720p, rising to 233 credits for 10 seconds at 4K. Pro costs 54 credits for 6 seconds at 720p and 128 for 10 seconds at 1080p. Audio to Video costs 9 credits per second of your track, or 12 on the Pro mode. Every price is shown before you generate.
How long can LTX 2.5 videos be? Up to 20 seconds in one generation on LTX 2.5 Fast, at 720p or 1080p and 24 or 25 frames per second. At 1440p or 4K, or at 48 or 50 frames per second, the limit is 10 seconds. Pro stops at 10 seconds at any setting.
Does LTX 2.5 generate audio? Yes. Fast and Pro generate sound in the same pass as the picture, and the price is the same with audio on or off. The Studio's web console starts LTX with audio switched off, so turn it on if you want a soundtrack. Audio to Video works the other way round: you supply the track and it generates picture timed to it.
What is LTX 2.5 audio to video? It generates video timed to an audio track you upload, such as a music bed or a line of dialogue, steered by an optional prompt and an optional image used as the first frame. It outputs 1080p and bills per second of the track: 9 credits per second, or 12 on the Pro mode. The model accepts tracks of 2 to 20 seconds, or 10 seconds at most on Pro, so trim a longer track before you upload it.
Is LTX 2.5 open source? Lightricks releases LTX-2.5 as open weights under a permissive licence with no mandatory branding, and says it is free to use below $10M in annual recurring revenue. You can download the weights and run them yourself, or run the model in the AIVideos Studio on credits with no hardware of your own.
Start With One Fast Shot
The cheapest useful test is a single 6-second Fast clip at 720p, for 42 credits, in the Studio. Keep the prompt to one subject, one move and one described soundscape, and switch the audio on before you press Generate.
If the shot is part of a bigger film and you would rather hand the whole brief over, that is what our production services are for.
Generate cinematic AI video, from €15
31 AI models. No subscription. Buy credits, generate on demand, own the results outright.
