Short answer: yes, when the shot is built on a voice — dialogue, a spokesperson, a narrator reading a line to camera — because LTX-2.5 renders speech, ambience and sound effects in the same pass as the picture and quotes you 80 credits for a five-second 720P keeper before a token is spent. No, when the brief is heavy action or expressive camera motion (that is Wan 3.0's contract), a clip past twenty seconds (Wan again, at 30), a reference kit heavier than a mood board (Seedance 2.5's 30/10/10 lane), or raw $0.09-per-second API time you are happy to babysit a pipeline around. LTX-2.5 is the current generation of Lightricks' audio-native family — released August 11, 2026, the successor to the LTX-2 line that shipped open weights in January 2026 — and on Moosky it bills on the shared credit balance like every other engine: 100 credits to the dollar, the exact cost on the button before you press it, failed provider runs returned. Its sibling guide Is LTX 2.0 Pro Worth It in 2026? runs the history and the licensing question for the previous generation; this note is the buying math for the one you would actually render today.

Quick answers

Is LTX 2.5 worth it in 2026? Yes, when the clip is carried by a voice: LTX-2.5 renders dialogue, ambience and effects in the same pass as the picture, bills a flat per-second price by resolution tier, and at roughly 13.85 credits per second at 720P it is the mid-cost talking lane that still quotes true 4K, which neither flagship neighbor can. No, when the brief is heavy action or complex camera motion (the model's own registry tags route those away), a single take past twenty seconds (the composer caps at 20), or a reference bundle larger than a start and an end frame.

How much does LTX 2.5 cost per clip on Moosky? A flat per-second rate by resolution tier on a 10-credit run base, converted at 100 credits to the dollar at the house margin, billed credits quoted up front. The ladder: 66 credits for a 4-second 720P draft, 80 for the 5-second 720P default, 149 for a 10-second 720P cut, 287 for the full 20-second 720P run, 210 for 10 seconds at 1080P, 303 for a 10-second 1440P clip, 472 for 10 seconds at true 4K. Failed provider runs return their credits.

What changed since LTX 2.3 and the 2.0 generation? Lightricks shipped LTX-2.5 on August 11, 2026 with better prompt understanding, sharper visuals and native multishot continuity over the LTX-2 line, and it is the generation Moosky runs as the ltx-2-5 composer. The old Pro/Ultra API tier naming from the 2.0 launch is not on the menu — on Moosky you pick the resolution tier, the length and the frames, and the price badge does the rest.

What is LTX 2.5 actually good at, and what should you avoid it for? Its strengths are tagged in the registry: talking-audio delivery, simple controlled motion, multi-shot continuity and first/last-frame control at 720P through 4K. The same registry explicitly routes heavy action, complex motion and animation to other engines — and it carries the platform's stricter photoreal-human policy tag, so face-heavy briefs can bounce at the gate.

Does LTX 2.5 do first/last-frame control, and what do retakes cost? Yes: the composer takes an optional start frame and an optional end frame that keyframes the interpolation between them, and portrait, landscape or adaptive aspect costs the same. A retake is the same flat math again — which is why the ladder starts at 66-credit drafts: the unit economics only work if you never explore at the tier you ship at.

How the lane bills, honestly

LTX-2.5 on Moosky is the fleet's cleanest price sheet: there are no tokens to model, no aspect-rate surcharges, no audio line item. The billing code (the lane's own price calculator, mirrored by its cost tests) charges a flat per-second rate for the resolution tier you pick, marks it up at the platform's fixed 35% margin, converts at 100 credits = US$ 1 and adds a 10-credit run base per generation. The published ladder, all figures the billed credits and not the pre-rounding products:

Run Resolution Bills About
Draft a hook 4 s at 720P 66 credits $0.66
The default clip 5 s at 720P 80 credits $0.80
The approved cut 10 s at 720P 149 credits $1.49
Longest 720P run 20 s at 720P 287 credits $2.87
Full-HD final 10 s at 1080P 210 credits $2.10
2K hero, capped 10 s at 1440P 303 credits $3.03
True 4K, capped 10 s at 4K 472 credits $4.72

Read it alongside the source. The published LTX API pricing page lists its 1080p tiers at roughly $0.06-$0.17 per second and 4K at $0.32-$0.39; Moosky's Fast-tier per-second rate is $0.09 at 720p and $0.13 at 1080p before margin, which is why a 5-second 720P keeper costs 80 credits — under a dollar, mid-cost by design, never the fleet's bottom rung. Three structural rules ride with it: the generator renders 4-to-20-second clips with a 5-second default; 1440P and 4K are capped at 10 seconds and the badge clamps your request to the ceiling rather than quoting a render the provider will not make; and there is no 480P tier to hide in — the ladder floor is 720P, and the price badge on the button is exactly what charges.

Where the talking lane earns its credits

Judged on what it was built for, this is the fleet's talking lane: the engine you route dialogue briefs to when the clip also needs a 4K option. The model is a joint audio-video generator: write the line into the prompt and the voice arrives lip-timed to the frame, ambience and effects folded into the same pass, with no separate voice tool and no lip-sync step. Three named jobs:

  • The spokesperson pass. A founder reading a hook, an actor on a script, a product pitching itself in twenty words. Talking-head and dialogue delivery is the registry's own strength tag for this lane, and an 80-credit five-second 720P attempt prices the try at pocket change (the H3 lane undercuts it at draft tier; H3 cannot quote you 4K or a start/end-frame keyframe). The talking-head craft lives in the AI talking-head guide.
  • The continuation keeper. Multi-shot continuity means the same subject surviving beat to beat, and first/last-frame control lets you pin the opening and closing compositions and let the model interpolate the middle — a start-frame clip inherits its aspect adaptively, so a still becomes a shot without re-cropping the brief. The tested camera-move and dialogue recipes are in the LTX prompt guide.
  • The 4K short cut. This is the honest differentiator: a 10-second 4K hero bills 472 credits, and the two obvious neighbors cannot quote it at all — Seedance 2.5 tops out at 1080P and refuses the request outright, and Wan 3.0 caps at 1080P too. The full cross-family split is walked in the model matchup.

The four-way decision: LTX 2.5, Seedance 2.5, Wan 3.0

Same balance, four purchases, decided by what the brief leans on. For a dialogue or spokesperson clip, LTX-2.5 is the seat: 13.85 credits per second at 720P on flat per-second billing, versus Seedance 2.5's roughly 51 credits per second for the same seconds, and its native audio arrives in the picture pass rather than via a reference kit. For length, LTX-2.5 caps at 20 seconds (and at 10 once you climb to 1440P or 4K); Wan 3.0's crown is the unbroken 30-second take, and Seedance 2.5's is the same ceiling plus edit and extend modes that reshape an attached clip — if the keeper is one continuous move past twenty seconds, neither of those two jobs belongs here.

For pixels, the fork inverts: only LTX-2.5 renders 4K among the three, at 46.15 credits per second up to its 10-second cap. The number that makes the comparison concrete is the same clip priced twice: a 10-second 1080P cut bills 210 credits on LTX-2.5 against 308 for the same seconds on Wan 3.0's no-base lane and 1,267 on Seedance 2.5's token ladder — you are not paying extra for the LTX lane, you are paying less, and what you trade for it is the motion ceiling. For motion, the registry is blunt: LTX-2.5's own tags route heavy action, complex motion and animation away from it and to Wan's operated-camera lanes. The honest failure mode of a cheap lane is spending it on the wrong brief — three 80-credit 720P takes of a fight scene you will ship from the draft tier anyway is not thrift.

The worked run: a four-scene spokesperson film

Here is a brief you can price against the ladder before booking anything — a four-scene product explainer carried by one presenter, all at 720P with native audio on. Scene one, the hook: two 5-second 720P drafts until the line lands naturally (80 + 80), then the keeper. Scenes two and three, the demo: two 10-second approved cuts (149 + 149). Scene four, the close: one 8-second sign-off (121). The whole pass — drafts included — bills 579 credits. Then the deliverable asks for Full HD: re-render only the approved scene two at 1080P (210), and the four-scene film plus its one hero upgrade closes at 789 credits, about $7.89 — one US$ 5 top-up (500 credits) plus the 300-credit remainder carried into your next brief. The same film with its 8-second close re-shot twice before approval, the realistic retake pattern on any dialogue lane, adds 242 credits; the drafts are why that stays survivable. This shape of job runs straight through Project Director, which prices the whole scene list before the first clip bills and holds its production floor at 500 wallet credits; the per-clip route is the same composer without the planner.

Where it does not earn its keep

Decline the lane in four cases. Heavy action, athletic motion and expressive camera work: the model's tags avoid them and Wan 3.0 owns them — the same 30-second envelope Wan bills at per-second rates starts here at 287 credits only because the ceiling is shorter. Complex multi-beat physical staging and animation: same routing note, and the Wan 3.0 buying guide has that lane's math. A reference build: LTX-2.5 steers from text plus a start and end frame, not from a bundle of identities and voice tracks — that is the R2V territory the Seedance 2.5 guide walks, with its 30-image kits and voice-sample locks. And photoreal crowds or sensitive human scenes: the composer carries the stricter photoreal-human risk tag, so face-heavy briefs can bounce at the gate — draft first, and read the rejection-reasons guide before assuming a policy miss is a prompt miss.

The credit wall, stated plainly

A new account starts with 10 credits at signup plus 50 planning credits — enough to walk the prompt path and see the real quote on the button, not enough to render even the 66-credit 720P draft. Production money starts at the credits page US$ 5 floor (500 credits), which buys the whole worked film above in one purchase. If you are weighing pay-per-clip against a subscription elsewhere, the pay-per-clip vs subscription math and the cost-per-usable-clip guide settle it arithmetically — on any lane the keeper rate, not the sticker, is what you are buying, and failed provider runs return their credits either way.

Frequently asked questions

Is LTX 2.5 a subscription I sign up for? No. There is no plan attached to the model on Moosky: every run charges the shared credit balance, the exact cost is quoted on the button before you generate, and credits come back when the provider's run fails. One balance spans the whole generator fleet, so the engine question is a per-clip choice, not a monthly contract.

Is LTX-2.5 the same as LTX 2.0 Pro? Same family, different generation. LTX-2.5 is the August 2026 successor to the LTX-2 line — better prompt understanding, sharper visuals, native multishot — and it is what the ltx-2-5 composer renders; the "Pro" label was an API compute tier from the 2.0 launch, never a consumer plan, and the full history is in the 2.0 Pro buying guide.

Does the price change with aspect ratio or audio? No. Billing is flat per second of output by resolution tier; landscape, portrait and adaptive aspect cost the same, and native synchronized audio rides the picture pass without a separate line item. The axes that move the number are seconds and tier — plus the one-time 10-credit run base.

Why do 1440P and 4K cost per second so much more? Because the Fast-tier ladder climbs: roughly 13.85 credits per second at 720P, 20 at 1080P, 29.23 at 1440P and 46.15 at 4K — and the provider caps the two top tiers at 10 seconds, which the badge enforces by clamping longer requests instead of quoting an impossible render. The sane path is approving at 720P and re-rendering only the winning take at the hero tier.

Can the free balance render an LTX 2.5 test clip? No. The opening 10 plus 50 planning credits stop short of the cheapest billable run, the 66-credit 4-second 720P draft. It is sized to reach the quote, not to produce — the first render starts at the US$ 5 floor.

Is it better than Seedance or Wan? It is better at its own job. For lip-synced dialogue on a controlled micro-motion shot with a 4K option, it is the lane the fleet routes talking briefs to (the budget talking seat is the MiniMax H3 lane, whose math its own guide carries); for one continuous 30-second camera move Wan 3.0 wins, and for 30-image reference kits and edit-and-extend passes Seedance 2.5 does. The per-second head-to-heads above are the boundary, and the family matchup walks it clip by clip.

The verdict

LTX 2.5 is worth it when the clip is carried by a voice and cut short: an 80-credit five-second talking take, a 149-credit approved ten, a 472-credit ten-second 4K hero no other lane in the fleet can quote at all — priced per second, quoted before you press, refunded on failure. Hand the brief to Wan 3.0 when the camera is the subject, to Seedance 2.5 when the reference kit is, and to the metered API only when you are building software rather than shipping video. Decided? Open the generator with the LTX-2.5 composer and read the tag on the button — it prices every run up front, so nothing ever bills by surprise.