GPT Image 2From $0.007/imagefal.ai's 18-page LTX-2.5 launch, explained -- and three video models you can call on HiAPI today
Choose a model, enter your prompt, and see the result.
HiAPI Blog
HiAPI
Generate it with HiAPI
fal.ai just gave LTX-2.5 a full launch treatment — dedicated pages for text-to-video, image-to-video, and audio-to-video, each split into Fast and Pro tiers. That's the kind of rollout that usually means one thing: a lot of developers are about to go looking for "LTX-2.5 API" and quickly discover it isn't the only option worth evaluating.
LTX-2.5 isn't available inside HiAPI's catalog, so we won't be walking you through calling it here. What we can do is break down what the launch actually claims, then show you three video models that are live on HiAPI's /v1/tasks endpoint right now — with native audio, longer durations, or 4K output, depending on what you need.
According to fal.ai's own model pages, LTX-2.5 ships with two new pieces of underlying tech:
The model is split into two speed tiers (Fast and Pro) across three generation modes: text-to-video, image-to-video, and audio-to-video, where the audio-to-video mode syncs generated visuals to an existing audio track rather than generating sound itself. Pricing on fal.ai's pages runs roughly $0.09–$0.30 per second depending on tier and resolution.
That's a real, credible launch — DFR and the new decoder are fal.ai's own technical claims, not ours to verify independently. But a brand-new model landing across 18 pages at once also means it's untested at scale, and if you're building on HiAPI, it simply isn't in the lineup. Here's what already is.
We pulled this directly from HiAPI's /v1/models endpoint and pricing feed — no guessing on names, prices, or availability.
| Model | Native audio | Durations | Resolutions | Price |
|---|---|---|---|---|
| wan2.7-video | No | 2–15s | up to 1080p, 5 aspect ratios | $0.167/s |
| seedance-2.5 | Yes | 4–30s | 480p / 720p, 7 aspect ratios | $0.1395/s (t2v/i2v) · $0.1214/s (reference-to-video) |
| veo-3.1 / veo-3.1-fast | Yes | Fixed 4s / 6s / 8s | up to 4K | $0.57/s (standard) · $0.25/s (fast) |
Prices reflect HiAPI's public pricing feed at the time of writing and may change — always confirm current rates on the pricing page before estimating a project budget.
wan2.7-video is the budget pick for silent b-roll, product spins, or any clip you're going to score with your own audio track anyway. It covers both text-to-video and image-to-video, with five aspect ratios and durations from 2 to 15 seconds — flexible enough for most short-form social formats without paying for audio generation you won't use.
Seedance 2.5 is the one to reach for when the clip needs its own soundtrack — ambient noise, dialogue timing, or sound effects generated alongside the visuals. It also supports a reference-to-video mode, letting you feed a source clip and get a stylistically consistent continuation back, at a slightly lower per-second rate than plain text-to-video. At up to 30 seconds, it has the longest single-clip ceiling of the three.
veo-3.1 is Google's flagship video model, also with native audio, and it's the only one of the three that goes up to 4K. Durations are fixed rather than a free range — 4, 6, or 8 seconds — which fits a punchy hero clip or ad cutdown better than a longer narrative sequence. The -fast variant trades some quality for well under half the price ($0.25/s vs. $0.57/s), which is the more practical default unless you specifically need the standard tier's extra fidelity.
All three go through the same async task endpoint. Here's a minimal example using wan2.7-video:
curl -s -X POST https://api.hiapi.ai/v1/tasks \
-H "Authorization: Bearer $HIAPI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "wan2.7-video/text-to-video",
"input": {
"prompt": "a paper airplane gliding through a sunlit office, slow motion",
"ratio": "16:9",
"duration": 5
}
}'
The response returns a taskId. Poll GET /v1/tasks/{taskId} until status flips to success, then grab the video from output[0].url — the same task-submit-then-poll pattern works for seedance-2.5 and veo-3.1, just with each model's own input fields. Full request/response shapes are in the quickstart docs.
-fast unless you specifically need the standard tier.Is LTX-2.5 available through HiAPI? No. As of this writing it isn't in HiAPI's model catalog. If you need an LTX-2.5-specific feature, you'd need to go directly to fal.ai for it.
Which HiAPI video model is closest to LTX-2.5's audio-to-video mode? seedance-2.5 and veo-3.1 both generate native audio alongside video, which is the closest match in capability, though neither works by syncing to a pre-existing audio track the way LTX-2.5's audio-to-video mode does.
Do any of these three support image-to-video? Yes — wan2.7-video and seedance-2.5 both support image-to-video in addition to text-to-video. veo-3.1's image-to-video support should be confirmed against the current model page, since availability varies by endpoint.
Why is veo-3.1 so much more expensive per second than the other two? It's Google's flagship model and the only one of the three offering 4K output — the price reflects that positioning, not a flat "audio" or "quality" tax across the board.
All three models — wan2.7-video, seedance-2.5, and veo-3.1 — are live on HiAPI behind a single API key, so you can prototype with the cheapest option and swap models by changing one string once you know which trade-off fits your project. Grab a key and check exact current specs on the wan2.7-video model page, or see how Seedance stacks up against its predecessor in our Seedance 2.5 vs 2.0 comparison.
