AI video models compared: 11 options on Cicadas.ai

Cicadas.ai runs ten AI video models end to end, plus a page for Sora 2 explaining what to use now that it is gone. This comparison is current as of September 27, 2026: input modes, durations, resolutions, audio handling and one example credit quote per model, plus what a small set of recorded outputs actually looked and sounded like. It is a reference for choosing a starting point, not a benchmark or a quality ranking.

Models compared

| Model | Page | Input modes | Durations | Resolutions | Audio | Example quote | |---|---|---|---|---|---|---| | Kling 3.0 Pro | /kling-3 | Text, one image | 5 s, 10 s | Fixed — recorded at 1920×1080 | Optional, +50% credits | 5 s silent = 140 credits | | Veo 3.1 Fast | /veo-3-1 | Text, one image, first+last frame | 4 s, 6 s, 8 s | 720p, 1080p | Optional | 4 s 720p with audio = 150 credits | | Seedance 2.5 | /seedance-2-5 | Text, one image | 4, 5, 10, 15, 30 s | 480p, 720p | Optional | 5 s 720p = 592 credits | | Seedance 2.0 | /seedance-2-0 | Text, one image | 4, 5, 10, 15 s | 480p, 720p | Optional | 4 s 720p = 303 credits | | MiniMax H3 Max | /minimax-h3 · /hailuo-ai | Text, one image | 5, 10, 15 s | 480P, 768P | Always on | 5 s 768P = 100 credits | | Grok Imagine Video 1.5 | /grok-imagine-video | Text, one image | 5 s, 10 s | 480p, 720p | Always on | 5 s 720p = 175 credits | | Wan 2.7 | /wan-2-7 | Text, one image | 5 s, 10 s | 720p, 1080p | Native track, no toggle | 5 s 720p = 125 credits | | Wan 3.0 Prime | /wan-3 | Text, one image | 5 s, 10 s | 720p, 1080p | Optional | 5 s 720p = 175 credits | | Gemini Omni Flash 1.1 | /gemini-omni-flash | Text, one image | 4, 6, 8 s | 720p, 1080p | Always on | 4 s 720p = 100 credits | | LTX-2.5 Fast | /ltx-2-5 | Text, one image | 6, 8, 10 s | 720p, 1080p | Optional | 6 s 720p = 135 credits | | Sora 2 | /sora-2-alternative | — | — | — | — | Not available since Sept 24, 2026 |

All ten active models take a text prompt or a single reference image; none of them take multiple images, a video, or an audio clip as input yet. Every quote above is for one duration/resolution pair, not the model's only option — Seedance 2.5 also quotes 221 credits for 4 s at 480p, and Seedance 2.0 quotes 170 credits for 5 s at 480p. Credits are estimated from fal's list prices; the live quote shown on each model page for your exact settings is the number that charges, not the table above.

Two entries need a footnote. Kling 3.0 Pro has no resolution selector — every output we recorded came back at 1920×1080 regardless of which duration we chose. Wan 2.7 has no audio switch at all: it returns a native audio track on every clip, which is different from Gemini Omni Flash 1.1, Grok Imagine Video 1.5 and MiniMax H3 Max, where audio is also always on but generated as a layer rather than described as "native."

Looking for Sora 2?

OpenAI closed the Sora 2 app on April 26, 2026, and the Sora 2 API on September 24, 2026. It isn't one of the models above, and there's no workaround on this site or elsewhere that restores it. If you landed here looking for it, see what to use instead.

One recorded clip per model, checked automatically

On September 27, 2026 we generated one example per model and ran each output's audio track through an automated pass — google/gemini-3.5-flash-lite describing what it heard, not a human listening review. Treat what follows as a directional read on one clip per model, not a certified transcript and not a benchmark: one sample tells you what happened once, not what happens every time.

The batch: a Seedance 2.5 wok over open flame at a night market (low rumble), a MiniMax H3 Max convertible on a coastal road (model-added, piano-like music), a MiniMax H3 Max hummingbird macro shot (ambient electronic music), a Grok Imagine Video 1.5 astronaut crossing red Mars dunes (calm music, though the prompt asked for wind and footsteps), a Wan 2.7 koi pond under lotus leaves (quiet water drips), a Wan 3.0 Prime shot of surf on volcanic rock (a loud roar), a Gemini Omni Flash 1.1 cat in a rainy bookshop (soft ambient music, though the prompt asked for rain and thunder), an LTX-2.5 Fast shot of paper boats in a wet gutter (wind-like rustle), and a Seedance 2.5 480p clay fox diorama (gentle music). Two older recordings are also on file: a Kling 3.0 Pro moonrise from a text prompt (silent) and a Veo 3.1 Fast desert scene from a text prompt (with audio).

Four of those, picked to show both sides of the pattern:

The two always-on clips above added a music bed instead of the specific ambience the prompt asked for; the two audio-optional clips, generated with audio switched on deliberately, came back closer to a literal read of the scene. Nine clips across nine models is not proof this holds every time — it's a reason to generate, listen, and adjust rather than assume the audio will match the brief.

Which to try first

These are starting points based on the numbers above, not a ranking:

• Long single takes — Seedance 2.5 goes up to 30 seconds without a cut, longer than any other model here.
• Lowest cost per second — MiniMax H3 Max quotes 100 credits for 5 s at 768P, 20 credits per second, the lowest per-second figure in the table.
• Audio-optional drafts — LTX-2.5 Fast (6 s 720p = 135 credits) or Veo 3.1 Fast (4 s 720p with audio = 150 credits) let you compare a silent and a voiced take of the same shot without switching models.
• Cinematic motion — Kling 3.0 Pro is built around camera moves and physical action; see the prompt guide for how to write for it.

Check the live quote and current availability on each model's own page before you generate — credits, durations and resolution options can change.

For the shot-writing side of this, including what these same recordings taught us about writing for audio, see the AI video prompt guide. For the recorded-output methodology behind both posts, see /methodology.