How long a clip each family states
FramePack states one minute at 30fps, LTX-Video up to 60 seconds, and SkyReels-V2 1457 frames for 60 seconds. Step-Video-T2V states 204 frames with no frame rate attached. Three families state no length at all. As of 2026-09-22.
| Item | What is published | Detail |
|---|---|---|
| FramePack 13B | 1 minute, 60 seconds | 1800 frames at 30fps, at the 6GB floor |
| LTX-Video 13B | Up to 60 seconds | Stated in the repository |
| SkyReels-V2 | 60 seconds | Given as 1457 frames in the vendor's own table |
| CogVideoX1.5-5B | 5 or 10 seconds | The vendor's own translation of up to 161 frames |
| Pyramid Flow | 10 seconds at 768p | 5 seconds on the 384p checkpoint, both at 24 FPS |
| Allegro | 6 seconds at 15 FPS | 88 frames, with the frame rate stated beside them |
| Cosmos Predict 2.5 | A 5-second clip | At 720P and 16FPS |
| Wan 2.2 TI2V-5B | A five-second 720P clip | The worked example behind a timing figure |
| Stable Video Diffusion | 4sec or less | 25 frames, described as rather short |
| Step-Video-T2V | Up to 204 frames | No frame rate in the requirements table |
| Open-Sora 11B | Under 129 frames | num_frames must be 4k+1 |
| HunyuanVideo | 129 frames | A measurement setting rather than a ceiling |
| LongCat-Video | Not stated | Described as minutes-long with no figure |
| MAGI-1 | Not stated | Frames are an argument with no maximum |
| Mochi 1 | Not stated | 480p described as the current capability |
Inclusion rule. Lengths stated in the vendor's own repository. A length reached by chaining generations is a different claim and is not recorded. Order. Repositories stating a length in seconds first, longest first; then those stating only a frame count, highest first; then families stating none.
1Frames and seconds are not interchangeable and the repositories know it
A frame count is exact; the seconds it becomes depend on the frame rate. The family that publishes frames also supplies the translation, which is the careful way to do it, and this register carries both rather than picking one.
Any table that converted everything to seconds would be doing arithmetic the vendors deliberately did not do.
2A supported length is not a usable length
Support for a duration means the code path runs to the end. Whether the result holds together over that duration is a separate question that nobody publishes an answer to, in either direction.
Treating the longest figure in this table as the best is therefore a mistake the table invites and the rule line tries to head off.
3Silence about length is common and worth recording
Three families state no length at all: one offers a resolution as its present capability instead, one treats frames as an argument with no maximum, and one calls its output minutes-long without attaching a figure. All three are recorded as not stated rather than as unlimited or as short, because a repository that has not addressed it has not committed to anything.
Against the hosted services in this register, which mostly sell tens of seconds per generation, several of the open figures reach a full minute.
- Length and resolutionCogVideoX1.5-5B states 1360x768 and up to 161 frames, described as 5 or 10 secondsper model
- Resolution480p, described as the current capabilitystated as a present limit
- LengthThe repository states support for up to 60 seconds of video on LTXV-13Bfrom the release notes
- HardwareTo generate a 1-minute video at 30fps, 1800 frames, using the 13B model, the minimal required GPU memory is 6GBa floor stated together with the output it buys
- Lengthnum_frames 257 for a 10s video, 377 for 15s, 737 for 30s and 1457 for 60sframes against seconds in the vendor's own table
- Length10-second videos at 768p resolution and 24 FPS, with a 384p checkpoint that generates 5-second video at 24FPSstated per checkpoint with the frame rate
- Resolution and length720 x 1280, 88 frames, 6 seconds at 15 FPSfrom the model info table
- Resolution and lengthProduces 720P video with 16FPS, as a 5-second clipboth stated on the card
- LengthThe generated videos are rather short, at 4sec or lesswritten as a limitation rather than a ceiling
- LengthVideos up to 204 framesin frames, with no frame rate attached
- LengthCandidate values for num_frames should be 4k+1 and less than 129expressed as the constraint the code enforces
4A hosted pipeline on the same models
Instead of local hardware for “How long a clip”, SceneMixer runs hosted models; its guides describe the result:
The shot seconds in the storyboard add up to the segment length, so editing them changes it.
SceneMixer, Fixing One Shot Without Redoing Everything — “Can I change the duration of a segment?” (read 2026-10-13)
Sixty to a hundred and twenty seconds is the working range for short drama, and that is what the outline paces for. A 90-second episode is typically seven segments of about fourteen seconds stacked together.
SceneMixer, Vertical AI Video Generator — “How long should a vertical episode be?” (read 2026-10-13)
5Sources
Repository statements and hosted rates are linked from the model pages and the method page, checked 2026-09-22. Hardware figures are set out on requirements. Nearby: Nobody sells these, Licence vs weights, Which version.