NodeValid

Published video-model weights, licences and hardware notes

A frame count against a full output row

Step-Video-T2V states videos up to 204 frames and publishes no frame rate. Allegro states 88 frames, six seconds at 15 FPS. One length cannot be converted into seconds and the other needs no conversion. As of 2026-09-22.

Step-Video-T2V and Allegro on the same columnFrames, seconds and a rate are three views of one value, and a release that publishes only one of them leaves the reader unable to recover the others.Step-Video-T2VAllegroLength, as each vendor statesitUp to 204 frames88 frames, 6 seconds at 15 FPSThe detail published beside itFrames, with no frame rate attachedFrames, seconds and frame rate all statedResolution, from the same pages768px768px and 544px992px720 x 1280Convertible, or not
Fig. 1 A count without a rate, beside a count with both.
Step-Video-T2V and Allegro on the length column, read from github.com/stepfun-ai and huggingface.co/rhymes-ai. Recorded 2026-09-22.
PointStep-Video-T2VAllegro
Length, as each vendor states itUp to 204 frames88 frames, 6 seconds at 15 FPS
The detail published beside itFrames, with no frame rate attachedFrames, seconds and frame rate all stated
Resolution, from the same pages768px768px and 544px992px720 x 1280
Hardware, from the same pages78.55 GB peak at 768px768px204f9.3G BF16 with cpu_offload

Inclusion rule. Two families whose published values in this column lead to opposite decisions. A pair that happens to agree is not listed, because agreement is already visible on the column summary. Order. The column itself first, then its detail, then the columns that qualify it.

1The missing rate is not a small gap

204 frames is somewhere between seven and thirteen seconds depending on a number nobody wrote down. For anyone planning a shot list that is the difference between one clip and two.

The register does not assume 24 FPS, because two releases here use 16 and 15. No default is safe when the published rates range that widely.

2Frames are the unit one repository thinks in

Step-Video's memory table labels its configurations 204f and 136f, so frames are consistent across the page. The omission is coherent rather than careless.

Allegro's card publishes its output as one row, with the memory figures for the same run beside it. Both are internally consistent documents with different vocabularies.

3Fifteen frames a second is itself worth reading

It is how 88 frames becomes six seconds, and it is lower than most. Output at that rate may need interpolation before it sits beside conventionally shot material.

A reader can only know that because the rate is stated. On the other row the same question cannot be asked.

4Sources

Read from github.com/stepfun-ai and huggingface.co/rhymes-ai on 2026-09-22. Full entries: Step-Video-T2V and Allegro. What this column is quoting is on length. Nearby pairs: A table or no ceiling, Priced, or unstated.