Eight frames a second against thirty
Two model cards publish a frame rate: eight frames a second on one, thirty on the other. The spread is the widest proportional gap in any column on these sheets, and no ad surface this register reads states a rate a file has to meet. As of 2026-09-22.
| What the pages state | CogVideoX | LTX-Video |
|---|---|---|
| Published rate | 8 frames per second | 30 FPS |
| How motion reads | Stepped | Continuous |
| What the rate is used for on the page | Explaining a six second length | Converting a frame ceiling |
Inclusion rule. Frame rate statements from each model card, with the figure each card uses the rate to derive. Rates reachable by interpolating afterwards are a post-process and not recorded. Order. The rate first, then its consequence, then its role on the page.
1Nothing rejects either file, which is the problem
Both ad specification pages this register reads are silent on frame rate. One publishes durations and billing behaviour, the other shapes, pixel floors, a size cap and a bitrate floor. Neither states a rate.
So this is not a compliance column. A piece at eight frames a second uploads, runs and is charged for, and the consequence lands on how it reads beside the filmed work in the same break. That is a judgement the register records the input to.
2On both pages the rate is doing a second job
On the low row the rate explains the length: forty-nine frames at eight a second is the six seconds the card prints. On the high row the rate is the divisor that would turn a frame ceiling into seconds, which is the only route to a length the card never states.
Neither card publishes the rate as a feature. It surfaces because the frame count is the quantity being budgeted, which is why this column is filled by model cards and empty on every rate card in the register.
3Raising a rate afterwards is a different deliverable
Interpolation exists and neither card publishes it. It adds a processing stage, introduces its own artefacts on fast motion, and moves the output away from the thing this register recorded.
The sheets stay on the published figures for that reason. What a pipeline can achieve with extra steps is unbounded and unverifiable, and these two cells hold one number each.
4The rest of the column behind these two
| Generator | Frame rate |
|---|---|
| Amazon Nova Reel | 24 fps |
| CogVideoX | 8 fps |
| Google Veo | Not stated |
| Grok Imagine | Not stated |
| Hedra | Not stated |
| HeyGen | Not stated |
| Kling AI | Not stated |
| LTX-Video | 30 fps |
| MiniMax | Not stated |
| Pika | Not stated |
| PixVerse | Not stated |
| Runway | Not stated |
| SceneMixer | Not stated |
| Vidu | Not stated |
| Wan 3.0 | Not stated |
Inclusion rule. Every generator this register carries, on the one column this pair disagrees about. Rows that state nothing keep their place and are marked hollow. Order. Alphabetical by generator name.
Why these two. A factor of nearly four in the property that decides whether motion reads as video at all, published by two documents of the same type, in the column almost nothing else in the register fills.
- Frame rate8 frames per secondfrom the model card table
- What the released model produces30 FPS videos at a 1216x704 resolution
5Sources
Read from the CogVideoX-5B model card and the Lightricks LTX-Video model card, both checked 2026-09-22. The column is described on frame rate, and each entry has its own cell page at CogVideoX and LTX-Video, and the rule that keeps surfaces and generators apart is on method. Nearby pairs: LTX-Video and PixVerse, Hedra and HeyGen on length.