CampaignReel

Ad placement specs against what generators output

CogVideoX and HeyGen, three hundred times apart

Six seconds at forty-nine frames against thirty minutes per scene with fifty scenes per video. The ratio is three hundred to one, which is the widest gap anywhere in this register, and it marks the boundary between two product categories rather than two grades. As of 2026-09-22.

Where each length figure comes fromOne page derives its length from a frame budget and the other from how long a speech file runs. The ratio between them says nothing about capability.A frame budgetA speech fileWhat sets it49 frames at 8 a secondHow long the audio runsPublished figure6 seconds30 minutes per sceneA sixty second slotTen runs, nine joinsOne run, no joinsRanking these would claim one is 300 times better at one job
Fig. 1 Two products answering different briefs, sharing a unit and nothing else.
What each page puts in writingFilled where the page states something in that column, hollow where it does not address it.What each page puts in writingCogVideoXHeyGenLongest single runLongest single run — CogVideoX: 6 seconds, fixedLongest single run — HeyGen: 30 minutes per sceneAssembly inside the productAssembly inside the product — CogVideoX: None statedAssembly inside the product — HeyGen: 50 scenes per videoWhat sets the lengthWhat sets the length — CogVideoX: A 49 frame budget at 8 fpsWhat sets the length — HeyGen: How long the supplied audio runs
Fig. 2 Filled where the page states something in that column, hollow where it does not address it.
The two length statements, side by side. Recorded 2026-09-22.
What the pages stateCogVideoXHeyGen
Longest single run6 seconds, fixed30 minutes per scene
Assembly inside the productNone stated50 scenes per video
What sets the lengthA 49 frame budget at 8 fpsHow long the supplied audio runs

Inclusion rule. The length figures each page states, and what each page says decides them. No figure here is converted or combined. Order. The single-run figure first, then what surrounds it.

1The gap is a category boundary, not a ranking

One figure comes out of a frame count: forty-nine frames at eight a second, rounded to six seconds. The other comes out of a speech file: the picture runs for as long as the audio does, and thirty minutes is where file handling stops.

Ranking them would say that one product is three hundred times better at the same job, which is not what either page claims. The honest reading is that a prompt-driven scene and an audio-driven performance are answers to different briefs that happen to share a unit.

2Against a sixty second slot the two behave completely differently

Ten runs and nine joins against one run and no joins. On the short row every seam has to hold a face, a key light and a background across separate inferences that shared nothing. On the long row the question of seams does not arise.

Impression billing sharpens that. A non-skippable slot delivers and charges for its whole length, so a visible join is a second the advertiser paid to show. The join count is the number this pair actually decides.

3What each page does not say

Neither publishes anything about how many attempts a usable take costs, and neither publishes an output file size. The short row states its frame rate at eight a second, which is a stronger constraint on whether the output belongs in a paid placement than the length ever was.

The long row publishes output dimensions as a range up to 4,096 pixels a side and says nothing about the size of the resulting file. Against a surface that caps an upload at 500 MB, that is the figure a delivery check would want and neither page provides.

4The rest of the column behind these two

Duration across the registerCogVideoX is marked; HeyGen sits at the top of it, at thirty minutes per scene in the same column. Recorded 2026-09-22.Duration across the registerDurationAmazon Nova Reel — Duration: 6 s increments, up to two minutesAmazon Nova ReelCogVideoX — Duration: 6 s, at 49 framesCogVideoXGoogle Veo — Duration: 4, 6 or 8 sGoogle VeoGrok Imagine — Duration: 1 to 15 sGrok ImagineHedra — Duration: 10 minutesHedraHeyGen — Duration: 30 minutes per sceneHeyGenKling AI — Duration: Not statedKling AILTX-Video — Duration: Frames, never secondsLTX-VideoMiniMax — Duration: 4 to 15 sMiniMaxPika — Duration: 5 s and 10 s, 20-25 s on PikaframesPikaPixVerse — Duration: 1 to 15 s on version 6PixVerseRunway — Duration: 5 to 15 s on Gen-4.5RunwaySceneMixer — Duration: Thirty seconds on the Wan tiers, fifteen on H3SceneMixerVidu — Duration: Not statedViduWan 3.0 — Duration: 2 to 30 sWan 3.0
Fig. 3 CogVideoX is marked; HeyGen sits at the top of it, at thirty minutes per scene in the same column. Recorded 2026-09-22.
Every generator in the register on the duration column, with CogVideoX marked. Recorded 2026-09-22.
GeneratorDuration
Amazon Nova Reel6 s increments, up to two minutes
CogVideoX6 s, at 49 frames
Google Veo4, 6 or 8 s
Grok Imagine1 to 15 s
Hedra10 minutes
HeyGen30 minutes per scene
Kling AINot stated
LTX-VideoFrames, never seconds
MiniMax4 to 15 s
Pika5 s and 10 s, 20-25 s on Pikaframes
PixVerse1 to 15 s on version 6
Runway5 to 15 s on Gen-4.5
SceneMixerThirty seconds on the Wan tiers, fifteen on H3
ViduNot stated
Wan 3.02 to 30 s

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. Three hundred to one is not a difference of degree. One page publishes a frame budget from a research release, the other publishes a performance bound by a speech file, and putting them in one column is only useful if the reason is printed beside the numbers.

5Sources

Read from the CogVideoX-5B model card and the HeyGen usage limits page, both checked 2026-09-22. The column is described on duration, and each entry has its own cell page at CogVideoX and HeyGen, and the rule that keeps surfaces and generators apart is on method. Nearby pairs: Wan 3.0 and Veo, Nova Reel and Runway.