NodeValid

Published video-model weights, licences and hardware notes

Context frame, and the size it forces

A context frame is the image an image-to-video model is given to start from. Where a release states that the frame must be the same size as the output, the resolution becomes a requirement on the input as well. As of 2026-09-22.

Where the resolution constraint landsFor an image-conditioned release the published size is a pipeline requirement, because source material has to be prepared at that size before anything runs.PrepareSource image at thestated resolutionmatched sizeConditionThe model starts fromthat framegenerated framesOutputFrames at the same sizeas the inputA constraint on what you feed inOne release in this register states the context frame and the output at the same resolution.
Fig. 1 One card here states the input and output sizes as the same value, by construction.
What a matched-size requirement adds to a pipeline. Recorded 2026-09-22.
StepAdded work
Sourcing imagesThey have to be at the target size
CroppingComposition decisions move upstream
Upscaling firstSometimes, with quality consequences
Batch preparationOne more processing stage

Inclusion rule. Work a matched-size context frame adds before generation. Steps that would exist anyway are excluded. Order. From the first pipeline step affected to the last.

1The constraint moves decisions upstream

If the output size is fixed by the input, then framing and composition are decided when the source image is prepared rather than at generation time. For a production that is a change in where the creative work happens.

Text-conditioned releases have no equivalent requirement, which makes this a different kind of resolution value from most of the column.

2Trained to generate is narrower than supports

One card in this register describes itself as trained to generate at a resolution, which describes the training rather than what the code accepts. Other sizes may run and were not what the weights learned.

The register keeps the wording rather than normalising it, because the distinction changes what a reader should expect from output at a different size.

3Where this constraint appears here

One card in the register states that the model was trained to generate at a resolution given a context frame of the same size, which turns the published output size into a requirement on the input.

No other entry carries that shape, so it is the one row where the resolution column describes a pipeline stage before generation rather than the artefact after it.

A reading note, not an entry: no vendor value appears on this page. Where the register records this term for a particular release, it is on the resolution column and on that family's own page. Nearby terms: Latent space, Upscaling.