Upscale after picture lock

Upscaling a shot you are going to regenerate is paying twice: once for pixels you will throw away, and again for the replacement. Generated pipelines make that easy to do because another take is always one click away, and an upscale looks like "finishing" while the cut is still a mood. It is not finishing. It is a delivery pass on files that are allowed to be final.

The before/after pair on this page is the same Extra studio example: a 480p source next to the upscaled keeper. That pair is the point of the tool — upscale video to 4K on the shot you are keeping. It is not a preview filter for a prompt you have not signed.

Picture lock when shots can still regenerate is the finishing contract: named files, a change-list, which picture changes reopen grade and mix. This post is the credit rule that contract implies. Lock the cut. Then upscale the keepers.

Upscale is a delivery meter, not an exploration meter

Exploration wants cheap, small, reversible files. A 480p or 720p take is enough to decide whether the motion, the identity, and the performance are right. Delivery wants the resolution the client, the platform, or the grade expects. Those are different jobs on different days.

Teams invert them because upscale is visible. A sharper frame feels like progress. A locked 720p timeline feels unfinished. So the sequence becomes: generate, upscale everything in the bin, pick a winner, regenerate the hero, upscale again. The discarded upscales are pure waste. The second upscale is the only one that belonged.

Treat resolution as a property of the export, not of the take. The take is a decision. The export is a file. Video editing is full of passes that only make sense after the decision: captions on the VO you kept, a dub of the master, an upscale of the keepers. Running those passes on candidates is how a credit balance disappears with nothing to deliver.

What lock means for this pass

You do not need a signed change-list to delay an upscale. You need a weaker, earlier lock: these shot IDs will not be regenerated for taste.

• The still was locked before motion. Identity is not in play.
• The take survived review. Nobody is about to "try it warmer."
• Duration is the duration. You are not about to extend, trim, or hold.
• Native audio in the shot is the audio you will ship, or you have already decided the shot is MOS and the mix will replace it.

If any row is still open, do not upscale. Another sample is a new shot. An upscale of the old one does not transfer. Colour-matching a regen "to the upscale we liked" is how you spend an afternoon restoring pixels you already paid for.

The picture-lock post is the client-facing version of this, with grade and mix in the matrix. Internally, the same matrix has a cheaper row: upscale reopens if the pixels change. They change on every regen.

Keepers only, at the end of the chain

Order of operations for a generated sequence:

1. Still → motion, at working resolution.
2. Cut. Approve. Write the shot list that is actually in the timeline.
3. Mix and grade against that timeline, or decide those departments are not in the job.
4. Upscale the shots that are still in the cut.
5. Burn captions, overlays, exports for each ratio.

Do not upscale before captions if the caption pass will re-render the file anyway — check whether your caption job writes a new master. Do not upscale a hook candidate you have not picked. Do not upscale a 50-variant batch so the client can "see them properly." Show working-res. Upscale the winner.

If the brief is a platform that encodes aggressively, upscale is still not a reason to do it twice. One pass on the locked file is the whole budget.

Hosted upscalers will not expose tile size or a seam-fix toggle. If a lattice shows up, it was in the source or in a previous local tiled pass. That is a different diagnosis. This post is only about when to spend the pass, not how to debug a grid.

What not to "sharpen" instead of locking

• A bad hook. Upscale does not hold the first second. Regen or recut.
• A soft identity. If the face is almost right, go back to the still. More pixels of the wrong person is still the wrong person.
• A temp VO. You will reprint the soundtrack and maybe the picture.
• Every angle in the bin. The bin is not the deliverable.

The Extra tools exist to finish a file, not to negotiate with it. Background removal is the same shape of mistake when you key a shot you are about to throw away. Upscale is just the more common one, because every AI clip looks a little small until someone pays to make it large.

If a client asks to "see it in 4K before we lock," send a single hero still or a five-second upscale of the leading candidate, labelled as a resolution check, not as the cut. Do not upscale the sequence. A resolution check answers "does this take hold on a big screen." It does not reopen taste. If they then want a different take, you are still in exploration, and the five seconds were cheap compared with a 4K bin.

FAQ

Should I generate at 4K so I can skip the upscale?

Only if the model, the duration, and the credit math actually support it, and you have already locked the still. Generating exploration shots at delivery resolution is the same waste with a different label. Most sequences should decide at working res and upscale keepers.

Does picture lock mean I can never upscale a replacement?

It means a replacement is a new shot. Upscale the new shot after it is accepted, not "to match" the old upscale. Quote the departments that reopen if a client asks for the replacement after mix.

Can I upscale a still instead of the video?

Yes, when the still is the thing you will animate, or when the deliverable is a frame. Do not upscale a still, animate it, then upscale the video of the same image as a ritual. Pick the pass that matches the file you ship.

Is there a reason to upscale before client review?

Only if the review is a projection, a grade session, or a spec that rejects the working resolution. A Slack H.264 is not that. Sending an upscaled temp trains the client to comment on sharpness instead of on the shot.