Upscale video for After Effects without resampling the comp

After Effects produces resolution problems from two directions, and they need different answers. The first is the composition itself: its Resolution menu lets you render and preview at a fraction of the frame size so the timeline stays responsive. That setting is not a preview-only convenience — a comp left at half resolution renders at half resolution. The second is a layer that was never big enough: you drop a small source into a large comp, drag the Scale property up until it fills the frame, and discover the picture has gone soft because you are magnifying a raster that was already the wrong size.

The Detail-preserving Upscale effect is the tool most people reach for here, and it is a good one: it pushes a layer to a much larger scale than a plain transform while keeping edges and fine texture from collapsing into mush. It is still working from the pixels the layer already has. The same is true of the Continuously Rasterize switch, which changes the resolution at which a pre-comp or a vector layer is rasterised inside its parent — that removes compounded resampling in nested comps, but it does nothing for video footage whose source is genuinely small.

Order matters inside a single comp too. Scale a layer up first and every effect after it — blur, glow, grain, keying — operates on the magnified, softened image. Apply those effects first and scale afterwards and they operate on sharper pixels that are then magnified. Neither order inserts detail, but the second keeps effects tuned at the size they were designed for, and working at a higher project bit depth keeps a heavy scale from banding as it lifts the contrast of the interpolation.

It is tempting to render the comp at half resolution because the render is long, then hand the file to an upscaler and let a model rebuild the missing detail. That trades render time for credits and loses the trade. The low-resolution render did not merely store less detail — it averaged neighbouring pixels together, deterministically, and threw the difference away. A model asked to restore it is not recovering those pixels; it is inventing plausible replacements for information that was already destroyed on purpose.

The comparison is not close, either. Raising the composition's resolution and rendering again costs machine time you already own. Rendering small and paying per submitted second costs money, and the best possible outcome is a frame that resembles the full-resolution render without being it. The only case where the pass is not second best is when the full-resolution render no longer exists to be made.

The same reasoning applies to the Output Module. Typing a larger size into the output's resize setting does not make the comp bigger; it resamples the finished frame on the way out, which is exactly the magnified-and-soft result you were trying to avoid. If the deliverable needs a bigger frame, change the composition's frame size, re-scale the layers that need it, and render that — the output module is the last place to solve a resolution problem, not the first.

The pass earns its place when the footage itself is small: phone captures, archival standard definition, downloaded clips, screen or drone recordings that were never high definition, or an older camera whose material now has to sit in a modern comp. There, nothing was thrown away that could have been kept — the detail was never recorded — and a model predicting plausible texture is doing something the app cannot, because the app has no information to work from either.

Run that pass on the raw shot, before it enters the composition. Not the composite, not the graded version, not a pre-render with motion blur baked in. A model fed a comp sees mixed sources at different scales, a look that was applied for taste, and effects it will treat as real texture — it has no way to know which parts of the frame are picture and which are treatment. Run it on the raw shot and the comp's blurs, keys, glows, grain and warps all act on a larger raster instead of magnifying a small one.

Judge the result in the comp, not in a viewer. Some effects depend on fine texture in ways a still frame will not reveal: a key pulls differently when predicted detail changes the edge, a tracker locks differently when grain has been smoothed, and a warp behaves differently on a sharper plate. Test one shot at the comp's full size, with the effects that matter to it switched on, before you commit the sequence to the same settings.

Export the shot at its native frame size, from the original media, once, with an intermediate codec — ProRes 422 or a comparable lightly compressed option exists to survive one more processing step without a visible generation loss. Do not export at the size you are aiming for: the export would fill the extra pixels with its own interpolation, and the model would then treat that interpolation as real image data. Native in, native out is the rule for anything you feed a model.

Bring the enhanced file back by replacing the footage item in the Project panel rather than rebuilding the layer. Every composition that uses the item updates at once, and the layer keeps its trims, keyframes, effects, masks and expressions. What does not adjust itself is the layer's size: if the pass doubled the linear dimensions, the new footage sits at twice the frame and the layer's Scale needs dividing by the same factor. Any scale keyframes need the same treatment, which is why it is worth writing down the factor before you start.

If the composition itself should now be larger — a 1080p comp that has to deliver a 4K master — change the composition's frame size rather than stacking a second scale on the layer. Nested comps and vector layers follow the new size and re-rasterise at it instead of being magnified, while raster footage gains nothing and should not be stretched to fill. Then re-check everything with a fixed pixel size: masks, blurs, grain, and motion blur, which are all defined in pixels and will read differently against a larger frame. Keep the original footage disabled rather than deleted until the final render is signed off.

Credits follow submitted seconds, rounded up from a five-second minimum, so the unit of cost is the file you upload rather than the length of the comp. A shot that is on screen for two seconds still costs its full duration if you submit the whole clip, and a three-second shot bills at the five-second floor. Trim the segment you actually need before uploading, and remember that one source can run up to 10 minutes, so a long take is split rather than submitted in one piece.

At the Standard model, 1080p bills 540 credits per minute of source footage, which is 32,400 credits an hour — the reference figure for comp work, because most After Effects shots that need the pass are 1080p sources that have to survive a 1080p or 4K comp. Converted at the Creator yearly rate of $0.0093 per credit, a minute is about $5.01 and an hour about $300. Creator's 106,800 credits cover roughly 197 minutes of 1080p Standard footage, or about 53 minutes of Standard 4K.

Read those numbers against how a comp actually breaks down. A piece with forty short problem shots of three seconds each is billed at the five-second floor forty times, so it lands closer to two hundred seconds than to the hundred and twenty seconds that appear on the timeline — still far less than the sequence length, which is why quoting a cost per finished minute is the wrong estimate. Count the shots that need the pass, trim each to the part that needs it, and add up submitted seconds.