AI product video upscaling: review the source before 1080p export

Inspect the source clip before you upscale it. Check product shape, small markings and movement, then compare the new file against that source. A larger frame gives you more pixels to inspect; approval still belongs to the actual product and wording you intend to publish.

We made two fictional product studies through Pickshot: cream headphones and a sage coffee mug. Each starts with an Indigo image, one MiniMax H3 motion request and a SeedVR2 upscale. Below are the real source and final files, with the complete frame preserved. These examples do not establish a real-product fidelity guarantee or an advertising result.

These operator-made studies use the same native upscale operation now connected to the dedicated Video upscale workflow. Choose an owned MP4, select Video upscale or SeedVR2, then review its measured input and current credits before submitting. A 5.167-second source is billed as 6 input seconds: 12 credits with Video upscale or 24 with SeedVR2 at 1080p.

This operation requires eligible 1080p plan access. Free is limited to 720p and cannot run it; current plans do not permit 4K, and paid checkout is not open. The workflow accepts one MP4 up to 10 MiB. Video upscale supports 1–30 seconds; SeedVR2 supports 1–10 seconds at up to 30 fps. Review the delivered picture and audio before publishing.

Headphones: inspect the inner band

The reference is the same cream headphone study shown on our home. We asked for a slow camera arc around a stationary product, with consistent ear cups, headband and hinges. Compare the outline and the inner band in both clips. Small markings remain visible there. They are generated image details, not verified branding or readable product information.

Watch the whole clip, then pause near its beginning, middle and end. Follow both ear cushions, the opening below the headband and the shadow on the plinth. If a detail is wrong in the motion source, record that issue before choosing an upscale. Treat any later render as another artifact to review.

The exact motion prompt used for this request:

A very slow physical camera dolly arcs ten degrees around the cream headphones. Keep the two ear cups, headband, hinge geometry and product placement exactly consistent with the reference; the product remains completely still on its plinth. A narrow sunlight highlight moves gently across the material. Restrained premium product film, one continuous shot, no object rotation or morphing, no people, text, added logos, transitions, cuts or camera shake.

Coffee mug: follow the handle and steam

For the mug we requested a gentle push-in, a steady silhouette and a thin wisp of steam. Compare the handle opening, rim, coffee level, glaze and stone contact. The steam is a generated visual choice; decide whether that movement suits the scene you want.

Play these files at normal speed as well as inspecting paused frames. A convincing poster does not show every later frame. Keep your verified product photograph beside the motion source when reviewing a real SKU.

The exact motion prompt used for this request:

A gentle continuous camera push-in toward the sage ceramic coffee mug. Preserve the exact mug silhouette, handle opening, glaze and coffee level. Only one thin natural wisp of steam drifts upward and dissipates. Morning light remains steady, curtain moves almost imperceptibly. Calm realistic product film, no boiling, splashing, new objects, hands, lettering, morphing, transitions or cuts.

Measure the file you actually received

The motion requests selected 512p, 16:9 and five seconds. Both delivered H3 files measure 896×512. The upscale option selected 1080p; both final files measure 1890×1080. Source and final streams each contain 124 frames at 24fps, or 5.166667 seconds. The table rounds duration to three decimals.

Study and stage	Actual dimensions	Frames	Frame rate	Duration
Headphones · MiniMax H3	896×512	124	24fps	5.167s
Headphones · SeedVR2	1890×1080	124	24fps	5.167s
Coffee mug · MiniMax H3	896×512	124	24fps	5.167s
Coffee mug · SeedVR2	1890×1080	124	24fps	5.167s

Both stages use a 7:4 frame. Check the required delivery format separately before placing the full image into a 16:9 or vertical composition. Our published files have no crop, padding or frame re-encode. The upscaler received the actual H3 file and the 1080p option; we supplied no text prompt for that operation. Its record therefore stores a null prompt and the actual operation parameters.

Approve the product detail, then the final composition

The official SeedVR repository notes that its research prototypes can produce unwanted detail and sometimes oversharpen lightly degraded inputs. That research context supports inspecting the result; it does not identify the exact weights or implementation behind this hosted request. These two clips are a small original production study, not a reproduction of a paper benchmark.

For a real product, compare any printed text and manufactured feature with verified references. Add approved copy as real text during composition. See exact ad copy on AI images and reviewing an AI product video for the handoff checks. In Cut Marketing, review the current render and wording before export; an edit needs review again.

Questions from these files

What should I review before upscaling a product video?

Review the source clip’s full silhouette, printed details, contact with the surface and motion artifacts. Inspect the new upscale against it, then approve only the current render and wording.

Why are these 1080p files 1890×1080?

Both delivered outputs measure 1890×1080. Their 896×512 sources and the final files use a 7:4 frame. The requested option was 1080p; inspect the actual dimensions before preparing a 16:9 handoff.

Can generated markings be used as brand text?

Use verified wording added during composition. Small generated markings in these examples are unverified visual details, not approved brand text. Review the final composition as well as the source.

Source and publication records

Produced on November 3, 2026. Each video’s production record links the source model, exact motion prompt or upscale parameters, reference image and checksums. The metadata-enriched public reference image has different file bytes from the generation input; its compressed image pixels are unchanged. Both hashes are labeled in the record.

Public videos are silent faststart MP4 copies. Each published file retains its own original stage’s encoded H264 picture packets and timing. The SeedVR2 render is a different output from the H3 source; packet identity applies only to publication of each individual file. Original downloads are retained privately. These operator samples do not define a customer price; check current model options and the credit estimate in the studio before generation.

Start from a product brief, or connect a reviewed creative with the creative testing workflow.