An archive-wide “upgrade everything to 4K” project sounds efficient. It is often the wrong procurement brief.
Resolution is an output specification, not proof that an asset is commercially usable. An old product film can be enlarged, denoised, deinterlaced and sharpened while still showing discontinued packaging, an expired claim, a talent license that does not cover new markets, or details that were never captured in the first place. More pixels do not create more historical truth.
On September 23, 2026, Adobe announced that it had completed its acquisition of Topaz Labs. Adobe said Topaz would continue as a standalone brand, that Topaz technology was already available in Firefly and Photoshop, and that more integration into existing creative workflows was planned. The announcement highlighted archival restoration, low-light refinement, mobile-footage enhancement and enterprise-scale modernization of legacy media.
That is a meaningful signal for brand archives. It is not evidence that every weak source file has suddenly become a valuable modern master. The transaction proves Adobe’s direction and the stated product footprint. The business value of any individual clip still depends on its source, content, rights, current product truth and likely reuse.
Enhancement changes technical usability—not historical truth
Topaz’s current documentation does not describe one universal enhancement button. It separates progressive, interlaced and incorrectly transcoded footage, then offers models optimized for different problems: noise, compression artifacts, faces, haloing, older analog sources, high-quality footage and computer-generated material. It also includes a control that mixes original detail back into an enhanced result.
The product’s own documentation describes important tradeoffs. Some processing can make background texture look too sharp or “crunchy.” Small faces in low-quality footage may become distorted when a model tries to interpret too few pixels. Detail controls can move toward extreme results quickly. Those are not reasons to reject enhancement. They are reasons to make human review part of the deliverable.
Primary research reaches a similar conclusion. The CVPR paper “Investigating Tradeoffs in Real-World Video Super-Resolution” found that complex real-world degradations can be amplified through temporal propagation, creating a tradeoff between detail synthesis and artifact suppression. Newer work continues to improve generative video restoration, but the underlying decision remains: a plausible reconstruction is not the same thing as recovered evidence.
Brand teams should therefore separate two approvals:
- Technical recovery: Does denoising, deinterlacing, stabilization, deblurring or upscaling make the original material more watchable?
- Commercial usability: Are the product, people, text, claims, rights and brand expression still accurate for today’s destination?
Models can help with the first. People and business records must decide the second.
The ONCE four-route archive model
Before sending files into a model, route every candidate sequence into Restore, Rebuild, Reshoot or Retire.
Restore: the truth is current and the defect is technical
Restore when the product and message remain accurate, the source genuinely contains the detail you need, rights are valid, and the main defects are noise, compression, interlacing, modest blur, instability or insufficient delivery resolution.
Factory processes, founder interviews, historical events and discontinued locations can carry value that cannot be recreated. The target is not a synthetic “new” surface. It is a more usable version that preserves the source’s credible texture.
Restoration should still be destination-specific. A clip that survives a small social placement may fail on a product page, retail display or large presentation screen. Approve the actual encoded outputs, not only a clean preview inside the enhancement tool.
Rebuild: the shot still works but the communication system does not
Some archives have usable product, process or human moments trapped inside obsolete packaging: a long intro, fixed 16:9 framing, dated captions, burned-in prices, an old voiceover or music that cannot travel to a new market.
The right purchase is not necessarily full-film restoration. Extract the accurate shots, then rebuild pacing, layout, captions, voice, sound and aspect-ratio variants around them. A small amount of new capture may connect the old material to a current campaign.
This route complements the Master–Delta–Log approach for multi-channel product content: treat recovered footage as a governed master asset, create only the necessary channel differences, and keep a record of every change and right.
Reshoot: the image carries proof and the underlying fact has changed
Do not ask restoration to conceal any of the following:
- the product’s construction, ports, dimensions, colors or packaging have changed;
- the shot must prove waterproofing, load capacity, accuracy, speed or real operation, but the source never captured it clearly;
- text, interfaces, certifications or safety information cannot be verified;
- likeness, voice, location or music rights do not cover the new territory, duration or paid-media use;
- only a heavily compressed social download survives and the claim-bearing detail is gone.
An AI model can produce a detail that looks reasonable. Advertising proof needs a detail that was actually captured and can be checked. When a shot carries product truth, comparative performance or safety responsibility, model inference cannot replace new evidence.
Retire: the recovery value is lower than the cost of continued use
Not every clip deserves modernization. A discontinued product, unknown rights, a completely superseded brand identity, no traceable master or a one-time low-value use may justify archival retention without commercial reuse.
Retirement should not mean deletion. Preserve the original, source, date and restriction note, but remove the asset from the current advertising pool. That prevents a future operator from mistaking “stored” for “approved.”
Buy representative tests before buying archive-wide processing
A polished hero shot is a poor test set. Even footage from the same production day can fail differently. A representative test should include at least five stress cases:
- close faces and small background faces;
- product edges, fine texture, transparency and reflective surfaces;
- packaging text, screens, certifications and interface elements;
- water, smoke, hair, leaves and fabric—random or high-frequency texture;
- rapid motion, camera movement and long low-motion passages.
Do not approve from still frames alone. Review real-time playback for flicker, detail popping, edge breathing and temporal inconsistency. Compare facial identity, text and product geometry. Check whether denoising turns material surfaces into plastic. Confirm that grain and sharpness remain coherent across shots. Then inspect the original, enhanced master and final channel encode side by side.
The procurement scope should retain the input identifier, model and version, settings or preset, test result, human corrections and final approval. Without that record, the brand cannot reproduce a result or explain why the same shot looks different in a later export.
Price the approved reusable second
Cost per source minute or processed file rewards throughput, including footage that never reaches a campaign. A more decision-useful variable is:
Cost per approved reusable second = (archive audit + rights and fact verification + test renders + AI compute + manual repair + temporal quality control + re-editing and delivery) ÷ seconds approved for the target destinations.
This is a scope model, not an industry benchmark.
A 20-second historical sequence with an irreplaceable person, location or brand memory may justify an expensive recovery. A generic product turntable showing obsolete packaging, built from a poor source, may be cheaper and more valuable to reshoot as a new multi-format master.
Teams can also compare the broader economics with a simple decision expression:
Net recovery value = avoided reshoot cost + time-to-market value + historical distinctiveness − recovery cost − quality-control and rework cost − residual truth and rights risk.
Do not force numbers into unknown fields. Use a representative test and rights review to convert uncertainty into observable work before committing the archive.
What changes in the production brief
The important consequence of Adobe’s Topaz acquisition is not that post-production replaces production. It is that high-end enhancement may become more accessible inside everyday creative workflows, making old media easier to inspect, route and reuse.
The buying unit should evolve with that capability. Do not buy “100 videos converted to 4K.” Buy an archive audit, four-route classification, representative tests, human truth review, temporal quality control, version records and a rebuild plan for assets that are approved for new destinations.
The valuable output is not the number of 4K files. It is the number of seconds that regained a current commercial purpose without rewriting what the product, person or event actually was.
Sources
- Adobe, “Adobe completes acquisition of Topaz Labs, bringing Emmy Award-winning AI enhancement capabilities to Adobe’s creative ecosystem,” September 23, 2026.
- Topaz Video Documentation, “Filters” and “Enhancement”, accessed October 5, 2026.
- Chan et al., “Investigating Tradeoffs in Real-World Video Super-Resolution,” CVPR 2022.
- Yesiltepe et al., “DTG-Restore: Training-Free Diffusion Refinement for Generative Video Super-Resolution,” CVPR 2026.
This article translates public company documentation and primary research into a procurement method. It does not guarantee the output, rights status or advertising compliance of any specific asset.
