A small puppet theater finds a compressed recording from a community performance. The main character has a crescent eyebrow and exactly one painted dot on its cheek. The footage is noisy and blocky. Better video might help visitors follow the expressions, but new painted features would falsify the puppet’s design. A restoration must preserve the object represented by the footage, not improve it into a different character.
Topaz describes Iris as an enhancement model for low- and medium-quality video, particularly face recovery, noise reduction and compression artifacts. Because Iris can regenerate fine detail, the vendor’s guidance encourages restraint with additional detail enhancement. The steps here are a restoration plan based on official references, not a claim that real theater footage was processed or that output quality was measured.
Identify the puppet and the authoritative visual record
Start by gathering the earliest surviving video, a photograph of the puppet, costume sketches, and any production notes about its paint. Record the recording format, frame rate, source dimensions, compression history and known edits. A later social-media repost is not equivalent to the first archived transfer. Work from the best lawful source available, with the appropriate permission from the company or rights holders.
Build a reference sheet showing the crescent eyebrow, single cheek dot, pale jawline, and seam of the fabric hood. Those are factual attributes to protect. If the original recording cannot show a feature clearly, label it uncertain rather than authorizing the model to decide. Restoration should not silently substitute a new face for missing historical evidence.
Match the enhancement model to the source
Examine a short representative segment with noise, compression blocks, motion and a clear view of the puppet’s face. In Topaz Video, consider Iris according to the current model list and the source’s progressive or interlaced nature. The official documentation characterizes Iris as useful for compression and face detail, but the available controls and hardware requirements may vary by version.
Set the target to a modest improvement that supports viewing at the intended display size. Do not begin with maximal sharpening or several enhancement passes. A large output frame does not automatically contain more trustworthy facial data. Save settings and exact source in the review log so a conservator can compare a candidate with the untouched recording.
Protect motion and frame continuity
Inspect a segment where the puppet turns its head. Pay attention to paint shapes across adjacent frames; an eyebrow that flickers or changes thickness with every head rotation creates new animation that was not present onstage. Watch the contrast around string attachments as well: a restored face that makes puppet strings disappear can subtly misrepresent the theater technique.
A side-by-side synchronized playback is more useful than two still images selected from different moments. Match timecodes, crop and viewing scale. Examine the original and processed versions at normal speed and frame by frame. If paint changes continuously through a shot, lower processing strength or retain the original instead of treating the invented detail as an aesthetic benefit.
Treat face recovery as enhancement, not discovery
A hypothetical rejected result may add a second cheek mark because texture synthesis interprets compression artifacts as freckles. That does not establish what the historical puppet looked like. Compare the generated detail with stage photographs taken around the same period. If the only evidence is a blurry video, preserve the uncertainty and avoid annotating the output as a reconstruction of fact.
Describe edits in ordinary language: ‘noise reduced; some features reconstructed by software.’ Never say a newly visible mark was ‘recovered’ without independent corroboration. If a showcase requires faithful paint texture, a less aggressive pass with visible noise is often the responsible alternative. An archival edition and a separate creative tribute may have different standards.
Check costumes, highlights, and background textures
Look beyond the face. Velvet curtains can become unnaturally smooth; stage lights may grow halos; painted wood seams may fragment. Review the puppet’s hand shape and the dark stage background as motion continues. Face-enhancement algorithms can modify regions outside the center of attention, and spectators may notice those changes before the editor does.
Compare performance details against a contact sheet from the raw clip. The objective is to maintain lighting cues and prop placement while making the image reasonably watchable. If a hand vanishes into a curtain or a string becomes a false edge, note the specific timecode. The review is about truthful continuity, not whether one freeze frame looks dramatically cleaner.
Preserve masters and document a reversible workflow
Keep the original encoded file, any higher-quality transfer, and the working enhancement in separate directories. Record settings, application version, intermediate exports and any creative color changes. For public archives, maintain a textual description of the treatment so visitors can distinguish the recording from a modern viewing copy. Do not overwrite the first transfer.
Use consistent filenames or repository identifiers and a checksum for the master. If the restoration is presented with captions, include the provenance note close to the player. Keep performance audio untouched by image enhancement. Separate soundtrack cleanup is another editorial decision that requires its own source comparison and artifact checks.
Approve the least misleading viewing copy
Show the result to an archive curator and, when practical, someone familiar with the puppet design. Ask them to inspect the eyebrow, cheek dot, costume seams and moving strings. If the viewers disagree about a faint detail, do not force consensus from a sharpened image. Refer to the physical prop or production photos when they exist.
Choose a mild enhancement for everyday viewing and offer an original-footage option for researchers. Avoid a release that presents regenerated paint as newly uncovered proof. The quality bar is not the highest apparent resolution; it is a watchable copy that remains honest about what the archive actually records.
How We Research
This article was prepared on November 4, 2026 from these named official vendor references: Topaz Video: Enhancement model guide; Topaz Video: Filters reference. We reviewed documented controls, availability caveats and stated workflow boundaries, and designed an independent hypothetical editorial check for the particular task described. We did not access a vendor account, upload live media or private records, generate candidate results, measure output quality, or claim that any illustrated failure occurred in a real tool. The WebP graphics are original teaching diagrams, not product screenshots. Plan restrictions, labels and capabilities can change. A responsible editor must still verify the underlying source, permissions, source data, accessibility, final exported artifact and sign-offs before public release.
Frequently Asked Questions
Topaz describes Iris as face-oriented but also suitable for noise, compression reduction and broader video enhancement. Its behavior on a puppet face still needs careful review.
Not reliably. AI-generated detail may be plausible rather than historically accurate. Verify distinctive paint marks against other records.
No. Choose the viewing target first and inspect whether upscaling adds artifacts or invented details. Sometimes a restrained 1x enhancement is better.
Play a synchronized before-and-after view, then examine the same eyebrow and cheek dot in adjacent frames at identical timecodes.
Retain an untouched master, source metadata, original photographs or designs used for comparison, the processing settings, and a clearly identified viewing derivative.