A cel is a sheet of acetate about a tenth of a millimetre thick. The line work goes on the front. The paint goes on the back, so the ink edge stays clean from the camera’s side. Then somebody stacks three or four of those sheets over a painted background, presses the lot flat under a glass platen, and photographs it. That was the job in 1993, and Ninja Scroll was made that way from the first frame to the last.
HIDIVE and Iconic Events are putting it back in roughly 600 North American cinemas on 4, 5 and 7 October — subtitles on the Sunday, the English dub on the Monday, an encore on the Wednesday in whichever format the theatre picks. Sentai produced the restoration and Yoshiaki Kawajiri, who wrote and directed the original, supervised it. The new master had its first outing at the 76th Berlinale back in February.
The distributor describes the work in three steps: a 4K scan of the original 35mm negative at 3840 × 2160, frame-by-frame damage repair, then colour correction to an archival-quality digital master.
The middle step is where these things live or die.
4K is the ceiling on a cel picture
A 35mm negative holds more than 3840 pixels across. People who scan live action argue about 6K and 8K for exactly that reason, and they are not wrong about the film stock.
Cel animation moves the bottleneck off the film and onto the drawing photographed through it — an ink line laid down by a person at a fixed physical size, on a sheet framed to a fixed field size under the camera. Scan far enough to resolve the edge of that line and the useful information runs out. Everything past it is grain and dust.
So 4K on a 1993 cel picture lands about where the artwork stops having more to give.
What the scan picks up that nobody drew
Photograph a stack of acetate and you photograph the acetate too.
Each sheet swallows a little light. A character painted on the third level sits fractionally duller and warmer than one on the first, and the grader on a 1993 production either matched the paint to compensate or let the shift ride. At video resolution the difference disappeared into the noise floor. At 4K it is visible, and it is a record of how the shot was assembled — you can sometimes read the layer order off a frame.
The platen leaves its own evidence. Dust that settled between takes, a hair, the faint bloom where the glass was not quite flat. Cel animation is full of this, and a good scan hands you all of it at once.
The second step goes after something else. All of that is the film keeping a record of having been shot on a table by people, and it should survive. The blog’s piece on the hand-painted backgrounds from the same era, seen at exhibition scale is the closest thing to looking at the source art with your own eyes, and the texture there is the texture a 4K scan starts resolving.
The thing an automated filter cannot tell apart
Dust removal works by comparing a frame against its neighbours. A speck that appears for one frame and vanishes is dirt. Flag it, patch it, move on.
Now run that over anime.
Japanese television and film animation is built on held drawings — the same cel photographed twice or three times before the next one arrives. A temporal filter looking for change across neighbouring frames sees long stretches where nothing moves at all, then a hard jump. And the effects work is worse. Impact flashes, sword glints, blood spray, a single white frame at the moment of a cut: these are deliberate one-frame elements that look, to the software, exactly like a piece of grit on the platen.
Kawajiri’s film is made of those. It is a picture about people being cut open in the dark by things that move faster than the camera can follow, and its grammar depends on marks that exist for a twenty-fourth of a second.
Push noise reduction too hard on top of that and you get the other classic failure, which anyone who has bought a badly remastered disc will recognise on sight. The grain goes. The colour fields flatten into plastic. And the ink line — the one element carrying all the drawing — turns soft and waxy at the edges, because the filter has been averaging it against the paint on either side. A line that a person put down in one confident pass ends up looking like it was traced by a machine. It is the same quality the Imageworks crowd spent years learning to protect when they moved ink lines onto a moving CG character: the mark has to keep the hand in it or the whole style collapses.
Which is why the supervision line matters
Every one of the decisions above is a judgement call rather than a repair, and the person who stood over the original photography is the one who can say which speck was dirt and which was a highlight somebody painted on purpose. That is what the supervision credit buys.
Colour is the same problem in a different coat. This film has been graded for a 1993 theatrical print, then for laserdisc, then for DVD, then for whatever Blu-ray master has been in circulation since — four different targets with four different ideas of what black is. Ninja Scroll is mostly night. Torches, moonlight, forest interiors, a lot of frames where the composition is a shape against something almost unlit. Lift those blacks to “reveal detail” and you have not restored anything, you have thrown away the staging.
Sentai’s description stops at “colour correction”, which tells you nothing about where they pointed it. That is the part I would want to see on a big screen before saying the restoration is good.
Two things I could not confirm and am not going to assert: whether the scan was done wet-gate, and whether the master runs at the original 24 or has been retimed. Neither has been published.
Restoration process, dates, formats and theatre count from the HIDIVE and Iconic Events announcement as carried by Anime News Network and Anime Corner. Production credits for the 1993 original per Madhouse. The description of cel photography and of temporal dust removal is general practice for the period and the technique, not a statement about this specific transfer, which has not been published in detail.