Managing a photographic archive

Keep every frame? Outtakes, variants and the case against weeding

Dark cloud layers frame a band of glowing amber sky.
Golden light through clouds.Photograph by Robert Harper. Robert Harper Photograph Library, Archive.London. From the archive, shown for illustration.

Our inventory counts 177,143 frames and 95,487 unique photographs. The 81,656 in between are the frames an editor did not choose, and they are the last thing an owner should throw away.

The chosen frame tells you what an editor wanted in 1987; the frames around it tell you what was actually in the room.

The inventory of the Robert Harper Photograph Library gives two numbers for almost every folder: frames and shots. Across the whole archive they come to 177,143 frames and 95,487 unique photographs. The difference, 81,656 pieces of film and files, is what a tidy-minded owner would call surplus. Advice on managing a photographer's archive usually says to edit it down to the best. We want to make the opposite case, from our own figures: the gap between those two numbers is where much of an archive's future usefulness lies, and nobody should narrow it without a written rule. This is the case against weeding, and a draft of the rule.

What do archivists mean by weeding and appraisal?

Two words, often confused.

Appraisal, in the Society of American Archivists' dictionary, is "the process of determining whether records and other materials have permanent (archival) value". It can take place at any level, from a whole collection down to a single item, and weighs origin, content, authenticity, physical condition and the cost of preservation. The dictionary is careful to separate it from monetary appraisal. This article is about the first kind only.

Weeding is "the process of identifying and removing unwanted items from a collection". The dictionary's sources describe it as item-level work done during arrangement and description, lighter than reappraising an entire collection, and note that in older practice it "was limited to obvious duplicates and totally irrelevant or very low-level information". Its synonyms are less polite: culling, purging, stripping.

That historical limit is the useful part. Archivists have always accepted that exact duplicates and stray irrelevant items may go. What they have not accepted is weeding by quality. Nobody in a record office removes the dull letters from a correspondence file.

What is the difference between an outtake, a variant and a duplicate?

Photographers use these words loosely. An archive should not.

Term What it is Is it redundant?
Duplicate A copy of the same exposure: a lab duplicate transparency, a second print, an identical file Yes, beyond a small number
Bracket The same composition exposed slightly lighter or darker No. It records how the film responded, and one bracket may have survived better than another
Variant A different arrangement, angle, expression or moment from the same session No
Outtake Any frame not chosen for publication No. It is defined by an editor's decision, not by its content
Reject A frame that failed technically: blank, fogged, badly out of focus Sometimes, but a blank frame can still date a roll

Brackets are worth a word, because a transparency archive is full of them. Colour reversal film has little tolerance for exposure error, so studio photographers routinely shot the same set-up at several exposures to be sure of one perfect frame. The practice explains a good deal of the arithmetic in our inventory. Our article on why studios used medium-format transparency film describes the working method.

Why would anyone want the frames that were not chosen?

Four kinds of people do, for four different reasons.

Historians and curators want process. Magnum's book of contact sheets, drawn from 67 of its photographers, was assembled to show what the agency calls the creative methods, strategies and editing processes behind well-known pictures, and to answer whether an image was "a set-up, or a serendipitous encounter". You cannot answer that from the selected frame alone. The chosen frame tells you what an editor wanted in 1987; the frames around it tell you what was actually in the room. Our guide to how to read a contact sheet shows the same thing at the scale of a single roll.

Picture researchers and publishers want alternatives. The published frame may have been used everywhere already, or be cropped for a cover, or show a logo the publisher would rather avoid. A variant with the model looking the other way is a new picture.

Conservators and scanning technicians want a spare. Dyes fade unevenly and film gets scratched. When the selected transparency has suffered, its neighbour in the sleeve may be the only way to recover the picture.

Data teams want variation, described honestly. A product photographed from four angles under shifting light is more informative than one hero frame, provided the four are identified as a group. The dataset documentation movement points the same way: the 2018 paper Datasheets for Datasets, by Timnit Gebru and six co-authors, proposed that every dataset be accompanied by a document setting out its motivation, composition and collection process. An archive that has quietly discarded half its frames cannot describe its composition truthfully, and one that has kept them can state the ratio. Ours is set out in the dataset card.

What does the gap look like in a real inventory?

Here are the figures from the published 503-folder inventory. Frames are everything in the folder; shots are distinct photographs.

Folder Frames Shots Frames per shot
Clothing: blouses 780 194 4.0
Drinks: glasses 710 168 4.2
Clothing: men's shirts 741 183 4.0
Cosmetics: lipsticks 764 223 3.4
Cosmetics: nail varnish 866 301 2.9
Polaroid prints 4,005 4,005 1.0
Digital photographs 42,803 42,803 1.0
Whole archive 177,143 95,487 unique about 1.9

Three things stand out. Studio product work carries three or four frames for each picture, which is what bracketing and variant set-ups would lead you to expect. Polaroids are one to one, because each is a unique object. And the digital material is counted one to one in the inventory, although digital shooting produces variants more freely than film ever did; that is a statement about how the inventory was compiled, not about how the pictures were taken. The counts come from the archive's inventory and have not yet been independently reconciled.

Imagine weeding the blouses folder to its 194 selected shots. Five hundred and eighty-six pieces of film would go. The saving would be a few sleeves in a box. The loss would be every alternative crop, every bracket that might have aged better, and any chance of telling a future researcher how the session ran. It is a poor trade.

When is weeding justified?

Sometimes. The case against weeding is a case against casual weeding.

  • True duplicates in bulk. Twenty identical lab duplicates of one transparency can become two, once you have confirmed they are identical and noted how many there were.
  • Material that is not the creator's work, such as other photographers' frames, suppliers' samples or blank stock, which should be separated and listed, not silently binned.
  • Material the owner has no right to keep, which is a legal question, not an archival one.
  • Film that endangers its neighbours. Decaying acetate gives off acid that attacks film stored beside it. Isolate it and copy it first; see vinegar syndrome and how film decays.
  • Born-digital redundancy, where the same file exists in five backup folders. Here weeding is simply good housekeeping, and file hashes make it safe.

Cost is the usual argument for going further, and it is weaker than it looks. Stable film takes up very little room. The real expense is scanning, and scanning can be selective without any destruction at all: scan the selected frame at full resolution, record that six variants exist, and leave them in the sleeve.

How do you write an appraisal rule for a photo archive?

Keep it to a page, and adopt it before anyone opens a box. Ours would read something like this.

  1. Default: keep. Every frame made by the photographer is retained unless a clause below applies.
  2. Describe before deciding. No item is removed until the folder has been counted and the count recorded, frames and shots separately.
  3. Duplicates. Exact duplicates may be reduced to a stated number, for example two, after comparison. For files, identity is established by checksum, not by eye. The free image hash generator produces a SHA-256 for exact identity and a perceptual hash for near matches, and our guide to finding duplicate photos with hashes explains the difference.
  4. Variants and brackets are not duplicates. They are grouped, not removed. When in doubt, the image compare tool shows whether two scans differ.
  5. Hazard. Deteriorating film is isolated, copied and assessed. Disposal requires a second person's agreement.
  6. Record. Anything removed is logged: what, how many, when, by whom, under which clause.
  7. Review. The rule is dated and revisited when scanning is complete.

The sixth clause is the one that makes the rest credible. A finding aid has an element for exactly this, "appraisal, destruction and scheduling", and an honest entry there is part of writing a finding aid. An archive that can say "nothing has been weeded" or "exact duplicates reduced to two per frame in 2027" is easier to trust than one that says nothing.

Where this leaves a working archive

More than half of the Robert Harper Photograph Library is still on film and has yet to be scanned, so most of its variants have not been looked at closely by anyone since they were sleeved. We count that as an asset. The frames and shots for every folder are published, and the ratio between them is part of the description, not something to tidy away.

AI and data teams who want to understand how variants and near-duplicates are counted and grouped can start with the dataset card and then read about AI training data licensing with Archive.London, which is available by negotiation with the single rights holder, subject to rights review.

Frequently asked questions

Should you keep outtakes in a photo archive?

In most cases, yes. Outtakes and variants show how a picture was made, supply alternatives when the chosen frame is damaged or restricted, and give researchers and data teams the variation a single frame cannot. Film that is stable costs little to keep. Discard only under a written rule, and only true duplicates, material outside the archive's scope, or film that endangers the rest.

What is weeding in an archive?

The Society of American Archivists defines weeding as the process of identifying and removing unwanted items from a collection. It is usually done item by item during arrangement and description, and is lighter than reappraising a whole collection. The dictionary's sources describe it as traditionally limited to obvious duplicates and irrelevant or very low-level material.

What is the difference between a duplicate, a variant and an outtake?

A duplicate is a copy of the same exposure, such as a lab-made duplicate transparency. A bracket is the same composition at a different exposure. A variant is a different arrangement, angle or moment from the same session. An outtake is any frame not selected for publication. Only the first is redundant; the others each record something the selected frame does not.

Do near-duplicate images matter for AI training data?

They matter in both directions. Variants of one subject under different lighting or angles are useful variation. Unflagged near-duplicates can also distort a dataset, for instance if almost identical frames fall on both sides of a training and evaluation split. The answer is to keep them and label them as a group, not to delete them.

When is it right to discard photographs from an archive?

When a written appraisal rule says so and the decision is recorded. Reasonable grounds include exact duplicates beyond a set number, material that is not the creator's work, items the owner has no right to hold, and film whose decay threatens neighbouring material, which should be isolated and copied first. Record what was removed, when, by whom and why.

Sources

Published by Archive.London. Corrections are welcome under our editorial policy.

Help digitise the archive

52,684 analogue originals, mostly medium-format transparencies, are waiting to be scanned and catalogued. We are speaking to scanning, cataloguing and tagging partners.

See the digitisation briefWrite to the archive