Why do all NFTs look the same?

Notes on the aes­thet­ics of spec­u­la­tion

A fig­ure pulled from Minecraft, Minions, Fortnite, Looney Tunes, DeviantArt por­trai­ture, the Are ya win­ning son meme, cor­po­rate il­lus­tra­tion, or some other slice of the American ver­nac­u­lar is shown in three-quar­ter por­trait on a field of bright colour, look­ing in­dif­fer­ent. The fig­ure is dec­o­rated with ref­er­ences to pop­u­lar en­ter­tain­ment (Harley Quinns base­ball bat, the three-eyed fish from The Simpsons, the float­ing orb of wa­ter from The Last Airbender, the face mask from Mad Max, Yugi-Yos hair, Venoms teeth), pop­u­lar con­sump­tion (bum bags, hood­ies, puffer jack­ets, sneak­ers, vapes, tinted glasses, branded head­phones, gold jew­ellery, take­out food con­tain­ers), and crypto-spe­cific sym­bols (laser eyes, di­a­monds, and cur­rency icons).

The im­age is ei­ther a vec­tor draw­ing or 3d-rendered, but in ei­ther case there is lit­tle sug­ges­tion of depth; every el­e­ment is evenly lit and de­picted in sharp de­tail as if it were pressed right against the im­age sur­face. There is a love for vi­sual de­tail: every hair is pre­cisely de­lin­eated, every piece of gold is pol­ished, and cloth is care­fully draped, lasers glow.

This de­scrip­tion cov­ers most non-fun­gi­ble to­kens at the top of OpenSea (the biggest web­site for buy­ing and sell­ing NFTs) on any given day. Why do all these im­ages look so alike?

The first level of re­sem­b­lence has to do with the fact that main­stream NFTs are gen­er­ally pro­duced by an op­er­a­tion called lay­er­ing. You start by mak­ing a list of el­e­ments like background”, clothes”, and hat”. Then you pro­duce (or pay a gig worker to pro­duce) a set of im­ages cor­re­spond­ing to each el­e­ment: A few dif­fer­ent back­grounds, some vari­a­tions of your char­ac­ter, and some dif­fer­ent hats. Finally you use a sim­ple com­puter pro­gram to stack these lay­ers on top of each other in ran­dom com­bi­na­tions, pro­duc­ing a set of fi­nal im­ages. The more el­e­ments and lay­ers you have, the more im­ages you can pro­duce, and the big­ger your pay­off will be if the col­lec­tion catches on.

The prac­tice of as­sem­bling an im­age or other me­dia ob­ject from a set of in­di­vid­ual el­e­ments is­n’t unique to NFTs. It has some par­al­lels to col­lage, but the more apt com­par­i­son is com­posit­ing, a processes that hap­pens every­where in con­tem­po­rary me­dia pro­duc­tion. The de­sign of any movie, video game, or other new me­dia ob­ject”, writes the critic Lev Manovich,

… be­gins with as­sem­bling a data­base of pos­si­ble el­e­ments to be used … Throughout the de­sign process, new el­e­ments are added to the data­base; ex­ist­ing el­e­ments are mod­i­fied. The nar­ra­tive is con­structed by link­ing el­e­ments of this data­base in a par­tic­u­lar or­der, that is by de­sign­ing a tra­jec­tory lead­ing from one el­e­ment to an­other. The nar­ra­tive is con­structed by link­ing el­e­ments of this data­base in a par­tic­u­lar or­der, that is by de­sign­ing a tra­jec­tory lead­ing from one el­e­ment to an­other. On the ma­te­r­ial level, a nar­ra­tive is just a set of links. 1

The most vis­i­ble ex­am­ple of this is the film in­dus­try. When Tony Stark walks through a hangar in Avengers: Endgame, the footage of Robert Downy Jr, the 3d-model of the cos­tume, the build­ing, the HDRI sky, the aero­planes in the back­ground, and even the lens flares are only tem­porar­ily brought into the same frame — in re­al­ity they’re sep­a­rate, in­de­pen­dent as­sets (both in the me­dia-in­dus­trial and fi­nan­cial sense of that term), ready to be re-as­sem­bled into other out­puts down the line.

Hey! This page con­tains em­bed­ded con­tent from Youtube, who might use cook­ies and other tech­nolo­gies to track you. To view this con­tent, click Allow Youtube con­tent.

VFX Breakdown for Avengers: Endgame


NFTs are a height­ened, dis­torted ver­sion of this. Making a col­lec­tion is mostly about fill­ing the data­base; the link­ing hap­pens al­most in­ci­den­tally, in the sim­plest way pos­si­ble (randomised lay­er­ing), fully au­to­mated in a few dozen lines of code.

If you want to make vi­su­ally co­her­ent im­ages in this way you have to make the el­e­ments in­ter­change­able; this lim­its the kinds of vi­sual ges­tures you can make and, to­gether with the fact that the same el­e­ments ap­pear again and again, pro­duces col­lec­tions of very sim­i­lar im­ages. It also causes the flat­tened, oc­ca­sion­ally shift­ing per­spec­tive and vis­i­ble seams be­tween el­e­ments that are char­ac­ter­is­tic of many NFT col­lec­tions.

In both cases, the com­pos­ite na­ture of the im­ages is­n’t a se­cret. Disney, who owns the Avengers as­sets, re­leases reg­u­lar VFX break­downs demon­strat­ing the fact to their share­hold­ers and every­one else, and OpenSea shows a list of the con­stituent el­e­ments next to every NFT for the same rea­son.

The sim­i­lar­ity be­tween NFTs across the field (not just in­side a given col­lec­tion) is an ex­ten­sion of this logic. A promi­nent sell­ing point of many col­lec­tions is­n’t so much that they’re as­sem­bled from a data­base, but that they might them­selves be­come one; their im­ages re-as­sem­bled into fresh me­dia prod­ucts like comic books, toys, TV shows, video games, ex­pe­ri­ences, and mer­chan­dise.

NFT pro­jects aren’t gen­er­ally pre­pared to do the work to ac­tu­ally make any of these things (it takes hun­dreds of artists work­ing 12-hour shifts to turn a data­base like Disney’s into a movie like Endgame, not to men­tion the peo­ple sewing the mer­chan­dise or staffing the theme parks), but that does­n’t mat­ter. The idea that such labour could be per­formed in the fu­ture, and that you would be able to pocket the sur­plus by own­ing a piece of the data­base is enough to sell it.

This piece of spec­u­la­tion starts out in roadmaps and other mar­ket­ing ma­te­r­ial, but it quickly seeps down and across into the im­ages them­selves, where it crys­tallises into the vi­sual cues we’re fa­mil­iar with:

Incidentally, this also ex­plains why there is such a big push to en­able NFTs as avatars on so­cial me­dia sites like Twitter and Instagram. It’s a growth hack (people see your avatar, they buy an im­age from the same col­lec­tion and set it as their avatar, more peo­ple see it, every­one prof­its), but more im­por­tantly it shores up the claim that NFTs can be com­pos­ited into other me­dia ob­jects.

Of course, pro­po­nents are quick to point out that this is only proof-of-con­cept. The mer­chan­dise, video games and all the rest will be here any minute now.

Three silver coins show a woman's face. The image varies slightly.

Ancient Greek coins show­ing Helios, the god of the Sun, c.a. 350 AD.

British Museum 1949,0411.781, 1949,0411.775 , 1955,1102.8, all CC BY-NC-SA 4.0

It seems likely that NFTs will im­plode for rea­sons ex­ter­nal to them: The un­der­ly­ing cur­rency might col­lapse, they might be reg­u­lated out of ex­is­tence for the en­vi­ron­men­tal fall­out or the wide­spread fraud (or both), or they might just run out of buy­ers. But there is a deeper ar­gu­ment against them: There is­n’t re­ally such a thing as a non-fun­gi­ble im­age.

In prin­ci­ple”, the philospo­her Walter Benjamin wrote in 1936, the work of art has al­ways been re­pro­ducible”. 2 People through­out his­tory, he ar­gued, made im­ages in mul­ti­ples: The an­cient Greeks mass-pro­duced pic­tures of their rulers by strik­ing them into coins, which they spread across the con­ti­nent. Around the third cen­tury, Chinese cloth-mak­ers be­gan to carve im­ages into wooden blocks, which they cov­ered in ink and pressed against silk, leav­ing a coloured im­pres­sion that could be re­peated over and over. The tech­nique spread, and soon artists in every town were churn­ing out thou­sands of play­ing cards, re­li­gious icons, por­traits, and scenes from na­ture and every­day life. Woodblock print­ing was joined by cop­per­plate en­grav­ing in the 15th, lith­o­g­ra­phy in the 18th, and pho­tog­ra­phy in the 19th cen­tury, each time in­creas­ing the ve­rac­ity and speed with which im­ages could be re­pro­duced.

The cur­rent stage of this de­vel­op­ment is the dig­i­tal im­age, where even the sim­ple act of look­ing en­tails mul­ti­ple acts of re­pro­duc­tion. When you open an im­age on your com­puter, it’s copied from your com­put­er’s hard drive into its work­ing mem­ory, parsed and trans­lated, un­til a spe­cific ar­ray of pix­els in your screen is lit up to ren­der the im­age. As soon as you close the win­dow, those pix­els are turned off and the pic­ture you were look­ing at is de­stroyed, only to be pro­duced afresh the next time you open it. When the im­age is stored on­line (as NFTs typ­i­cally are), this process hap­pens every­time any­one looks at it.

Naturally each of those coins, bolts of silk, packs of play­ing cards, printed pho­tos, and ar­rays of pix­els on your screen are as real” and authentic” and valuable” as all the rest of them — there is­n’t re­ally an original” to speak of.

It takes an enor­mous amount of labour to sus­pend be­lief in this fact, even tem­porar­ily. In a clas­sic es­say, the art critic John Berger de­scribes the lengths to which the National Gallery in London has to go to main­tain the no­tion that their ver­sion of a paint­ing by Leonardo is in fact the orig­i­nal”:

[The cat­a­logue en­try] on the Virgin of the Rocks” is one of the longest en­tries. It con­sists of four­teen closely printed pages. They do not deal with the mean­ing of the im­age. The deal with who com­mis­sioned the paint­ing, le­gal squab­bles, who owned it, its likely date, the fam­i­lies of its own­ers. Behind this in­for­ma­tion lie years of re­search. The aim of the re­search is to prove be­yond the shadow of a doubt that the paint­ing is a gen­uine Leonardo. The sec­ondary aim of the re­search is to prove that an al­most iden­ti­cal paint­ing in the Louvre is a replica of the National Gallery ver­sion. French art his­to­ri­ans try to prove the op­po­site. 3

Similarly, Disney em­ploys an army of copy­right lawyers, im­age recog­ni­tion soft­ware, lob­by­ists, a whole ju­di­cial ap­pa­ra­tus to main­tain the no­tion that the 3d-model of Iron Man in their data­base is in fact the orig­i­nal”, and that every­one else only has a tem­po­rary view­ing li­cense.

NFTs are an­other at­tempt at this. People who make them recog­nise it’s dif­fi­cult to ar­gue that a dig­i­tal im­age can be original” on any ma­te­r­ial level, so they sug­gest a kind of au­then­tic­ity-by-proxy: Buy an NFT and you get a unique en­try in our spe­cial data­base say­ing you own the im­age. That data­base en­try has ef­fec­tively the same func­tion as those fancy art his­to­ri­ans and copy­right lawyers: Establish au­thor­ship, keep track of prove­nance, au­tho­rise de­riv­a­tive works, me­di­ate roy­alty pay­ments, and so on.

Critics ar­gue that this does­n’t work: There is no way of know­ing, for in­stance, if some­one who mints an NFT re­ally made the im­age, and buy­ing an NFT does­n’t re­ally give you own­er­ship of the im­age in any legally recog­nised form.

They’re clearly right, but if there is any­thing to learn from the his­tory of im­age-mak­ing, it’s the no­tion of the at­trib­ut­able, own­able, date­able original” is it­self pretty shaky. Images were al­ways pro­duced col­lec­tively and in abun­dance; the re­cent drive (historically speak­ing) to en­close them for in­di­vid­ual profit must be over­come.4

Explosion diagram of a cartoon monkey's head. Skull, skin, fur and eyes are spread horizontally on white ground.
  1. Lev Manovich (2001): The Language of New Media, p. 231 ↩︎

  2. Walter Benjamin (1936): The Work of Art in the age of Mechanical Reproduction. Available at marx­ists.org/​ref­er­ence/​sub­ject/​phi­los­o­phy/​works/​ge/​ben­jamin.htm ↩︎

  3. John Berger (1972): Ways of Seeing, p. 22. Available at archive.org/​de­tails/​waysof­see­ing00berg/​page/​22/​mode/​2up ↩︎

  4. This post also ap­peared on Medium ↩︎