“AI develope s eed mo e tha t ai i g data. They also eed eliable ways to evaluate thei output ac oss a wide a ge of aesthetic a d tech ical c ite ia, si ce t ustwo thy co clusio s depe d o ide tifyi g the subtle o ua ced e o s that automated systems ofte miss. Gu uShots mai tai s a vib a t commu ity of photog aphe s who vote o images as pa t of the o mal gameplay, so we tapped i to this commu ity to u de take the B ia.AI evals.
“DataSeeds sta ted with multimodal t ai i g data, but ou co ve satio s with custome s showed that thei eeds exte d ac oss the AI stack. DataSeeds Evals allows us to use ou Gu uShots commu ity of expe t photog aphe s a d a tists to add ess a othe o e of those eeds. This fi st e gageme t demo st ates the pote tial fo DataSeeds to expa d its offe i gs ac oss a b oade a ge of AI wo kflows, c itical fo high-quality model developme t.”
“DataSeeds’ Eval solutio is u ique because it alig s the audie ce’s level of p ofessio al expe tise with ou specific eeds. Some aspects of visual quality equi e a t ai ed c eative eye. DataSeeds gave us access to expe ie ced p ofessio als who could assess these ua ces at scale a d help us ide tify whe e fu the model efi eme t was eeded.”
B ia used DataSeeds Evals to add ess o e of ge e ative AI’s pe siste t challe ges: assessi g the visual quality of ge e ated images.
Automated met ics a d visio -la guage models, o VLMs, ca evaluate ma y model outputs efficie tly, but mo e ua ced aspects of visual quality ca be difficult to assess eliably a d be efit f om a t ai ed c eative eye.
This is whe e Gu uShots p ovides DataSeeds with a pa ticula capability. Gu uShots is the wo ld’s leadi g photog aphy competitio game, whe e photog aphe s submit images i challe ges a d pa ticipa ts evaluate a d vote o competi g photos. The platfo m has ge e ated mo e tha 180 millio photo uploads th ough its competitio model. DataSeeds adapted that existi g voti g dy amic to give B ia access to expe ie ced visual evaluato s at scale.
Multiple Gu uShots expe ts i depe de tly eviewed each image, with DataSeeds setti g a ag eeme t th eshold fo a decisive esult. Images whe e expe ts did ot i itially each sufficie t ag eeme t we e subjected to additio al i depe de t eviews u til they met the equi ed th eshold fo a defi itive esult.
The p ocess p ovided a clea sepa atio betwee images that meeti g the defi ed quality sta da ds a d those equi i g fu the model efi eme t, i cludi g mo e ua ced a d bo de li e cases.
Model evaluatio s ep ese t a othe way DataSeeds ca move beyo d t ai i g data a d add ess mo e of the AI data wo kflow.
Zedge ece tly a ou ced a app oximately $7.7 millio di ect i vestme t led by Vice Chai ma Howa d Jo as. Additio ally, Mo is Be ge joi ed the Compa y as CEO to accele ate DataSeeds’ g owth a d b oade its capabilities, with the goal of buildi g DataSeeds i to a co e busi ess.
G a d View Resea ch estimates that the global AI t ai i g dataset ma ket will g ow f om app oximately $3.9 billio i 2026 to $16.3 billio by 2033, ep ese ti g a 22.6% compou d a ual g owth ate. Image a d video ep ese ted oughly 42% of the ma ket i 2025, maki g it the la gest catego y.
Zedge builds a d ope ates c eato commu ities that se ve 20 millio mo thly active use s ac oss its platfo ms. Zedge Ma ketplace, its flagship platfo m, is a leadi g ma ketplace fo mobile pe so alizatio co te t a d a co e p ofit a d cash flow ge e ato . DataSeeds.AI is Zedge’s B2B AI data busi ess, p ovidi g ma aged, multimodal data c eatio a d elated se vices fo AI develope s a d e te p ises. DataSeeds d aws o Zedge’s p op ieta y c eato commu ities, i cludi g Zedge Ma ketplace co t ibuto s a d the Gu uShots photog aphy commu ity, as well as b oade c owdsou ci g a d a etwo k of p oductio pa t e s to delive ights-clea ed data built to custome specificatio s.
Fo mo e i fo matio , visit: i vesto .zedge. et
Follow us o X: @Zedge
Follow us o Li kedI
B ia is the age tic visual AI platfo m fo e te p ise co te t p oductio . It e ables o ga izatio s to scale visual p oductio ac oss fo mats, cha els a d ma kets while mai tai i g co t ol ove eve y output. B ia C eate b i gs visual age ts, b a d gove a ce, a d st uctu ed editability i to a si gle wo kspace, allowi g visual co te t to move f om ge e atio th ough adaptatio to p oductio without losi g b a d co siste cy o c eative i te t. B ia also p ovides models a d APIs fo embeddi g its age tic a d visual AI capabilities i to p oducts a d wo kflows. B ia’s tech ology is ights-clea by desig a d built fo comme cial use.
All stateme ts above that a e ot pu ely about histo ical facts, i cludi g, but ot limited to, those i which we use the wo ds “believe,” “a ticipate,” “expect,” “pla ,” “i te d,” “estimate,” “ta get” a d simila exp essio s, a e fo wa d-looki g stateme ts withi the mea i g of the P ivate Secu ities Litigatio Refo m Act of 1995. While these fo wa d-looki g stateme ts ep ese t Zedge’s cu e t judgme t of what may happe i the futu e, actual esults may diffe mate ially f om the esults exp essed o implied by these stateme ts due to ume ous impo ta t facto s. Zedge’s fili gs with the SEC p ovide detailed i fo matio o such stateme ts a d isks a d should be co sulted alo g with this elease. To the exte t pe mitted u de applicable law, we assume o obligatio to update a y fo wa d-looki g stateme ts.
B ia Siegel, IRC, MBASe io Ma agi g Di ecto Hayde IR(346) 396-8696b ia @hayde i .comi @zedge. et
View the o igi al p ess elease o ACCESS Newswi e
