Izdelava sekvence slik s pomočjo generativne umetne inteligence

Vitežnik, Laura (author), Gabrijelčič Tomc, Helena (mentor), Hladnik, Aleš (comentor)

V diplomskem delu smo preverjali, ali je z različnimi modeli generativne umetne inteligence možno ustvariti sekvenco slik, na katerih je vidno gibanje. V teoretičnem delu je predstavljena umetna inteligenca ter vsi koncepti, ki spadajo pod njo. Natančneje je opisano delovanje generativne umetne inteligence, v nadaljevanju pa delovanje modela iz besedila v sliko (ChatGPT) ter modela iz slike v sliko (Stable DIffusion). Predstavljeni so tudi začetki animacije ter njeno delovanje. V eksperimentalnem delu smo najprej pripravili originalne slike, ki smo jih ročno narisali ter nato digitalizirali za nadaljnjo uporabo. Nadaljevali smo z ustvarjanjem sekvence z modelom generativne umetne inteligence, ki iz besedila ustvari sliko. Najprej smo se spoznali z modelom ter ugotovili, kako mu podajati informacije, da bomo dosegli želene rezultate. Sledilo je ustvarjanje slike za sekvenco. V drugem delu smo slike ustvarjali z modelom, ki iz slike ustvari novo. Predhodno pripravljene slike smo podali dvema modeloma. S prvim modelom nismo dosegli pričakovanih rezultatov. Kljub številnim poizkusom, model ni ustvaril primernih slik glede na podane originalne slike in ukaze. Nato smo uporabili še drug model, ki se je izkazal za učinkovitega. Z njim smo ustvarili slike za končno sekvenco. Rezultat diplomskega dela sta dve sekvenci slik, ustvarjeni z generativno umetno inteligenco. Prav s slednjima smo dokazali, da je postopek ustvarjanja z zgoraj omenjenim modelom, mogoč. Potrdile so se nam domneve, da je modelom potrebno zelo natančno podajati ukaze. Poleg tega pa smo opisan postopek uporabili zgolj za kreiranje enostavnih likov.

• Language: Slovenian
• Keywords: generativna umetna inteligenca, model iz besedila v sliko, model iz slike v sliko, sekvenca slik
• Work type: Bachelor thesis/paper
• Organization: NTF - Faculty of Natural Sciences and Engineering
• Year: 2025
• PID: 20.500.12556/RUL-173041
• Publication date in RUL: 12.09.2025

Secondary language

• Language: English
• Title: Creating sequence of images using generative artificial intelligence

Abstract

In this thesis, we examined whether it is possible to create a sequence of images that show movement using various generative artificial intelligence models. The theoretical part presents artificial intelligence and its related concepts. The functioning of generative artificial intelligence is described in detail, followed by the operation of the text-to-image model (ChatGPT) and the image-to-image model (Stable Diffusion). The beginnings of animation and its functioning are also presented. In the experimental part, we first prepared original images that we drew by hand and then digitized for further use. We continued by creating a sequence with a generative artificial intelligence model that creates images from text. First, we familiarized ourselves with the model and determined how to provide it with information to achieve the desired results. This was followed by creating images for the sequence. In the second part, we created images with a model that generates new images from existing images. We provided the previously prepared images to two models. We did not achieve the expected results with the first model. Despite numerous attempts, the model did not create suitable images based on the provided original images and commands. We then used another model, which proved to be effective. With it, we created images for the final sequence. The thesis results in two sequences of images created using generative artificial intelligence. With these, we demonstrated that the creation process using the aforementioned model is possible. Our assumptions were confirmed that models need to be given very precise commands. Additionally, we used the described process only for creating simple characters.

• Keywords: generative artificial intelligence, image sequence, image-to-image model, text-to-image model