Exploring Instant Photography utilizing Generative AI: A Design Probe with the UnReality Camera

The system takes two enter channels: visible and auditory. For visible enter, the video digicam module mechanically captured the environment, which have been then frivolously processed (e.g. adjusting decision) earlier than being displayed on the LCD. For auditory enter, customers pressed a button to begin and cease recording to seize audio knowledge. The audio enter transcription was displayed on the LCD for consumer reference. The audio knowledge was transcribed utilizing OpenAI’s whisper-1 mannequin, and subsequently translated into a picture era immediate utilizing OpenAI’s GPT-4o-mini mannequin, with the immediate “provide ONLY a comma-separated prompt with image generation tags for stable diffusion for the input” and the position “You help me write prompts for Stable Diffusion”. This course of helped translate uncooked spoken language, which might embrace filler and ambiguous wording, right into a structured keyword-based syntax which aligns with coaching. Keyword-based prompts are extensively adopted within the generative AI group to create extra controllable outputs (Oppenlaender, 2024) akin to immediate enlargement strategies (Datta et al., 2024). We examined this informally throughout implementation; nonetheless, we acknowledge that this provides a degree of interpretation to the consumer’s intention.

We selected to make use of the generative AI mannequin of Stable Diffusion v1.5, offering a mix of velocity and high quality, whereas additionally being accessible by way of value, domestically runnable, and configurable by the researcher. Although this particular mannequin was chosen for our system, in essence, any form of stochastic generative mannequin would have been enough for our probe. To interface with the era mannequin, we used the API offered by the favored Stable Diffusion Web UI, which ran domestically on a pc (Intel Core i7-10700K processor and NVIDIA GeForce RTX 3080 graphics card) and was uncovered utilizing ngrok. In specific, we used the img2img endpoint, which takes a picture (of the camera-captured environment) and a payload (with the picture era immediate) to generate a brand new picture.

To introduce various prospects for picture era, we added 4 fine-tuned variants of Stable Diffusion v1.5 to create totally different inventive kinds. We assigned them every a one-word descriptor — Anything V3 (anime), DreamShaper V8 (whimsical), OpenJourney V4 (fantastical), and Realistic Vision V6.0 B1 (reasonable). Users might cycle by means of totally different fine-tuned kinds for the picture era course of, and the presently chosen model was displayed on the LCD with its one-word descriptor.

The parameters for the picture era, such because the variety of steps (20), denoising power (0.55), and cfg_scale (20), have been subjectively tuned by means of trial and error till the values appeared enough for creating a picture that largely retained qualities of the captured environment however was nonetheless totally different; see Figure 3 for examples. Given the unpredictable nature of AI-generated photos, we additionally integrated detrimental prompting to filter out NSFW and dangerous content material.

The closing generated picture was printed as a bodily output by the Ivy 2 printer, related to the Raspberry Pi by means of Bluetooth. This served as a tangible artifact for the customers to interpret and use as they see match. As our system takes time to carry out all of the processing steps, the printing doesn’t begin instantaneously after the button click on. Instead, a casual take a look at on 10 printed snapshots indicated that the printer would start printing (i.e. an audible printing sound) at a median of roughly 21 seconds after clicking, and the complete print may very well be collected after roughly 63 seconds. This is comparable, if not sooner, to the speeds of typical Polaroid cameras (Buse, 2010).

Participants adopted methods to navigate the stochastic, but rule-bound house of prompting, whereas missing a full understanding of how the prompting and era course of labored. When individuals had a particular unreal scene that they wished to seize in a sure imagined method, their method of developing it was by means of verbal communication with the gadget. This course of, distinctive to our system in comparison with different prompt pictures techniques (Page and See, 2025), turned some extent of communicative friction. For some individuals, trial-and-error processes have been used to barter their meant purpose with the algorithm:

“I think I sort of struggled to like, verbalize what I wanted to say, and I fell back to these like one or two word prompts that were very short and nonspecific.” – P10

This highlights one of many key difficulties of human-AI co-creation — the alignment of intention, particularly by means of phrases because the medium for nuanced intention communication. Participants resorted to trial-and-error to regulate their prompts to align with their desired outputs. As P13 acknowledged: “I’m looking at how I can change my input to get the result that I want” (P13). To assemble the pictures they wished, individuals engaged in a dialogue with an AI system, one that usually couldn’t completely perceive what they wished (both resulting from lacking context clues, lack of environmental understanding, and so forth). This dialogue induced friction within the layers of authorship that will not sometimes exist in conventional prompt pictures, resulting from misalignment at probably a number of steps — the precise human intention, their linguistic expression of their intention, the algorithmic interpretation of the phrases, and the generated picture primarily based on that interpretation. This friction in authorship was additionally exacerbated by temporal friction. P2 and P10 point out the “iterative” means of inventive creation, which is misplaced within the time it takes to “wait [for] it to be printed” (P2).

Participants’ experiences, together with their expectations and interpretations of the unreal photos, have been depending on their shifting perceptions and understandings of generative AI inside the job. Participants analogized the AI as having all kinds of various roles relative to themselves, e.g., as an unsupervised intern (P1), an worker (P3), a unhealthy assistant (P5), a director (P7, P11), a collaborator (P12) or co-creator (P2), and so forth. These roles have been established primarily based on the interrelated ideas of expectations, management, and perceived authorship, and have been harking back to these present in prior works on generative AI techniques (Oh et al., 2018; Shojaei et al., 2024; Panchanadikar and Freeman, 2024; van Es and Nguyen, 2025).

Participants interpreted the hierarchy of roles between themselves and the AI primarily based on how they perceived the stability of labor in creating an output. Many of the individuals initially anticipated themselves to keep up the “main role of the photographer” (P8), with the AI appearing as an “assistant to help me shape what I want to show” (P8). Yet, individuals ended up having various perceptions concerning their enter. Some individuals thought that they had a excessive diploma of enter: “I am the creator because I was the one who said the model and the prompt” (P5); others felt like that they had minimal enter: “My role is simply prompting and the AI did everything else” (P3). This was subjective, because the individual’s procedural involvement was fixed — they pointed the gadget, selected a mannequin, gave a immediate, and clicked the button to take a picture. Relatedly, individuals’ interpreted roles have been tied to their perceived degree of management. For occasion, if the AI was perceived to have extra of a job within the closing picture, then the participant could really feel “I was the assistant” (P12).

Participants’ assignments of the AI’s position additionally relied on their orientation in direction of the AI — whether or not it was to construct a particular imaginative and prescient or to brazenly interpret an thought. Again, this various largely participant to participant — some individuals had robust expectations on what the output ought to be, e.g., “add things into the photos” (P3), “tweak up the image” (P4); others had free preliminary expectations, e.g., “I didn’t have really any picture in my head about what it would have looked like, I didn’t really have any expectation about it” (P13). The expectation that the participant envisioned for the output picture affected their notion of it as useful or disruptive. With a looser set of expectations, individuals might backfill the generative AI’s interpretation of their immediate:

“For this one, where [I prompted] ‘I’m tired, heading home’, it interpreted the kind of staircase, plus the walls on the side, plus the railing, as a subway train home.” – P14

Given the context of individuals’ building and interpretation of the generative AI imagery, individuals skilled and expressed a variety of various emotional responses in direction of the ultimate unreal picture. At one excessive, individuals felt opposed emotions reminiscent of disappointment and vexation — e.g. “frustration, irritation, … lead someone who’s feeling that way to not use it” (P11), and “it could be frustrating that someone wanted a specific thing” (P12). These emotions arose when the consumer had a particular imagined output that they wished:

“[if] it does meet your expectations and if you enjoy AI, it’s satisfying. In this case, I found like, if it generates completely different than what I wanted to, it’s a bit off-putting.” – P14

When the individuals had a stronger sense of their imagined output — a extra goal-focussed interpretation — they might reply positively when AI met these expectations (e.g. for P2, when the pictures “I [felt] are going the exact direction that I expected, in this case I’m happy”), and negatively in any other case (e.g. for P3, the pictures have been “jarring” as a result of they “completely changed the background… I felt like I was just generating AI images”). Stochastic era essentially creates a chasm between human intentions and algorithmic interpretation, an inherent hole in human-AI co-creation, as subjective interpretations could by no means align completely. When people approached their use of the UnReality Camera with a robust intention and expectation, they usually reacted negatively to interpretation: “in general it’s negative because rarely do I have something in mind and then it is better than what I have imagined” (P3).

Yet, when the individuals had a looser, open interpretation of their imagined output — a extra exploratory interpretation — they have been extra prone to reply positively. In this case, it was much less about matching their imaginative and prescient however utilizing the generative AI to extra freely interpret an idea, with an open curiosity in what it could provide you with. For occasion:

“The prompt was ‘I am hungry’, which I didn’t have like expectations about the outcome. For these cases, I kind of think that it was more fun and serendipitous.” – P5

P13 discovered that it was extra fascinating than detrimental to “see what an AI brain interprets as what I said”, and P15 acknowledged that “it didn’t bother me that they all turned out different” indicating that “the surprise factor is there… they’re all creepy in their own way” (word: creepy was used positively right here, as a way of eerie delight). P8 acknowledged that, though the immediate signifies what they need to see, they didn’t anticipate the AI to have priors and simply wished the AI to supply its interpretation, indicating that “because you never know the outcome, you can see some very interesting or shocking stuff”. From this attitude, unpredictable stochasticity turns into a function that probably serves higher to discover and encourage quite than create a particular imaginative and prescient. All in all, individuals’ emotional responses hinged on not solely the precise output picture, but additionally their strategy, interpretation, and orientation in direction of the system and in direction of uncertainty.

As foregrounded by participant responses already, artwork is as depending on the creation course of as a lot as the end result. We revisit the consequences of the tangible, bodily dimension of each the photo-taking system and the resultant output picture. Firstly, the bodily type of the system supplied a way of enjoyable and whimsy, as P3 talked about that this induced them to strategy it with a “whimsical and candid nature of a Polaroid camera versus trying to take the best picture possible”, and P1 acknowledged that it felt extra “like a toy rather than a software”. The sense of getting a weighty, materials system that they might maintain offered customers with a quite indescribable sense of worth, private expressiveness, and intentionality:

“Personally, I love the more physical interaction that holding a ‘camera’ provides, and the tactility of the buttons… significantly different than holding a phone. Fresh, novel, more liberating, whereas the phone feels kinda like a ‘sink’, no responsiveness and very 1-dimensional.” – P11

“I personally think holding a physical, camera-shaped object makes me feel like the photo is more intentional… however, taking photos on a phone feels more casual. If the study was done on a phone, I feel like the sense of control over the result would be even less than having the study done with the camera-box.” – P12

Participants indicated that physicalizing each the output and the gadget made the method really feel extra intentional, extra proximate, and extra important when in comparison with the ubiquity of a cellphone. For P10, “a physical camera is more meaningful” and for P14 “the AI is in your hands right now… It’s less significant when you consider it on your phone”. Part of this significance arises from having a tool with a sole goal of taking photos, as P10 acknowledged that with such a tool, they’re “less likely to get distracted and ‘taken out of the moment’ ”; P5 prolonged that “holding the object that is specifically designed for photography makes me feel more engaged, because when I take pictures using a phone it’s just one of many functions”. The physicality of an intentional gadget supplied a solution to recapture which means, significance, and resonance with the particular follow.

The bodily nature of the printed output remodeled participant responses in direction of the end result of pictures. A bodily output heightened the sense of possession and belonging that the individuals felt in co-creating the picture, e.g. “printing out the pictures makes me feel stronger ownership” (P5). Participants acknowledged that:

“Having like a physical representation of what you took is more… it’s not just using the visual sense. It’s also like you’re holding it and you can see it and share it in ways that aren’t strictly just visual.” – P13

“You just get this instant physical thing in your hands that in some ways has more value to it than just a little digital photo on your phone’s camera roll.” – P10

Yet, this sense of possession is available in pressure with the inherent lack of emotions of management and authorship that is available in AI-generated artwork within the first place, outlined in 5.1.2. We discover that physicality helps offset the lack of management or possession that comes with generative AI-assistance and transforms stochasticity right into a type of private uniqueness. It transforms the generated picture into nearly a random reward, or as P2 cash it, a “collectible”; one the place “there’s only one copy of that photo in the entire world, and you can’t really replicate it” (P12).

Another necessary dimension of prompt pictures is temporality. The first temporal interval we contemplate is the ready interval between the consumer’s click on of the button and the precise printed output being obtained. Initially, we thought of this ready interval an inconvenient inevitability, as filler time for the generative algorithm to provide an output and for the info to be despatched and printed. Yet, we noticed this ready interval to turn into an necessary a part of the photo-taking expertise as a type of “stochastic friction” — which we outline because the anticipatory, suspenseful ready interval the place folks hope, surprise about, and speculate about their “reward”, of what the end result may very well be. As highlighted in 5.1.3, the randomness of the output meant that individuals’s reactions can vary from shock and inspiration to disappointment; emotions on this ready interval constructed as much as these feelings:

“It’s almost like a one-time interaction; you provided something, and you are waiting for feedback. So in that moment, I do have some expectation and curiosity in seeing the result.” – P2

“I guess I am expecting… imagining what the result the system might give. A sense of anticipation before the actual printout.” – P8

We additionally contemplate the temporal dimension of the machine within the broader context of human needs. For individuals, the quite instantaneous ready time provides a solution to “try again” if the primary picture didn’t come out the best way that they imagined — this was a follow that we noticed for some individuals, who took a number of footage, maybe with a barely tuned immediate, to get one thing nearer to what that they had in thoughts. Yet, the ready time of the gadget additionally made retries troublesome and effortful. This provides a really candid, one-shot strategy to capturing the second much like prompt pictures, over which you’ve got minimal management over the only output picture — P7 signifies that this instantaneity permits one to “try and capture something in the moment without so much planning behind that”, P11 analogizes the attraction as “being in the moment and you don’t have a second chance”. P15 lastly states that for Polaroids, you by no means knew what you have been going to get, drawing a parallel to the attraction of our system in exactly that “it’s a crapshoot”. This inherent randomness introduces an aesthetic and an emotional threat, when it fails (e.g. “By the time you get [the output], you’re not there anymore” – P7), the second is misplaced.

Our findings underscore pressure concerning the temporal aspect of generative stochastic AI. As AI techniques goal for elevated velocity and effectivity, in addition they strip away components of anticipation (by means of friction) and momentary significance (by means of effort), which contribute to the “wonder” of applied sciences reminiscent of prompt pictures.

Putting all of it collectively, the method of utilizing the UnReality Camera was human-focused and located. Instead of creating it straightforward for the consumer to think about and create their desired output, it requires them to hold a tool, stroll round, wait, and discover. We recall individuals strolling round and exploring totally different locations, such because the institute’s rooms and hallways, the grassy strolling paths open air, and even empty rest room stalls. Participants tried many alternative framed pictures and quite a lot of prompts, reminiscent of attempting so as to add a porcupine into the grass, or imagining the buildings 100 years from now (some examples are discovered within the supplemental supplies). Even when the outcomes have been generally disappointing, individuals contrasted with a a lot simpler and tunable means of prompting a mannequin:

“I think even if I sat for like an hour and prompted a model and like created the exact thing that I was wanting to, I think I would have trouble feeling like I created this content just because I didn’t go through the process of actually making it.” – P10

Calling forth among the individuals’ definitions of artwork, we once more spotlight the significance of the method past the end result — P11 talked about that two related photos made in numerous methods are “inherently very different”, highlighting how the course of of creation reframes notion. P5 acknowledged that “art is creating something fun with a purpose… in terms of that, I think this is art”, and P14 states that “it’s more personalized to you because you took the effort and time to [make] that one photo”. This contrasts with the controversy of possession that comes from AI utilization in artwork, which removes the human authenticity from inventive endeavours. This pressure was evident in different contrasting participant responses — “I don’t consider [it] art because I don’t think I had creative direction or control” (P11). Synthesizing the whole lot, our findings underscore the intrinsic worth of the expertise of prompt pictures. Time, effort, and human involvement have more and more turn into abstracted away, but these underscore emotions of surprise, anticipation, possession, and resonance.

Our findings that individuals usually felt a lack of management in comparison with conventional pictures agree with prior work on generative AI within the arts. Prior analysis highlights the tradeoff in autonomy for collaborative assist, and that individuals have usually wished to be primarily in management (Oh et al., 2018; Johnston and Thue, 2024; Bryan-Kinns et al., 2024; Guo et al., 2024; Sun et al., 2024). The sources of end result management in our work have been blended. Similar to the A(I)Cam (Page and See, 2025), our digicam takes human enter from the environmental seize and voice enter, but the ultimate step of mixing these two human inputs into a visible picture is delegated to an unpredictable and stochastic AI. However, though the system end result was unpredictable, its behaviour was steady throughout individuals. Instead, we discover that participant responses and emotions in direction of the end result have been led by means of primarily particular person perceptions of the roles of generative AI, the power of their expectations, and their alignment in direction of the purpose of taking such a photograph.

When we mix the expectations formed by means of a comparatively excessive diploma of management (e.g. by means of framing the shot and passing in a immediate) with the inherent lack of management of generative AI, the ensuing co-creation is commonly misaligned with what customers need. People’s robust expectations have been knowledgeable by prompting, the place they primarily explicitly advised the AI what to do. Such expectations distinction techniques with extra summary outcomes (e.g. the Dream Sticker Machine (Bønlykke et al., 2024)) or techniques with weaker human management (e.g., the A(I)Cam (Page and See, 2025)). Although the prompting mechanism allowed for exploration of prospects (aligning with (Dang et al., 2023b)), it might additionally create a stronger alignment of expectations, which impacts the judgment of company (Didion et al., 2024). Our findings counsel that elevated enter management created extra expectations and consequently extra frustration when expectations weren’t met reliably. In distinction, we hypothesize that much less management, like in A(I)Cam (Page and See, 2025), could align higher with stochastic, summary, and unpredictable outcomes. This probably ties to psychological possession as properly — elevated effort and time expenditure (i.e. by means of prompts) leads to the individual feeling like they “own” the generated end result (Joshi and Vogel, 2025), maybe as an expression of their management and identification (Pierce et al., 2001).

One short-term method of addressing the expectation hole is to enhance the reliability and efficiency of the AI. Challenges in aligning intent in human-AI co-creation are presently widespread (Dang et al., 2022). We draw inspiration from prior analysis on how the design of prompting techniques may be improved to raised align consumer expectations. Having a number of output choices might supply extra probabilities to satisfy expectations (Dang et al., 2023b), and having the ability to quickly make incremental, reversible edits (Masson et al., 2024), e.g. by means of localized inputs (Gholami and Xiao, 2023), would assist customers guarantee the end result meets their imaginative and prescient. Interactive parts and incorporating human decision-making may be helpful for creating satisfying outcomes that align with the consumer’s inventive imaginative and prescient (Huh et al., 2025; Gero and Chilton, 2019). These options would assist the system outcomes meet the expectations implied by our design (Palani et al., 2022).

While our examine revealed challenges that arose from robust consumer expectations, it additionally revealed {that a} optimistic expertise — serendipity — might come up when consumer expectations have been reframed. Unexpected outcomes and imperfect concepts can spur shock, innovation, and creativity within the arts and past (Caramiaux and Fdili Alaoui, 2022; Wan et al., 2024; Dang et al., 2023a; Chen et al., 2024); in a pictures case, this agrees with Page et al. (Page and See, 2025). In our examine, this usually required the customers to not have a robust intention of the output. We discovered this to be a subjective aspect primarily based on notion greater than precise management, since participant involvement was largely fixed, but their interpretations and degree of expectations differed; such expectations might probably be formed by means of previous experiences, cognitive bias, and particular person character (Riedl, 2022; Nourani et al., 2022). Thus, when expectations have been misaligned with outcomes, it led to frustration; when expectations have been loosened, it led to potential discovery. Expectations may very well be noticed by means of the precise prompts that individuals used all through the research, starting from extra focused, goal, prompts with extra concrete expectations (e.g. “add a porcupine here” – P3), to extra summary ones (e.g. “I am hungry” – P5)

Stochastic era of imagined realities shifts the expectation of reflecting actuality into reflecting prospects, transitioning from a photographic expertise of documenting actuality to speculating on it. Much like previous experimentation with pictures, such shifts straight affect the paradigm of pictures (Laxton, 2016; Newhall, 1941) and its indexicality. We interpret capturing generative realities as entailing a special set of motivations, use instances, expectations, and behaviours when in comparison with capturing actuality; thus, a prerequisite to discovery-based utilization is to speak to the consumer to set their expectations appropriately. For LLMs, this may increasingly embrace speaking early that it hallucinates (Metze et al., 2024); analogized for our system, it could be speaking that the mannequin may not at all times comply with your directions precisely. This is also achieved extra abstractly — reminiscent of mentioning that the system is supposed to think about prospects quite than make particular edits. Through communication, the number of roles that have been assigned to the AI may turn into extra distilled, lowering the confusion and frustration probably generated from AI’s ever-changing roles (Zhang et al., 2024).

Aligning with earlier work, AI-supported serendipity can induce emotions of inspiration and new prospects (Oh et al., 2018; Wan et al., 2024), probably serving to pull artists out of psychological blocks (Wan et al., 2024). We discover that such stochastic shock additionally varieties a novel supply of anticipatory pleasure and engagement (much like a gacha pull (Yin and Xiao, 2022)), the place individuals really feel a spread of feelings in direction of a randomized consequence.

We revisit the type of our system, contemplating its visible and bodily look (Jung and Stolterman, 2012), as materials type is necessary to symbolism which means of the expertise. Fuchsberger et al. (Fuchsberger et al., 2013) mentioned how the materiality of a system impresses people and interplays with human actions and experiences. The type of our gadget, harking back to a Polaroid, captures sides of expression and which means (Jung and Stolterman, 2012). We contemplate how merely having a tangible bodily type enabled emotions of meaningfulness, significance, and possession over the expertise and artifact in our examine.

These emotions could partially stem from the symbolic worth of fabric type — individuals differentiated from a cellphone, which comes with particular connotations (Chopra-Gant, 2016; Lou et al., 2022). Smartphones are multipurpose and ubiquitous, the UnReality Camera, in distinction, turned purposeful and intentional. Furthermore, the anthropomorphism of the generative AI part of the digicam labored on high of human tendencies to anthropomorphize bodily objects usually (Boyer, 1996; Wan and Chen, 2021), making the notion of the digicam a lot totally different in comparison with a special photo-taking gadget, reminiscent of a cellphone — they ascribed it anthropomorphic roles and felt a connection to the bodily gadget. Connecting anthropomorphism to AI, Rozendaal et al. (Rozendaal et al., 2018) mentioned the type of AI-incorporated objects particularly, discussing intelligence as character and contemplating their familiarity, authenticity, and sociability; strongly tying to our examine’s depiction of AI as enjoying sure roles. Aligned with prior work, individuals described the AI utilizing anthropomorphic language and assigned the AI agent within the digicam with numerous character traits (Salles et al., 2020).

With this reframing in symbolic understanding and cognitive notion of the photographic object, individuals adjusted their emotions to the method accordingly. We tie this to psychological possession, as having a bodily gadget might imbue a stronger private worth past the extra digital smartphone (Atasoy and Morewedge, 2018); the custom-built gadget being distinctive provides a way of shortage as properly. When we juxtapose this towards the much less “authentic” means of generative AI, which entails a way more nebulous definition of, and emotions concerning possession (Xu et al., 2024), this harks in direction of how bodily interfaces can paradoxically improve human possession or connection even towards one thing as indifferent, uncontrollable, and non-interpretable as generative AI.

The precept of bodily possession extends not solely in direction of the bodily digicam, but additionally the bodily output as a printout. When these outputs require human effort and time, imbueing their ideas and emotions, and are scarce (i.e. restricted by printing paper), they turn into artifacts that individuals felt a way of psychological possession in direction of (Atasoy and Morewedge, 2018; Pierce et al., 2003, 2001). Embedded in generative AI, the bodily supplies assist folks in feeling extra related to the supplies of the expertise, taking stronger possession of the output, and ascribing significance to the method even when motion stays largely fixed. Thus, whereas physicalizing a digital system to be used in an embodied course of requires extra effort and time, its combination with a stochastic course of can reframe the method so as to add individualization and possession.

We additionally contemplate the temporal dimension of our system. The shift from viewing design as revolving round “things” to “events” has underscored the significance of time, which might have an effect on how folks replicate and derive which means from their experiences (Wiberg and Stolterman, 2021). Much work has studied the idea of sluggish expertise, which provides a paradigm to design reflective and amplified experiences (Hallnäs and Redström, 2001). Although prompt pictures got here with the promise of speedy outcomes, it nonetheless took time for the photographic consequence to turn into seen. This temporal dimension is mimicked by the UnReality Camera, which takes a comparable period of time resulting from its era and printing velocity. We discovered that this ready interval, i.e. “stochastic friction”, was a boon for constructing anticipation, much like Odom et al.’s Photobox (Odom et al., 2014), serving to construct anticipation in direction of an end result and induce reflection (Odom et al., 2014).

Similar to Photobox (Odom et al., 2014), which additionally incorporates a level of identified randomness, the stochasticity of generative AI in our work exacerbated anticipation. Participants, not realizing what the output is likely to be, used this time to think about and speculate. Yet, this contrasts with present AI techniques that prioritize velocity. This delay additionally made it troublesome for folks to “redo” the picture when it turned out in a method that they didn’t like, contrasting present AI techniques that permit for re-generation. The UnReality Camera introduces slowness into the method, which inhibits usability however establishes a way of significance and uniqueness. During ready instances, folks find yourself pondering (Buçinca et al., 2021; Cox et al., 2016; Tepponen et al., 2025), disrupting senseless use (Cox et al., 2016) and shifting right into a System 2 paradigm underneath dual-system idea (Kannengiesser and Gero, 2019). For goal quite than inventive duties, this pondering can materialize as analyzing for the “correct” reply (Buçinca et al., 2021). On the opposite hand, imbuing stochasticity into prompt pictures, a inventive, summary course of with no “correct” reply, interprets this pondering as an alternative into anticipation and surprise. Waiting instances may be detriments to usability and effectivity of generative AI instruments and might exacerbate already present points with negotiating authorship (Section 5.1.1). Thus, whereas slowness in an inventive course of can create frustration and inefficiencies, deliberate incorporation of slowness may also get better the magic and surprise by means of anticipation.

Altogether, the UnReality Camera integrates generative AI into the embodied, sluggish means of prompt pictures to probe into the way it impacts the expertise. Calling again to prior design probes, e.g. (Pierce and Paulos, 2014; Seo et al., 2025), such exploratory probes are necessary as a result of they present what’s gained or misplaced by means of paradigm shifts — modifications in materials, velocity, and human involvement.

By imbuing generative AI in a sluggish and tangible interface, we in the end sacrifice some degree of usability and ubiquity – the method can not be shortly carried out, simply iterated on (if the output is disliked), and digitally processed with out the necessity for a paper output. However, by making the system sluggish, deliberate, and bodily, we imbue the method with a way of non-public significance and uniqueness — a way that non-public effort and time have been spent to create this expressive output. Even if the visible output was not essentially loved, we present that generative AI additionally performs a job in shaping the expertise of making this output, which is necessary as properly. This agrees with how such elements can enhance the aesthetics of deliberation and the high quality of motion from Seo et al.’s work (Seo et al., 2025). Altogether, we spotlight that modifications in temporal and bodily qualities essentially change the objectives of how folks understand, interact with, and surprise about pictures. Designers can leverage modifications in type and time inside stochastic practices as a solution to commerce off between an anticipatory and important expertise towards a very usable one.

As customers body the distinctive second, they seize a second that’s troublesome to seize once more. The expertise turns into a performative (Jacucci, 2015) expertise to precise one thing private and distinctive, even when the end result of such expression is ambiguous in tone and subjective in which means resulting from embedded generative AI. In essence, the UnReality Camera is ironic. It materializes prompt pictures, a nostalgic course of intertwined with human autonomy and insurrection towards digital modernity (Lathrop Ligueros, 2020; Minniti, 2020), whereas incorporating generative AI, a up to date, digital course of that complicates emotions of human company (Allred and Aragon, 2023; Oh et al., 2018; Halperin and Rosner, 2025). By fusing these, the UnReality Camera re-enchants generative AI, a expertise usually stripped of its magic by its ubiquity and effectivity, by giving it a bodily type, a novel materials output, and deliberate temporality. At the identical time, it re-explores prompt pictures as not only a web site for one-shot seize and materials output, however a supply of hypothesis, unpredictability, and surprise.

Our dialogue has explored how the stochasticity of the method and its integration inside the temporal and bodily expertise of prompt pictures form notion and interpretation of each the method and the end result. We imagine that this embedding may also essentially shift the idea of photographic which means, tied to our preliminary idea of “unreality”. By subverting conventional photographic indexicality, the UnReality Camera opens up new potentials of speculating on the totally different summary “unreal” prospects — individuals mentioned how utilizing the UnReality Camera enabled hypothesis on what may very well be, quite than reflection on what’s, appearing akin to a Monte Carlo simulation of actuality.

Changes in inventive course of and which means can hark in direction of shifting actions and cultural zeitgeists, essentially shifting how folks produce and interpret artwork. For occasion, Dadaism took benefit of chaos and randomness in inventive processes to signify existentialist themes (Dzhimova and Moura, 2024; Rosen, 2014; Kristiansen, 1968). Thus, inventive worth and which means arrive each from its end result and the method of creation. While the UnReality Camera departs from conventional indexicality, the incorporation of stochastic randomness (much like Dadaism) and temporal delay in each course of and end result can probably open up a slower, reflective layer of understanding, meaning-making, and interpretation, essentially reshaping which means pushed by the design of the method.