Simulated Dreams and Digital Neural Paintings: Orkhan's Rêveries

In this conversation, Orkhan Mammadov discusses his upcoming Artist Rooms series, Rêveries, a generative AI series that blends neural networks, cultural memory, and real-time simulation. He reflects on his artistic evolution, the role of machines in creativity, and the emotional depth of algorithmic painting. This conversation took place on Twitter, hosted by Ivan Zhyzhkevych and Holly Wrench, and features a brief segment wher two AI agents provide an initial introduction to the series.

HW: Hi everybody, and thanks for joining us. I'm joined tonight by my colleague Ivan and artist Orkhan Mammadov —welcome! For anyone who doesn't know, Orkhan's series Rêveries will be released next week on May 21st, curated by Ivan a.k.a. Lonliboy.

Rêveries is a new and technically complex series. Orkhan had the idea to run a sort of AI podcast at the start of the space to break it down. I’ve listened to it—it’s a brilliant intro and conceptually ties in nicely to Orkhan’s broader practice, which examines the role of AI in society and its impact on our culture and psyche. So, this feels like the perfect way to begin the conversation. After that, we’ll talk to Orkhan directly and hear more about the series in his own words.

AI 1: Welcome to the deep dive. This is where we take a whole load of information and boil it down into key insights—essential takeaways, aha moments, without all the noise.

AI 2: Today, we're diving into the work of media artist Orkhan Mammadov and his generative neural painting series, Rêveries. We’ll explore the art itself and the ideas and tech behind it.

AI 1: So our mission is to unpack how Orkhan uses AI—not just as a tool but almost like a collaborator.

AI 2: Exactly. The art explores memory, texture, and the whole evolving relationship between us and machines. Let’s start with the core concept. Orkhan repurposes neural networks—typically used for tasks like image classification and finding patterns—and uses them to generate these really rich painterly textures.

AI 1: Right. And wasn't there a specific paper that sparked this?

AI 2: Yes, the 2015 paper “Texture Synthesis Using Convolutional Neural Networks” by Leon A. Gatys, Alexander S. Ecker, and Matthias Bethge. It was foundational. So instead of using this tech to analyze things, Orkhan wondered: what if we used it to create things instead? And conceptually, he links this to the idea of dreamscapes—where forms dissolve into visual thoughts. The result is neither purely human nor purely algorithmic either, it’s something new.

AI 1: Definitely. And Deep Dream by Google was an early influence. It opened up the idea that neural networks could generate rather than just classify.

AI 2: And Rêveries isn’t actually Orkhan’s first work with AI art. Back in 2016, his studio—the Orkhan Media Art Lab—was already experimenting. They even wrote custom neural code to generate carpet textures. So this isn't new ground for him, Rêveries is more of a return to that space and original model, refined with better tech and experience. It's an evolution.

AI 1: So beyond the tech, what’s the bigger picture here—what’s the conceptual goal?

AI 2: He's using these computational tools very consciously to explore our relationship with culture, heritage, collective memory, very human things. He’s not just showcasing cool AI outputs—he’s probing how AI affects us psychologically and culturally.

He sees art as a meta-discipline that brings together philosophy, tech, and critique. And he is inspired by people like, you know, Nam June Paik, Jody, even theorists like McLuhan and Manovich. Mm-hmm. People who were thinking about media's impact way back. It's part of a longer conversation about technology and society.

AI 1: And didn't he show AI work at the Venice Biennale too?

AI 2: He did, yes. At the 2019 Venice Biennale, he represented Azerbaijan with Circular Repetition, his first major AI installation. That work also explored cultural heritage through generative techniques.

AI 1: So Rêveries builds on that, right? Let's get into the specifics then of how the texture synthesis actually works, the nuts and bolts.

AI 2: So Orkhan uses convolutional neural networks (CNNs), which as we mentioned are typically used to recognize features like edges or patterns. Instead of asking "what is this?", he trains the network to create an internal map of the feature space and generate new textures by capturing the stats, the eternal relationships between these features across different layers of the network.

AI 1: So it's not just copying, it's learning the, the essence, the rules of the texture, and then generating something new based on those rules.

AI 2: That's a great way to put it. It learns the statistical essence, and as the process goes deeper through the network layers, recognizable, shapes fade out, leaving just this distilled texture. So the algorithm is his tool, but it's also like a window into how the AI itself represents visual stuff. There's almost an emerging quality there peeking into the machine's mind. As the process goes deeper into the network, recognizable shapes fade, leaving behind distilled textures.

AI 1: So it learns the essence of the texture and creates something new—not a copy, but a generative imitation. And there’s also movement—Rêveries isn’t static?

AI 2: Right. That’s where it gets dynamic. It goes beyond just making textures to actually bringing them to life. Making them evolve using feedback loops powered by GLSL shaders—fast, real-time programs that simulate evolving textures. Imagine it starts with a generated image and a depth map that tells it about distance from that it calculates a vector field, basically. Arrows showing direction and speed of flow everywhere in the image, like currents. Then it uses masks like stencils to control how different areas blend and get pushed around, and the image's pixels get shifted along those currents.

It actually does the shifting in a a couple of slightly different ways. One sharper, one smoother, and blends them. Then crucially, it reintroduces bits of the original image and the whole thing loops back. Starts again. So it's constantly generating, shifting, blending, feeding back, creating this continuous transformation. The source provided called it a 'visual ecology', as it really does capture that feeling of things flowing, fragmenting, reforming. And because it's using those fast GLSL filters it happens right there in front of you. You're literally watching the machine, well, dream. You're watching a process unfold, not just seeing the end result.

AI 1: So, okay. What does Mammadov want us, the audience to get from this? What's the intended engagement?

AI 2: He wants it to be a space for contemplation, to make you think about how machines are synthesizing memory, texture, form, these really fundamental things. It's meant to spark wonder, maybe some introspection about what seeing even means now with AI and how we square, you know, natural textures with these sometimes weird AI generated versions. Bridging our perception with the AI's.

AI 1: And the sources listed some really interesting questions this work raises, really deep questions like are there parallels between how these networks work and how our own memory or thinking works? Or how does that feedback loop compare to say, organic growth? Or even our own stream of consciousness. Big questions. And then the idea of a machine subconscious or dream life, is that what we're seeing?

AI 2: It pushes you to ask that. Definitely. And then there are the more technical, artistic questions, like how crucial is that feature space idea for generative art? And does seeing it happen in real time change the experience compared to just a static AI image? It must.

And then placing it historically. How does it echo, say, Impression trying to capture fleeting moments, or how does it connect to generative art pioneers like Harold Cohen, or even abstract painters like Richter or Rauschenberg, but in a post-digital way.

AI 1: And that huge question about cultural memory. How is heritage being reinterpreted through code here? Could this become a whole new visual language? It's a lot to chew on.

AI 2: Okay. Let's circle back to Venice 2019, Orkhan's Circular Repetition piece and how it helps us understand Rêveries.  Circular Repetition used traditional Islamic geometric patterns as its input data. Thousands of them. And the AI generated this endless shifting stream of new decorative forms, projected, immersive. It clearly shows his early interest in using AI to engage with and transform cultural heritage.

That theme is still strong in Rêveries, just maybe more abstract. And there was another piece there too, Muraqqa. Miniatures. Based on Safavid dynasty miniatures. But he added subtle animation to make you think about fake news. Basically how images, even historical ones can be manipulated, how truth gets distorted.

AI 1: Ah, so even then he was exploring that line between the real, the historical, and the technologically mediated or altered.

AI 2: Absolutely. That tension between ancient craft and machine creativity, questioning our perceptions. It's a consistent thread. He seems really fascinated by how our memories personal and collective interact with machine intelligence and how that blurs the real and the simulated. He wants us to think about how fragile our knowledge systems are and how tech can both preserve and warp the past.

AI 1: And he is also building tools for others, right? genie.art?

AI 2: Yes. genie.art is described as a conversational node-based IDE. So giving artists more control over these complex AI workflows. Node-based, like visual programming, making it more accessible perhaps. And he is also working on ethical licensing for museums using AI. So he is not just making the art, he's thinking about the whole ecosystem around AI and culture, really considering the infrastructure, the ethics, not just the output.

AI 1: Okay. So wrapping up our deep dive on Orkhan's Rêveries, what are the big takeaways?

AI 2: I'd say first the really innovative way he uses neural networks for texture, synthesis, and simulation. That dynamic aspect is key, right? Second, the deep exploration of what AI aesthetics mean. How they connect to or challenge your own ways of seeing. And third, just the sheer number of profound questions it raises about art, tech, memory, perception. What it means to see today. Maybe a final thought to leave you with as AI like this develops its own ways of well expressing itself, like in Rêveries, how does that change our own understanding of creativity or memory or even consciousness? What does it mean if machines can, in some sense, dream in a ways we're only starting to grasp?

IZ: Wow. Thank you so much, Orkhan. Getting that kind of overview and dive with AI is really phenomenal and I think it's a really special tie into the series as a whole and your focus with neural networks stemming back all the way to 2016.

Now that we've had that kickoff with two AI speaking to each other, I'd love to hear you summarise Rêveries in your own words and then perhaps to relate it to your practice more widely.

OM: Thank you Ivan. Yes, it's quite, uh, scary right to listening to AIs talk.

IZ: Oh, they speak perfectly. It's one of the most seamless conversations I've heard.

OM: Yeah, it was fascinating for me. So, back in 2016 when I kind of first started to learn and add this, uh, methodology to my creative practice, we found the Textual Synthesis paper. And I was also fascinated by the DeepDream by Google. As I kind of researched, like the first implementations of AI, it was CNNs called the conventional neural networks. As you heard before, the core idea of that was, analyzing the vast data sets and the finding correlations to similar patterns, recognition of objects or common, let's say the patterns among datasets among images. Me and my team, we kind of reverse engineered that textual synthesis and, uh, using this feature space, kind of like finding the more repeated pattern or more unique, more distinguished pattern among the given image. We just multiplied and created a new, uh, let's say learn from that pattern and correlation. We created new, uh, iterations. We additionally added GLSL shaders, which custom written, uh, takes all information from image and turns it into this, uh, fluid simulation. That's it.

IZ: So not only with the GLSL shaders that, you know, they're, they're pushing and pulling everything, right? Being able to give that dynamic nature, that everlasting movement to it, that's running in real time. I love the fact that it pulls back the original base image from the synthesis model in the first place. Was that an idea that you always had from the start, or is that something that came to you during the process of implementing the generative system?

OM: If you look to my other artworks, I'm never stopping with the initial creation or initial generation. For me, the data is always a source and then I use, or generate any kind of, uh, image or video using AI. I always want to add second, third layers on top of it. Imagine you are taking a photo with a camera and there's a raw image output. It needs a color correction. It's needs cropping. So in our context, uh, I always love to animate the paintings using the data of the painting or the initial source image.

So in Rêveries as in my other collections, like a Muraqqa, it is also the same. We have a initial source image, and we build on top of it a simulation that uses the image's information.

IZ: So in a sense it's kind of like a collage within a collage within a collage, and it keeps building up to become the final form. But the final form never stops for Rêveries, it keeps going on forever.

How critical would it be for viewers to understand these technical foundations of convolutional neural networks in order to engage meaningfully with the series?

OM: It's very straightforward, I would say. Let's imagine there's a narrow network model or an AI model that is, uh, designed to analyze, and finds several items in the picture, either a pattern, or an object. These CNN networks could train with several artworks, as I said, to extract information from that. Um, this simple algorithm is just a reverse. Instead of like analyzing an image, it is, analyzing and finding a pattern to create new iterations of.

Overall, I don't think it's too much necessary to know what that actually means. I would like explain this process or workflow with simple words like, um, generative neural data paintings. So there's a paintings that generated from neural networks, hidden layers and, uh, animated with the generative algorithms.

IZ: See, I love that simplification where you call these pieces paintings when they're created through code and live systems. Could you dive into a bit more of your perspective of what a painting is and you know, in this era now where we're so technologically ingrained, and how that perspective may have changed, especially since 2016 with the work that you were doing then - would you consider those outputs paintings or is that something that has developed through time now?

OM: I would say develops in the time. I wouldn't call myself as an artist in 2016, my whole ideology and, uh, let's say the practice formulated under one statement or under my vision around 2019 when I represented my country in the Venice Biennale. So before that, I will call myself a researcher, an experimenter.

IZ: And a lot of your work really stems back to cultural heritage. So within Rêveries, in what ways do your cultural roots shape your use of pattern, texture, and memory with a neural network as the collaborator?

OM: It's an amazing question. Thank you. So, um, as many you know, from art history, there's several chapters in artists' creative careers. For me, researching my own cultural heritage and roots was very crucial in the beginning of my career because it was not all about becoming an artist. It was also the reason I became an artist, because I asked a lot of questions to myself and needed to find an answer.

During this five years that I experimented with my own cultural heritage, I learned so many things about my country, my people, my ancestors, but as an artist and as a builder, I cannot stop this creative urge inside to research about nature. I have one collaborator, which always works with me and our team, he's like obsessed with nature. He's obsessed with the natural patterns, how nature builds this procedural, uh, systems.

I'm fascinated by memory the how the brains works, how we perceive information, how we get manipulated by the several factors, by algorithms or by the media. So many research, in the next five years I will love to dedicate to memory and neuroscience and, uh, more societal aspects of the usage of AI and this changing shift in the technologies in our daily lives.

IZ: Are there any memories in your life that you feel you've ingrained within the series, and do you see it as a kind of self portrait in the sense of instilling your personal essence, memories and guidance within the work?

OM: Interesting. There is aspect of self-portrait, with Rêveries as the starting point of new chapter, kind of remembering my first, let's say sketch or idea around AI. AI started with deep dreams, if I'm not wrong, and has evolved now to advanced algorithms that we all use. So not self-portrait, but visualizing of kind of the journey, I would say, with inspiration from 2016 and with the knowledge from 2025. Rêveries stands to that research from 2016, but it is made with the current technologies that we have now.

IZ: With Rêveries we are building up movement that is dynamic, ongoing, forever, what distinguishes simulation from representation in the context of these generative textures?

OM: Uh, so when you take the raw output, it asks 'please make me something' like 'animate me, make me alive'. When we generate image, the algorithm also generates a tape map, also creates the vector fields. Everything is actually data to use and enhance and make it alive. Just the raw output is kind of soulless, I would say. It's like your brain tries to mimic or imitate, tries to understand what's under this pattern because it learns from the pattern of the image and, tries to construct a new pattern.

So when a user is observing the image, it's trying to reconstruct in the brain that what it looks like. But, because the all output is like a dream space - it's not like a dreamy, like style-GAN algorithms, I would say - but it's more, uh, you are in, as you said, in the moment, you feel like some part of the memories from painting. So just keeping it raw and without any activation on top of it, it will not deliver its emotion to the viewers.

IZ: So this animation actually helps you try to build emotional connection with an image? So we're being thrown into the river and then the entire time the river's endless and we can see to the right and left of everything, the mountains changing. We see everything changing through time, but the experience is the same. We're still in the river and it's everlasting and it's ongoing. I think that's absolutely beautiful and it's a sensation that I feel when viewing every piece from Rêveries. Tying a bit more into these base textures that are created by the neural network, would you say that the series continues the lineage of the Renaissance era, of Neoclassicism, and Impressionism?

OM: Yep. So there was a question first. What will be the data source? I didn't want to create an abstract, random noise pattern that I will use in the animating of it. So, we carefully, hand created, let's say, the core image dataset, from the 10,000s of the library. Each one is uniquely choosen to represent the right color palette and right compositional aesthetic.

And why I chose this data set - as you know, before I was using mostly Orientalist paintings because the Orientalism paintings created about the Middle East without ever visiting the Middle East by the European painters, uh, mostly French, Dutch and then the German, uh, Italians. So, the idea of that paintings were kind mimicking the memory of unvisited lands. But as I updated or opened the new dimensions for my arts practice, uh, I choose this time the mythology from the biblical images, for more storytelling. This vibrant imagery, is not just the figurative images, but tells the stories. So I was hoping that these stories also will somehow translate into the my paintings.

HW: I might jump in for a second here, because that was a question I was very interested in when we first started talking about this series. The base images and why you chose them. And I think it's interesting that you use classical images rather than historical or let's say non-fiction leaning, because with these dreamscapes you are storytelling and creating something that isn't tangible. Does that relate to the dream scapes you're creating, how humans have imagined these stories throughout history and now the machine is using its own imagination while transforming these images?

OM: Yes. I fully agree with you, you amazingly explained the process, but also I would love to add that as we are integrating AI to our daily life, and it becomes a part of our life that we cannot deny anymore, we're sharing this cultural context among like a collective memory among us, and it's interesting for me to understand how we are sharing this memories with AI and, once we see our cultural memories reinterpreted or imagined as dreamscapes by AI, can we build or can our memories mimic or remember.

HW: I also love the idea of how different fragments of the textures are recycled and fed back into the system, almost like how when we dream, we take fragments of images that we've seen in our lives and our minds stitch them together in new compositions in the same way that the machine is operating here. I suppose that's not really a question, but I find that quite interesting.

IZ: To jump into that, I've been thinking about dreaming a lot lately because of Rêveries, and for the last few months I haven't been able to remember dreams too well, but I've been taking note of kind of the experience of dreams and the experience of waking up afterwards. There's always the sensation that the dream carries into the day a little bit, and then sometimes you'll get these flashbacks to the dream, like echoes. The more I've been thinking about it, the more I try to stew in that feeling when I wake up, and the longer that I spend diving into what you've created here, it's impacting my life personally. So I wanna say thank you Orkhan for that.

Secondly, I wanted to ask, do you think machines are capable of creating beauty or is it always our own reflection that we recognize through their output?

OM: Hell yes, definitely. But, uh, alone machines will not go great. So there will be always a human factor, of course. There will be the AI agents, bots, etc. But whatever you do, they are all controlled by the human, there is a human factor. Machines are nothing without humans, and it'll be always like this. There is no way that it'll replace that. We just need to adapt and understand how we can make our process workflow and everything better.

Why I dedicated to almost half of, more than half of my life to building the Genie (genie.art) is that all companies are trying to create an ideal software that with one click you can create, and it'll be visually appealing, but then it becomes a really AI slop, without soul. But as artists, we deserve much more control over the AI, um, that we fully could express our feelings, our emotions, our ideas, not just tied to the limitation of the prompt, but could go deeper to set the parameters and play with the parameters of that AI.

IZ: You know, you've pretty much ended any sense of doomer-ism I felt towards AI with what you've just said. Changing lanes slightly, are there any hidden or subtle things in these works that only you as the creator would notice?

OM: Mm, no. There's like a golden rule in our studio. If the world is burning, if we are not satisfied fully with a result, we are not shipping it, like ever. I have canceled so many collaborations, so many projects because I don't feel it.

We are not building application that's 'oh, let's run the second update, third update, fourth update'. The artwork that we create is something that's forever, once you're done, it's forever. So for that reason, we spend a lot of time on the creation, for perfect in the outputs, for the code. We spend almost the same amount of the time to create a presentation of the artwork to creating a teaser, the visual narrative, the presentation that shows the whole story around that work.

HW: So I have a question less about Rêveries, I suppose, but more about you as an artist. When you're working so much with machines and pattern recognition, and have done so long, does it in any way feed into the way that your brain works? Do you become better at recognizing patterns or more method methodical in your thinking?

OM: Yeah, so when I was young, when I first started to create and design websites around like 11 years old, I don't know if it's like OCD or like a neurological disorder, but my nickname was a pixel Orkhan. So I was obsessed with, uh, pixel perfect layouts, I was like measuring everything. It had to be golden ratio. The amount of space between the letters need to be mathematically perfect. Um, so it's kinda my lifestyle or my brain how it works. All of my projects are structured, all my, um, exhibitions, planning, everything is run by the production company structure. I'm coming from the more advertising studios, design studios, uh, or enterprise companies, and I learned a lot of like management skills, how to build things and but also opening your world with the ideas. And, it's hard to find the right people to collaborate with this. And, it created this, uh, trust in all process that they understand you fully, or you create things not just the for the sake of creating, but you are creating based on the divisions that you share together.

So, um, back to your question, yes, like it's just the mindset. I had an education in computer science, later in visual communication design and, and finally of fine arts. So through this three, the discipline that I studied, by time I just start to combine them and create the intersections, took items from one and just combine to apply to another one.

So I would say yes, my life is works like AI sometimes. I have a friend, he says to me that you're literally cyber work. There's AI inside you. I don't believe you are a full of human.

HW: I love that. A lot of personality traits feed into each other, to be quite perfectionistic is quite in tune with how a lot of artists operate. And when you were saying earlier, you know, before 2019 you would describe yourself more as a researcher and experimenter, those also go hand in hand with creating artistically. That's 90% of what an artist does is experimenting and looking at new ways to articulate what you have to say. And that was one of the first things I noticed when we started speaking, was how organized you are.

IZ: As a curator, I must say it's absolutely thrilling because there's just so much context that you bring to everything that you make. Finally, I want to ask what feels more intense to you, trading neural networks and implementing generative systems from scratch, or the rollercoaster of the digital art movement the last four years in Web3?

OM: Please guys, don't cancel me. But the only thing is the level of degen-ism in this space. Like, it's hilarious. I worked as a design director in the leading bank in my country. I was the technologist advisor of the president administration in my country. I was graphic designer, product designer, like I worked with so many clients. The only place here makes me cancer with dealing with this. Being a patient of like how we want to be rich in the one second, you know, how we want to change our life. How everything's, the attention is so small here. Like you do one thing and tomorrow no remembers that. This is the only thing that really annoys me.

But this space is, we are so early, like incredible early and we will evolve so much. We just need corporations or the communities who will build this bridge between the artists and the right communities. Because this space puts such big responsibilities on the shoulders of the artist. That what is your floor price? Why it's like this, why you need to this, oh you need to burn your supply. There is a artist whose work is not selling for like six, seven years and works all around the fairs, and in the end it finds the right collector. So this space is a little bit not understanding the long-term narrative. But of course there is a big majority of people who considers this as very temporary, not thinking about the long term.

I have a friend who once told me to imagine that you are a person in the sixties in New York and he's making the printmaking, you know? And you need to understand the scale of the revolution that we are in and how the world is gonna change in the next few years as an in the digital art context. You just need to be patient, build, create, collaborate, collect and, and support each other instead of gatekeeping. So we are not big millions of community. We are like maximum around like a few thousand people per year. So people online just want that being friendly, and just creating together. Patience, passion and focus. That's all it's about work on.

IZ: I definitely agree. Sometimes the degen-ism in the space, it feels like a JG Wentworth commercial. You know, 'I want my money and I want it now'. And that's just not what makes it thrilling, it's being a part of it all. So I appreciate your perspective on everything.

HW: Just before we wrap up, Orkhan are there any final bits and pieces that we haven't covered yet that you'd want to share about the series?

OM: So the series is hopefully going to be interactive. This even I was hiding from the Verse team. You can play with artwork with the mouse. You will have several states to display the source image and additionally the animation. Each artwork is incredibly high resolution for printing. There will be the blueprints of how you can create, uh, the physicals of these paintings, but just to focus on the beauty, choose your most favorite artwork.

HW: Just to recap, Rêveries will be coming out next week on the 21st. It'll be a 200 piece series, and there will be 250 works to select those 200 from. Thank you so much everyone for joining us. Thank you especially to Orkhan for bringing us these works. I've really enjoyed this conversation and I hope the same goes everybody else.

OM: Thank you everyone for your time and thank you for amazing hospitality. It's a pleasure to work with you guys.