#neutropia

[selinasharmapractice1]

Taking one of the assignments further - Neutropia (part 5): Exploration with AI to take concepts further (attempt 2)

In my previous AI exploration, I made reference to copyrighted IPs in my text prompt which the Netflix AI guidelines (2025) note is something to be avoided to prevent potential plagiarism. Further to this, ChatGPT has since updated their image generator. I decided to take the AI exploration with this project further to suit the Netflix guidelines and to test the capabilities of the new image generator.

Using the same prompt, I removed the references to the copyrighted IP. This time the results were less dramatic in its colour grading - this suggests that it may have previously heavily referenced the IPs that I had mentioned for the grading. These results are better creatively for the grading process as it is relatively flat in colours, so it can be given to a colour grader to creatively alter the image.

The new image generator is better at photorealism and does not appear to have artefacting - the words in the image are comprehensive and legible, including even on the signs and the vending machines in the back. The previous image generator had some artefacting with the letters and signs, and with the assets themselves. This is seen a lot less in the results of this new ChatGPT image generator.

Figure 1 - first output

The first output followed my render closer than the previous outputs, but it forgot to have doors for the barriers and made the station platform too wet. While I think it would be wet because of the weather conditions and river stream, I imagine drains would prevent this to make the platform safer. I asked ChatGPT to make the floor slightly less wet, turn the windows into sliding doors, and to move the boat closer to the platform so that passengers can logically get on/off. I also asked that there be a bucket where the cones are to signal leaks to the viewer as this was an asset the AI missed from my render.

Figure 2 - second output

The second output completely removed the wet floor rather than making it 'slightly less wet' - this highlights how the newer AI model still struggles with control (like making specific changes). It also removed my wet floor sign and cone assets from the scene completely, recontextualising the image despite me not requesting this. It did however, move the boat closer and even allowed room on the platform for when the doors open for passengers. It also included the bucket I requested, but it created a weird asset in the roof for the leak to come specifically out of. This logically does not make sense. It is clear that, while artefacting is less common and photorealism is more achievable, the AI still lacks control and reasonable judgement, which requires human input.

I requesting that the cones and wet floor sign be brought back into the scene and that the floor around that area be wet. I also requested that the strange where the leak was coming from to be removed.

Figure 3 - third output

The third output successfully brought the requested assets back and removed the unwanted asset. Overall, it has done a good job at following the brief, albeit a few iterations were required. Even still, this was a speedier process than traditional texturing and rendering.

Reflection and Evaluation

The issue of control is still present with ChatGPT's newest image generator. It makes creative decisions itself to remove assets, change positioning of assets, or create new ones. This makes art direction difficult to manage when the AI cannot stay consistent or follow simple change requests. However, for post-vis or look-dev, this is a decent workflow to get an idea of how final shots could look. Using them for final outputs can be problematic as a result because it will struggle with consistency between shots if it cannot be consistent within one shot. Furthermore, where the AI gets it information is a concern as it has made this station 'Northwood' in the image despite this in fact not being Northwood station. There was no mention of Northwood in the prompt, yet the AI made the decision that it would be. This highlights the practical issues of using AI for final outputs, despite its technical advancements in photorealism and preventing artecating.

[selinasharmapractice1]

Taking one of the assignments further - Neutropia (part 4): Exploration with AI to take concepts further

My time with the project began to run slim for texturing, lighting, rendering, compositing and grading. Thinking back on Andy McNamara's talk on 'Generative AI and The Future of VFX' (2026), I remembered how he suggested using AI as a tool for visualising what untextured renders could look like. Since this is the stage I have reached with the project (Figure 1), I wanted to take this as an opportunity to explore the capabilities for AI currently to understand how it can be used in a workflow (not to replace workflows or to be used as final outputs).

Figure 1 - raw render of my scene

I started by writing a prompt in ChatGPT to explain the art direction for this scene. The prompt was as follows: "Taking inspiration from the mood boards attached, I would like this plain render to be visualised into a station 100 years in the future where, instead of train stations in the UK, we have 'boat' stations', an adapted mode of transport that has become the norm due to extreme climate change. The weather in the UK 100 years from now is severe rain, flooding what used to be train stations. I have modelled and laid out a 'boat station', where the 'boats' are on water. They mechanically work in a similar way to how water rides work at theme parks. They move along a stream of water, where tracks would have been when it was a train station before. There are barriers with sliding doors that only open when the 'boats' have arrived. Take inspiration from how Elizabeth Line stations like Tottenham Court Road have done this in the current day to inform the barriers and sliding doors in this image. Have adverts on the barriers in sensible places that reflect the rainy society 100 years from now. Some examples could include: insurance for weather damage; waterproof sneakers; designer rain coats and umbrellas. The adverts should be futuristic as the society is 100 years from now, so imagine holographic imagery. In the back of the station, have vending machines for umbrellas, rain coats and rain boots. The cones are surrounding a leak in the station. There is a wet floor sign to bring caution and awareness to this. The inside of the station should overall look futuristic (but in a mundane way rather than a sci-fi/apocalyptic way), while the outside of the station should look worn down slightly for contrast. This can be done by having wear and tear to the brick wall outside, greenery growing as a result of rainfall, and leak stains. The image should aim for photorealism in style, looking indistinguishable from a photo. The outside should be foggy/misty and gloomy. There should be rain hitting the glass barriers and the river stream. The inside of the station should be futuristic yet bland, having a mundane approach. The atmosphere should take inspiration from films like 'The Batman' and 'Blade Runner 2049' in colour grading and in tone."

The mood board I am referring to here is the one I created during the planning stage (in part 2).

Figure 2 is the result:

Figure 2 - ChatGPT's interpretation

The atmosphere and style is what I was envisioning; however, the logic is not perfect. The AI has created rain in the station which does not make sense. I asked ChatGPT to amend this. Figure 3 is the result:

Figure 3 - ChatGPT's amendment

Figure 3 is better logically now. However, there are still issues with this image:

the 'boat' is not a safe or sensible distance away from the platform

it moved where I placed the cones in the scene

my seat design has been changed

When I asked it to make small changes to fix these issues, the entire image would alter slightly (Figure 4). This demonstrates the lack of control currently with Generative AI.

Figure 4 - the result of asking ChatGPT to move the boat closer to the platform barrier

Overall, it has done a good job interpreting the prompt but it is difficult defining details and having control of the details. Artistic direction becomes difficult to control. This is why AI is best used to brainstorm and visualise rather than to use as final output.

To take visualisation further, I took the output and asked Sora 2 to simply 'Make it rain outside'. Sora 2 was able to do this, although it completely removed the water stream outside and had a train run inside the station which is completely nonsensical (Figure 5).

Figure 5 - screengrab of Sora 2's first attempt

This time, I specified to have the train be outside (al

Figure 6 - Sora 2's second attempt

Reflection

I approached using GenAI following the guidelines of Netflix and the teachings of McNamara. However, I realised in hindsight that I did forget Netflix’s guidelines makes it clear that reference to other IPs in prompts should be avoided.

Bibliography

McNamara, A. (2026) Generative AI and the Future of VFX. Lecture delivered online, 11 March, 14:30.

Netflix (2025) 'Using Generative AI in Content Production'. Netflix | Partner Help Center. Available at: https://partnerhelp.netflixstudios.com/hc/en-us/articles/43393929218323-Using-Generative-AI-in-Content-Production (Accessed: 1 April 2026).

[selinasharmapractice1]

Taking one of the assignments further - Neutropia (part 3): 3D Modelling

After planning the scene, I began blocking out the layout of it and where assets will go.

First, I blocked out where the platform barricade doors would go (Figure 1 and 2). I started with a cube that I scaled up and positioned to be at the edge of the platform as a safety procedure.

Figure 1

Figure 2

To make the scene feal photoreal, scale is important. Therefore, I put a human reference into the scene as a scaling reference (Figure 3 and 4). I scaled the human reference to match the humans in the original photographic plate (Figure 5). The human reference was useful to then plan the size of the benches which I could block out with cubes. I then cut edgeloops planning where the doors and panels would go, before extracting the faces for the doors to make them separate objects (Figure 4).

Figure 3

Figure 4

Figure 5

I then examined the scene in perspective view so get an idea of how the whole scene looks as a whole outside of the shotcam (Figure 6 and 7).

Figure 6

Figure 7

Back in the shotcam view, I examined the scene in a wireframe view to see the tracks behind the barriers against the scene (Figure 8).

Figure 8

Happy with the block out so far, I made a new maya scene to build the assets in a clean environment. I imported the human reference so that the assets could have a scale reference as I model.

As I was making the models, I made a moodboard to get ideas of what 'futuristic design' could potentially look like taking inspiration from what designs currently exist (Figure 9). From here, I began modelling.

Figure 9

Figure 10 - bench process

Figure 11 - bench

Having finished the bench, I began modelling the contextual assets that will fill the scene: a traffic cone for areas of leakage; a bucket to catch leaks; a caution sign for wet floors; a vending machine that supplies umbrellas, raincoats and rain boots; a drain; an umbrella; an an umbrella helmet which I thought would be adapted into everyday life to deal with the weather conditions like how face masks were worn by society during the COVID pandemic (Figures 12-16).

Figure 12 - cone

Figure 13

Figure 14

Figure 15

Figure 16

Once I was happy with the assets, I brought them back into my scene and placed them throughout (Figure 17 and 18).

Figure 17

Figure 18

Next, I UV mapped my assets to prepare them for the texturing stage (Figure 19).

Figure 19

Before moving onto texturing, it is good to light the scene so you can assess the textures against the scene lighting. I created mesh lights in the station and had an skydome for the outside (Figure 20).

Figure 20

Reflection

I enjoyed modelling the assets and populating the scene in this way. The next stages would be to texture the assets in the scene, create narrative lighting, and colour grade for atmosphere.

[selinasharmapractice1]

Taking one of the assignments further - Neutropia (part 2): Mood-boarding and Planning

To help steer the project in a cinematic and believable direction, I made moodboards of both film references and real life references to base my inspiration on tangible existing things. Collating such references can help ground the project to a society we are familiar with rather than going too futuristic and going off of sci-fi tropes.

Figure 1 - Overall look at the moodboard

On the left in Figure 1, I have made a moodboard putting real life references and inspiration for the project. These references will help the project to feel based on reality. On the right in Figure 1, this is the moodboard for creative projects that I can take inspiration from: stylistically, contextually and compositionally. These references will be a reminder to make my project visually interesting.

Figure 2 - LINE Mobility (Art&Graft, 2026) on the left; Heavy Rain (Quantic Dream, 2010) on the right.

The LINE Mobility project (Art&Graft, 2026) presents a unique new system of travel, where people travel in pods along tracks, being a combination of the car and the train. Studying how they approached presenting a new method of travel can help with my approach in this project that combines boats and trains as a method of transportation.

Heavy Rain (Quantic Dream, 2010) is a good reference for rainy conditions and rain patterns. Rain going in different directions, falling at different speeds and different sizes will all contribute to believable rainfall.

Figure 3 - Blade Runner 2049 (Villeneuve, 2017)

Blade Runner 2049 (Villeneuve, 2017) is a great cinematic reference for a rainy environment set in the future. The mood and atmosphere set by the weather conditions adds to the storytelling. Further to this, the way they think about lighting in this film is a great reference on a technical level and a contextual one too. The lights would be bright and colourful to be seen through foggy and misty conditions, and the light would travel through the fog/mist creating a halation effect. These are effects I would like to implement in my project.

Figure 4 - The Batman (Reeves, 2022)

The Batman (Reeves, 2022) provides great reference for rainy conditions and it is used atmospherically.

Figure 5 - Death Stranding 2 (Kojima Productions, 2025)

I really liked the rain on the camera lens in Death Stranding 2 (Kojima, 2025). This heightens photorealism as well as immersion for a viewer as it is mimicking how a camera lens would pick up rainy conditions. This is something I would like to implement in my own project if I do a camera angle that is sat outside of the train platform.

Furthermore, the greenery resulting from rainfall is something to remember to implement in my work where it makes sense and it plausible, like an outdoor brick wall for example.

Figure 6 - Photorealistic flooded subway (Pepe Baze, 2021)

This artist Pepe Baze is a perfect example of a photorealistic space that has been flooded presented in a cinematic way. I will be drawing heavily from this artist's work in this project.

Figure 7 - Roller coasters with water

These water rides have a mechanical system where boats are carried along water streams. This could be useful when thinking about how the actual boat system would work mechanically.

Figure 8 - Primary photography

Figure 8 is a photo I took on a misty morning which provides reference of how light looks through a camera lens in these conditions: bright, exaggerated, halation. This will be a good reference when trying to get a photorealistic look.

Figure 9 - Flooded stations

I imagined the boat system replacing train stations from flooding of the tracks creating bodies of water. This is something that is already happening, especially in the UK. When searching for references on Google Images, simply searching 'flooded stations' brings up examples that are clearly in the UK, and even local examples that are in Hertfordshire. These examples provide a grounded reference to base the project off of.

Figure 10 - Rainy conditions paraphernalia

Smaller details that tell the story of an extreme rainy environment would be drains to prevent flooding, wet floor signs, wet/muddy floors indoors, leak buckets, advertisements capitalising off of the rainy society etc.

Some advert ideas that can be scattered around the environment subtly to contribute to the environment's storytelling:

charities for families affected by the weather conditions

shoe adverts boasting 'anti-slip' or 'anti-wet' technology

fashion adverts promoting designer umbrellas or raincoats

'TFL Boats' (company name I am assigning to the travel system) encouraging premium 'first class boat' upgrades.

vending machines with umbrellas, raincoats and rain boots.

dehumidifier adverts

Figure 11 - Train platforms with sliding doors

Finally, to prevent people swimming in the water streams and to prevent drunk people from falling in and drowning, the stations will have these sliding doors to ensure safety, as well as to prevent water splashing onto the platforms and causing slip hazards.

Evaluation

Moodboarding has helped me to think about the reasoning behind my decision making. It has also helped to give me a better idea of the cinematic look I am going for in this project. My next blog posts will make a start on the project, continuing on from where I previously left the Neutropia project.

Bibliography

Art&Graft (2026) LINE Mobility. Available at: https://artandgraft.com (Accessed: 1 April 2026).

Baze, P. (2021) Photorealistic flooded subway. ArtStation. Available at: https://www.artstation.com/artwork/1835we (Accessed: 1 April 2026).

Kojima Productions (2025) Death Stranding 2 [video game]. Tokyo: Kojima Productions.

Quantic Dream (2010) Heavy Rain [video game]. PlayStation 3. Paris: Quantic Dream.

Reeves, M. (2022) The Batman [film]. Burbank, CA: Warner Bros.

Villeneuve, D. (2017) Blade Runner 2049 [film]. Los Angeles, CA: Warner Bros.

[selinasharmapractice1]

Taking one of the assignments further - Neutropia (part 1)

Reflection on Neutropia

Having completed all of the assignments and reflecting on them, I will be taking Neutropia further. After having a tutorial with Saint, he recommended thinking more about the cultural logic behind images. For example, he suggested taking inspiration from stations along the Elizabeth Line that use barriers to prevent people falling/jumping onto the tracks. He also suggested a hierarchy boat system (like how we have first class and standard, for example) where different boats can represent social status. Finally, thinking about the small everyday mundane things can add so much to the storytelling of an image.

Figure 1 - how I left the Neutropia project last

Due to time restrictions, I will look into kitbashing different types of boats using pre-existing assets. This way, I can still make something using my creative vision while also saving on time.

My plans for this scene so far:

kitbash a boat in 3D.

add more greenery to the brick wall to signal to the viewer that the weather conditions have promoted unintentional growth.

Add barriers in 3D to the platform to provide logic for preventing people jumping/falling into the water stream.

Have the inside of the station (waiting platform) appear maintained in comparison to the outside of the station.

raincoat/umbrella/rainboots vending machine perhaps.

Adverts promoting gadgets or charities to help with the extreme weather to provide context of societal needs.

buckets placed where there are leaks through the roof.

A couple of other stations to show a variety of what stations look like in this society.

Bright lights and signs to help people see through the fog and mist (potential light pollution in the sky as a result).

Grey overcast and mist

Potentially rain simulation

Some things I can take inspiration from to help guide the project:

Rollercoasters with water (Figure 2)

Art&Graft's Line Mobility project (2026) (Figure 3, which I got involved with during work experience with them)

Heavy Rain (Quantic Dream, 2010) (Figure 4)

Elizabeth Line stations with platforms.

Figure 2 - an example of a ride with water

Figure 3 - Line Mobility (Art&Graft, 2026)

Figure 4 - Heavy Rain (Quantic Dreams, 2010)

In the next blog post, I will begin mood-boarding and planning further.

Bibliography

Art&Graft (2026) LINE Mobility. Available at: https://artandgraft.com (Accessed: 1 April 2026).

Quantic Dream (2010) Heavy Rain [video game]. PlayStation 3. Paris: Quantic Dream.

[harshansai3dandvfx]

Neutropia 3 : Reflection: What I Learned

This assignment taught me that believable futurism comes from both technical integration and conceptual consistency. Technically, I learned that camera matching is not a quick step that can be approximated; it is the foundation that controls scale, realism, and projection accuracy. I also learned that lighting is a major realism trigger, and that even small mismatches in exposure and shadow softness can break the illusion faster than low-detail modelling. Conceptually, I learned that “futuristic” is not a style it is a system. If I want Neutropia to feel real, I need to design the future with rules, reference real sustainable design developments, and build a coherent visual language before I open Blender.

How I Will Improve Neutropia:

To improve this project, I would begin by collecting targeted references for sustainable campus design, EV infrastructure, solar integration, and accessible public technology, then extract a design rule set from those references before modelling anything. I would redo the camera matching more carefully by validating the fSpy alignment using simple proxy geometry early, checking scale with real-world measurements, and ensuring the verticals and horizon line remain consistent. After that, I would plan the additions as a connected system rather than separate objects—for example, positioning solar elements where they logically receive light, designing mobility platforms around real circulation paths, and making the technology feel embedded into architecture. Finally, I would treat the composite as part of the design process by matching colour and exposure intentionally, using projection and material tweaks only once the camera and lighting fundamentals are secure. In short, my next version of Neutropia would be less about adding “future objects” and more about designing a future that feels inevitable, functional, and visually consistent.

Bibilography:

Kitbash Assets : https://kitbash3d.com/pages/cargo

Blender Foundation (2024) Blender Manual: Rendering, Compositing and Colour Management. https://docs.blender.org/manual/en/latest/

fSpy (n.d.) fSpy: Still Image Camera Matching. https://fspy.io/

Okun, J.A. and Zwerman, S. (eds.) (2020) The VES Handbook of Visual Effects: Industry Standard VFX Practices and Procedures.

[harshansai3dandvfx]

Neutropia Process : 2

Lighting and HDRI (Trying to Match the Plate) :

Once the basic forms were in place, I began working on lighting, and this is where I felt the gap between “I know the steps” and “I can get the result.” I introduced an HDRI to create a natural base light, then adjusted the intensity and rotation to match the directionality of the original plate. My aim was to make the 3D additions inherit the same daylight logic as the photo soft shadows, consistent contrast, and a believable brightness range. However, my first lighting pass looked too clean and CG-like. The shadows did not feel embedded in the environment, and the highlights on objects felt slightly disconnected from the real-world exposure of the plate. This forced me to slow down and treat the lighting like a matching task rather than a “nice render” task. I began tweaking until the light direction and shadow softness felt closer, and I also learned that even minor mismatches in exposure make the composite immediately suspicious. This step was challenging, but it made me more aware that realism is often lost through lighting mismatch before anyone even notices modelling detail.

Adding Models Using Camera Projection :

After establishing a more stable camera and lighting direction, I explored camera projection to integrate textures and details more efficiently. I approached this as a practical VFX technique: instead of fully modelling and texturing every surface, camera projection lets me “borrow” the photographic realism of the plate and project it onto geometry that matches the scene. When I tested this, I understood why projection is considered powerful in production workflows. It allows fast, photoreal detail because the source is real-world information, and it can help sell complex surfaces without building everything from scratch. At the same time, I also noticed the limitation: projection is dependent on camera angle, and it can break if the geometry deviates from the plate perspective. This was another learning point—projection is not a shortcut that replaces fundamentals, but it is a method that rewards accurate camera matching and careful modelling. In my case, it helped certain elements sit more convincingly, but it also exposed where my perspective alignment still needed improvement.

After Image, What I Wanted, and Why I Still Felt I Failed:

When I brought the elements together into the “after” version, I could see progress but I still did not feel I achieved the Neutropia vision I had in my head. The additions were present, but the scene did not fully transform into a believable future. The main reason I consider it a partial failure is that the overall image still reads like a present-day building with a few futuristic props placed on top, rather than a campus designed around future systems. I also noticed my confidence dropped during this stage, because the more I tried to force the concept, the more it felt like I was losing creativity instead of gaining it. However, reflecting on it honestly, I think the issue was not “no creativity,” but a lack of structure in my concept development. I started building too soon, before I had a clear design language, references, and a consistent rule set for how this future would look and function.

[harshansai3dandvfx]

Assignment 2: Neutropia - Concepts and Ideas

For this assignment, I wanted to move beyond simply making something “look futuristic” and instead build a concept that felt believable and socially grounded. The core idea I started with was a world set roughly a hundred years from now where humans have deliberately reshaped everyday life to be genuinely eco-friendly rather than performatively sustainable. In my mind, Neutropia is not a dystopia; it is a future that tries to repair damage through practical design choices products built to last, transport systems shifting to EV infrastructure, public spaces supported by solar power, and technology integrated into the environment in a way that feels normal rather than flashy. Alongside this, I also wanted the world to feel futuristic in a way that still reads as possible, where the “future” comes from small systems and services that change how people move, access information, and live day-to-day. These two directions - eco-resolution and believable futurism—became my anchor when I started planning the image work.

Attempt and Failed (First Plan) :

I began with a real photographic plate of a university building exterior, because I wanted my final image to sit in that border between photography and VFX, where the audience might accept it as real if the integration is strong. My first step was to attempt camera matching using fSpy. I imported the photo into fSpy and tried to align the perspective lines to the building geometry, setting the vanishing points and exporting the camera data for Blender. After importing the fSpy file into Blender, I quickly realised my initial match was not as accurate as I thought. The camera alignment looked “close enough” in isolation, but when I started placing blockout geometry, the scale relationships didn’t sit properly and the verticals began to drift. This is where the first failure happened: I was pushing forward without validating the match, and the errors became more obvious the more I built on top of them. Even though it felt frustrating, it was useful because it showed me that believable realism starts with the camera, not the models.

Blocking Out in Blender :

Block-out

After the camera match attempt, I moved into a simple blockout stage in Blender to test whether my concept could sit naturally within the scene. I kept the geometry basic at first and focused on forms that communicate “future infrastructure” without overwhelming the plate. I experimented with adding ideas like an outdoor lift system, solar-powered elements, Wi-Fi or signal modules, and an automated moving platform concept, treating them like design interventions added to a current-day campus. While doing this, I noticed that my design choices were initially too random; they looked like disconnected props rather than part of a coherent future system. That was another small “failure” moment for me because I realised the concept needs rules. I started rethinking the objects as part of a single logic—energy capture, accessibility, and mobility—so they read like they belong to the same future city. Even at block out level, I could see that photoreal integration is not only about rendering; it is also about whether the design language makes sense inside the world.