#pixel science

[harshansai3dandvfx]

Pixel Science : 4

Reflection and Evaluation:

Through this exercise, I learned that photorealism is not achieved through a single technical choice but through the careful balance of resolution, colour depth, file format, lighting, and post-processing. High-quality formats such as EXR and PNG preserve realism, but even these can be undermined by poor lighting or excessive processing. Conversely, heavy compression, low resolution, and reduced colour depth quickly reveal the digital structure of an image. This experiment reinforced the importance of understanding pixel science as a foundation for believable VFX work.

Bibliography :

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

Blender Foundation (2024) Blender Documentation: Rendering and Compositing. https://docs.blender.org

Adobe (2024) Photoshop User Guide. https://helpx.adobe.com/photoshop

[harshansai3dandvfx]

Pixel Science : 3

6. JPEG, 1920 × 1080 p, 8-bit colour depth, RGB, Quality 60%, Pixelate Sampling 0.900

#image6

This render explores pixel degradation using Blender’s compositor. After rendering at full resolution, I used scale and pixelate nodes to reduce sampling quality before exporting as a JPEG.

Although the resolution remains high, the pixelation introduces visible softness and uneven pixel structure. This image demonstrates how post-processing alone can degrade realism, even when resolution and composition remain unchanged.

#image7

This screenshot documents the compositor node setup used to create #image6. It shows how scaling and pixelation nodes were applied to manipulate the image after rendering.

Including this process image helps demonstrate technical understanding and transparency in workflow, showing how pixel science can be explored non-destructively through node-based compositing.

7. JPEG, 1920 × 1080 p, 8-bit colour depth, RGB, Quality 60%, Pixelate Sampling 0.100.

#image8

In this render, I further reduced the pixel sampling value in the compositor. This extreme pixelation causes individual pixels to become clearly visible, removing fine detail and smooth transitions.

At this stage, the image is no longer photoreal. The pixel grid becomes dominant, and the scene reads as a digital abstraction rather than a physical space. This demonstrates the lower boundary of visual credibility.

#image9

This screenshot shows the compositor setup used to generate the posterisation and pixel manipulation effects in later renders. It documents the technical steps used to intentionally reduce realism.

8. PNG, 1920 × 1080 p, 16-bit colour depth, RGBA, Compression 15%, Overexposure.

#image10

This render explores how lighting alone can break realism. By overexposing the lights in Blender while maintaining a high-quality PNG output, highlight detail is lost and surfaces appear unnaturally bright.

Although the file format and resolution remain technically strong, incorrect lighting immediately reduces believability. This demonstrates that photorealism depends as much on lighting balance as on pixel fidelity.

9. PNG, 1920 × 1080 p, 16-bit colour depth, RGBA, Compression 30%, Posterize Effect.

#image11

In this render, I applied a posterize effect using Blender’s compositor. Posterisation reduces the number of tonal values, replacing smooth gradients with harsh colour steps.

This significantly impacts realism, as natural lighting relies on continuous tonal variation. The image begins to resemble a stylised illustration rather than a physically plausible render.

10. PNG, 1920 × 1080 p, 16-bit colour depth, RGBA, Greyscale Bitmap Effect

#image12

The final image was processed in Photoshop using a bitmap and greyscale conversion. This removes colour information entirely and reduces the image to stark tonal contrasts.

At this point, realism is completely lost. While the composition remains readable, the image no longer attempts to replicate real-world vision. This final render clearly exposes the digital nature of the image.

[harshansai3dandvfx]

Pixel Science: 2

1. EXR, 1920 × 1080 p, 16-bit colour depth, RGBA.

#image1

The first render was exported as an OpenEXR file at full HD resolution with 16-bit colour depth. EXR is a high-dynamic-range, lossless file format widely used in professional VFX pipelines because it stores extensive pixel data across multiple channels. This allows subtle lighting information, reflections, and shadow detail to be preserved without compression artefacts.

In Blender, I set the output format to OpenEXR (Multilayer), increased the colour depth to 16-bit, and rendered using Cycles. The resulting image appears highly realistic, with smooth gradients, accurate material response, and stable lighting. This render represents the most photoreal version of the scene and acts as the visual reference point for the following tests.

2. PNG, 1920 × 1080 p, 16-bit colour depth, RGBA.

The second render was exported as a PNG file while maintaining the same resolution and colour depth. PNG is a lossless format, meaning it does not introduce compression artefacts, but it does not store the same extended dynamic range as EXR.

#image2

This render was created by changing only the output file format in Blender while keeping lighting and render settings identical. When compared to #image1, the image still appears photoreal, although it contains less flexible colour information for post-production. Visually, the difference is subtle, showing that high-quality PNG files can still maintain realism for final images.

3. JPEG, 960 × 540 p, 8-bit colour depth, RGB, Half Resolution, Quality 100%.

In the third render, I reduced the resolution to half HD and exported the image as a JPEG at maximum quality. JPEG is a lossy compression format that reduces file size by discarding pixel data.

#image3

This image was rendered by lowering the resolution in Blender and switching the output format to JPEG with quality set to 100%. While the image still reads as believable at a glance, fine details begin to soften, and subtle gradients are less smooth. This demonstrates how resolution and bit depth play a significant role in maintaining realism, even when compression quality is high.

4. TIFF, 1920 × 1080 p, 8-bit colour depth, RGBA, Deflate Compression (No Noise).

#image 4

The fourth render was exported as a TIFF file using lossless Deflate compression. TIFF is commonly used in professional workflows because it preserves image quality while allowing moderate compression.

In Blender, I kept the full resolution but reduced the colour depth to 8-bit. The image remains clean and visually convincing, but compared to earlier 16-bit renders, subtle banding can begin to appear in smoother lighting gradients. This highlights how colour depth affects tonal continuity, even when compression artefacts are absent.

5. JPEG, 256 × 144 p, 8-bit colour depth, RGB, Quality 90%.

#image5

For this render, I significantly reduced the resolution to 256 × 144 pixels and exported the image as a JPEG at 90% quality. This was achieved by lowering the resolution in Blender before rendering.

At this stage, realism begins to noticeably degrade. Edges appear blurred, surface detail is lost, and the image no longer holds up under close inspection. When compared to #image3, the impact of extreme resolution reduction is clear, demonstrating that spatial resolution is one of the fastest ways to break the illusion of realism.

[harshansai3dandvfx]

Assignment 1 – Pixel Science : 1

Photorealism and Pixel Science:

Photorealism in pixel science refers to the ability of a digital image to convincingly replicate the visual characteristics of the real world through pixels, colour, light, and spatial resolution. Rather than relying purely on artistic skill, photorealism is built on technical decisions such as colour depth, file format, compression, sampling, and lighting accuracy. Even small changes in these parameters can determine whether an image feels believable or immediately artificial.

For this assignment, I explored photorealism through a practical workflow in Blender. I used multiple Suzanne (monkey) models, each assigned different shaders, to observe how materials, light interaction, and pixel data contribute to realism. I rendered ten still images while systematically altering resolution, file formats, colour depth, compression, and post-processing using Blender’s render settings, compositor, and limited Photoshop adjustments. This process allowed me to analyse where the threshold between photoreal and unconvincing imagery begins to break.

[chandrups-developingskill]

#Blog 2 : Pixel Science

Exploring photorealism through image variations,

For this project, I explored the concept of photorealism by creating ten manipulated versions of a single photograph that I originally captured in Kerala, India. I used Adobe Photoshop to alter elements such as colour, lighting, resolution, sharpness, and filters. The purpose of this exercise was to understand how digital changes influence the perceived realism of an image and to identify the point at which an image begins to look artificial rather than photographic. Through this process, I realised that photorealism is not about making images look overly impressive or exaggerated. Instead, it focuses on accuracy, attention to detail, and subtle visual decisions. Achieving photorealism requires a balance between technical precision and restraint, where choices are made carefully rather than overtly. It is this careful balance that allows photorealism to be effective, ensuring that every element feels considered and accurate.

------------------------------------------------------------------------------

Original Image : The original photograph was used as the reference point for all variations. It represents the scene as it appeared in real life, with balanced colours, natural lighting, and appropriate contrast. This image acted as a baseline to compare how each manipulation affected realism.

Version 1 – White Balance Adjustment : In this version, I shifted the white balance towards a cooler tone. As a result, the image developed a strong blue cast and appeared unnatural. This demonstrated that accurate colour temperature is essential in maintaining realism, as incorrect white balance can entirely change an image from its original environment.

Version 2 – Overexposure : By significantly increasing the exposure, the image lost detail in the highlights and appeared washed out. This experiment showed that correct exposure is crucial for realism, as excessive brightness removes depth and visual information from the image.

Version 3 – Low Resolution / Pixelation : For this variation, I reduced the resolution of the image and then scaled it back to its original size. This caused visible pixelation and a loss of sharpness. The outcome highlighted the importance of image resolution and pixel density, as realism relies heavily on the presence of fine detail.

Version 4 – Colour Correction : Using colour balance, curves, and sky correction, I introduced a more stylised and dreamlike appearance. While this enhanced the mood of the image, it also reduced its realism. This version helped me understand that colour grading can be expressive, but excessive adjustments can make an image feel artificial. Subtle colour correction is more effective when aiming for photorealism.

Version 5 – Background Blur : In this version, I used the Camera Raw filter to isolate the subject and apply background blur. This created a shallow depth-of-field effect similar to that produced by professional camera lenses. The result enhanced realism by guiding focus naturally without overprocessing the image.

Version 6 – Stained Glass Filter : Applying the stained glass filter simplified the image into block-like sections, flattening shadows and highlights. This clearly demonstrated how smooth tonal transitions and colour depth are essential for photorealism, as heavy stylisation quickly removes photographic authenticity.

Version 7 – Noise and Grain : Here, I added visible noise to simulate the look of an older camera. While the texture added an artistic quality, it significantly reduced realism. This experiment showed that noise can shift an image from realistic to expressive when it is applied excessively.

Version 8 – Incorrect Lighting : In this variation, I intentionally added mismatched lighting and shadows using the brush tool. The resulting image appeared unnatural and visually confusing. This confirmed that realistic lighting depends on consistent light sources and accurate shadow placement.

Version 9 – Excessive Sharpness : By increasing sharpness beyond normal levels using the Unsharp Mask filter, the image developed harsh edges and unnatural textures. Fine details, such as water surfaces, began to look artificial. This showed that moderate sharpening can improve clarity, but excessive sharpening damages realism.

Version 10 – Posterization : The posterize filter reduced the number of colours in the image, resulting in flat shadows and highlights. This gave the image a graphic, almost cartoon-like appearance. This version of the image showed that smooth colour transitions are really important if you want the image to look like a photo. The posterize filter and its effect, on colours made this very clear. Through these ten variations, I learned that photorealism is achieved through a precise balance of light, colour, detail, and subtle imperfections. While controlled adjustments such as depth of field and slight colour correction can enhance realism, excessive digital effects can ruin it. This project improved my understanding of how digital tools influence visual perception and clarified the importance of these variation when it is aiming to a create realistic imagery.

[selinasharmapractice1]

Assignment 1 – Pixel Science

My Tests + Results

The first assignment was a short exercise investigating what goes into making an image photoreal, and where we cross that threshold. From my previous research into the different pixel science techniques, I used it to inform my practice with this assignment.

Taking a pre-existing scene in Maya where I explored materials, I trialled different techniques, rendered in Arnold, and documented the result. The following stills explore file output; resolution; colour space; lighting; sampling; and camera.

1. EXR, 1920 x 1080 p (HD), 8bit colour management, samples 3/2/2/2/2/2, focal length 35:

Figure 1

Figure 1 displays what happens when you try to open a .exr file. EXRs are a special file that are multi-pass and multi-channel, meaning that render passes and channels are stored in the file format itself. It is also lossless in compression. This makes the file format best for photorealistic renders and compositing, which makes sense as it was designed by visual effects artists at Industrial Light and Magic to help with their pipelines (Elwyn, 2023). With the mass of information stored within the file, it cannot just be opened in an image viewer, as it is designed to be opened in an editing or compositing package.

Figure 2

Figure 2 is a screenshot of the opened EXR file in the software ‘Krita’. The results of the EXR render are a very crisp render, free of artefacts. However, one point of note is that the colours appear washed out in the EXR format, as opposed to the render view in Arnold. This is because the colour information in an EXR is linear, with no gamma correction; while PNG, JPEG, and TIFF are typically sRGB, with gamma correction applied on monitors. This means that EXRs should be gamma corrected.

Another point to keep in mind is that this screenshot may not be a true representation of the EXR as the screenshot itself is a PNG file, even though PNGs are usually lossless in compression.

The sampling was put at 3/2/2/2/2/2 for camera (AA)/diffuse/specular/transmission/SSS/volume indirect (see Figure 3). Sampling will influence shot noise. Figure 4 shows the EXR image zoomed in. The quality is still acceptable. However, there is a small amount of noise in the transparent object. This could be improved by increasing the samples slightly.

Figure 3

Figure 4

The image file is 32,427KB and the render time was 3 minutes and 59 seconds.

2. JPEG, 960 x 540 p, 8bit colour management, samples 3/2/2/2/2/2, focal length 35:

Figure 5

Figure 6

The JPEG format has compression loss. Furthermore, 960 x 540 p is half HD. The results in Figure 5 are fuzziness and noise in the objects, particularly the glassy transparent materials. The reflections in these objects are not as clear. However, the colours are true to the render view in Arnold.

Overall, the lower quality of this image is enough to break illusion of the glassy objects being real. Furthermore, the colour correction pre-applied to this render oversaturates the colours, which also breaks photorealism. Although, the image gives a good overview into what a render should look like, without having to wait for long render times, as this image took 22 seconds at the lower quality settings. Its file size is also much smaller, being 229KB. Therefore, the JPEG format and half resolution is best when sharing updates on a scene because it is quick and efficient, but it is not good for final outputs and renders. This is because the low quality will be even more obvious and noticeable on a big screen; this can instantly take an audience out of an immersive scene. We can especially see this demonstrated when we zoom into the image (Figure 6).

Figure 7

TIFFs can be compressed with and without loss, have no compression, and have a colour range up to 32bpc. This makes TIFFs very powerful. The focus of this render was to see how sampling affects renders. Here, we can see that setting every sample value to ‘1’ has caused the image to have noise and fireflies everywhere in the scene. With the TIFF file format, 16bit colour management and HD resolution, this render should be clear and rich with colour once we target the noise.

To target noise, I isolated the AOVs in the render view. The diffuse, specular and transmission appeared to be where the noise was particularly coming from (see Figures 8, 9 and 10).

Figure 8 – Diffuse, samples 1

Figure 9 – Specular, samples 1

Figure 10 – Transmission, samples 1

The file size is 16,201KB and the render time was 5 seconds. In the following test, you can see the results of increasing sampling for these AOVs.

4. TIFF, 1920 x 1080 p,16bit colour management, compression none, colour space RAW, samples 3/3/3/3/2/2, focal length 35:

Figure 11

The results in Figure 11 are very clear and the reflections are sharp. By strategically targeting the AOVs, I increased samples where necessary. There is hardly any noise. The overall image is darker than the render view in Arnold, like was the case with the EXR result. This is due to TIFFs having a vast amount of colour information like the EXR. Furthermore, increasing colour management to 16bit provides more colour in the scene that was not scene in the Arnold render view, nor in the JPEG render (see Figure 12 and 13).

Figure 12

Figure 13

Figures 14, 15 and 16 illustrate the results of increasing the samples for diffuse, specular and transmission (refer to Figures 8, 9 and 10 to see the results before this). These AOVs were each raised to 3 samples:

Figure 14 – Diffuse, samples 3

Figure 15 – Specular, samples 3

Figure 16 – Transmission, samples 3

The results demonstrate the importance of sampling on renders. The file size is the same as the previous render, 16,201KB; but the render time was 3 minutes and 29 seconds, unlike the 5 seconds of the previous render. This suggests that increasing samples will not affect you file size, but it will increase your render time. Therefore, it is important to only raise samples where necessary.

5. PNG,1920 x 1080 p, 16bit colour management, samples 3/3/3/3/2/2, focal length 12:

Figure 17

The PNG file format at HD and higher samples has retained a clear look. The colours look the same as the Arnold render view, like with the JPEG, so the colour range is not as vast as EXRs or TIFFs. The PNG format is lossless in its compression. The decreased focal length zooms the camera out dramatically and warps the backdrop. This demonstrates how a camera in the same position can warp perspective by playing with the focal length. This can make camera movements more dynamic and interesting rather than keeping the focal length at the standard 35.

The file size is 5,800KB and the render time was 1 minute. The file size is much smaller than the EXR and TIFF, and renders at a fraction of the time. It can be a good option for final render output because of its high quality and lossless compression. However, its colour depth is not as broad as the TIFF and EXR format, which makes those the better option for compositing.

6. TIFF, 1920 x 1080 p, 16bit colour management, samples 3/3/3/3/2/2, focal length 35, exposure of AiSkydomeLight:

Figure 18

As the results of test 4 seemed to be the best outcome so far, I decided to keep the next few tests the same in its settings with slight tweaks to observe the effects of different factors. In the previous tests, each render used a plain AiSkydomeLight with an exposure of 1. With Figure 18, I increased the exposure to 3. Figure 18 shows the results, which are blown out. The translucent objects have completely been lost in the light; the light emitting objects are still visible because they are not absorbing light; and the plastic material is only half visible still as the underneath is not absorbing light while the top is.

The size of the file is 16,201KB and the render time was 3 minutes and 21 seconds. The results suggest that exposure will not affect file size and will hardly affect the render time. Exposure will easily fill an entire scene, which is why you should not go too high with it or else you will blow out elements in the scene.

7. TIFF, 1920 x 1080 p,16bit colour management, samples 3/3/3/3/2/2, focal length 35, HDRI:

Adding a HDRI image to an AiSkydomeLight can transform the light and colours in a scene., changing the way a scene looks. Using the HDRI in Figure 19, Figure 20 shows the results. The HDRI has filled the scene with an overall cool tone. It is very saturated which detracts from photorealism.

Figure 19 - Wronkowski, G. (no date) Modern Evening Street HDRI • Poly Haven, Poly Haven. Available at: https://polyhaven.com/a/modern_evening_street (Accessed: 10 October 2025).

Figure 20

Figure 21

Figure 21 is a close-up of the results. There is notably grain across the render. We can observe that the information of the HDRI can cause noise in renders.

The file size is still 16,201KB, and the render time was 3 minutes and 45 seconds. Like with exposure, it appears that HDRIs do not change file size and hardly affect render times. However, with the added noise from the HDRI, samples will need to be increased. As samples are increased, render time will be increased.

8. TIFF, 1920 x 1080 p,16bit colour management, samples 4/9/8/3/2/2, focal length 35, HDRI:

Figure 22

Targeting the AOVs as I did with previously with test 3 and 4, I increased the samples to 9 for diffuse, 8 for specular, and 3 for transmission. There is a very slight difference in the results, with a small decrease in noise. However, there is still some despite the very high sampling, as observed in Figure 23:

Figure 23

From observing Figures 24, 25 and 26, we can see that diffuse is still causing noise even with samples being increased to 9. Furthermore, transmission could have had samples increased slightly more to remove the noise of the transparent ball.

Figure 24 – Diffuse, samples 9

Figure 25 – Specular, samples 8

Figure 26 – Transmission, samples 3

While sampling made a slight difference with the noise problem and does not add to the size of the file (still at 16,201KB), the render time was 34 minutes and 37 seconds. This is unreasonably long to observe a small difference, taking 10 times longer than previously.

A better way to tackle this should have been to increase the samples on the AiSkydomeLight itself.

9. TIFF, 1920 x 1080 p,16bit colour management, samples 3/3/3/3/2/2, focal length 35, no HDRI and no AiSkydomeLight:

Observing the effect of HDRIs and light again, I removed the AiSkydomeLight from the scene, only keeping the area lights so that we see the objects.

Figure 27

With no HDRI, there is hardly any grain in the scene. The area lights doing the lighting work also empathises the highlights and shadows in the objects. As a result, Figure 27 can pass as being photoreal.

The file size is 20,091KB and the render time was 3 minutes and 45 seconds. When comparing to test 4’s results, we can observe that lower lighting hardly affects render time, but it does increase file size.  

10. TIFF, 320 x 240 p, 16bit colour management, samples 3/3/3/3/2/2, focal length 35:

Figure 28

Figure 28 has all the same settings as test 4, except it is at a much lower resolution than any of the other renders. 320 x 240 p is an outdated resolution that has a 4:3 aspect ratio rather than the HD 16:9. The lower resolution means that there are fewer pixels to display the image, therefore the render is fuzzy and grainy. This is not a good output format as of 2025, when we now have 2k, 4k, and 8k resolutions.

The file size is 601KB and the render time was 10 seconds. The low file size makes resolution renders good for sending and sharing, so it can be best for sharing updates on a shot. However, I would recommend a JPEG format and half HD at the very least (as seen in test 2) as the file size will be even smaller and the quality will be more acceptable.

Evaluation + Reflection

The results highlight that TIFFs at a minimum resolution of 1920 x 1080 p are the best for achieving photorealism, apart from EXRs that are the best for compositing because they store all render passes in the file. However, samples, lighting and HDRIs can affect the believability. Samples can affect noise in a shot, which will be especially noticeable on a big screen. Exposure, if set too high, can blow out objects within a scene, making readability difficult. HDRIs can make a scene more photoreal in some instances, but they can also make a scene oversaturated in other instances, such as in test 7 and 8.

This exercise has made me more conscious of the technical skills behind what makes renders photoreal. This is something I will be testing more with my future projects.

Bibliography

Autodesk (no date). Removing Noise – Arnold for Maya. Available at: https://help.autodesk.com/view/ARNOL/ENU/?guid=arnold_for_maya_tutorials_am_Removing_Noise_html (Accessed: 10 October 2025).

Camber Film School (2019). Video Aspect Ratio Explained – How Different Aspect Ratios Affect Your Video Style [video]. YouTube. Available at: https://www.youtube.com/watch?v=E-dUveGINTg (Accessed: 11 October 2025).

ExplainingComputers(2022). Explaining Image File Formats [video]. Available at: https://www.youtube.com/watch?v=WblPwVq9KnU (Accessed: 10 October 2025).

Goulekas, K. (2021) ‘Acquisition/Shooting’, in Okun, J.A. and Zwerman, S. (eds.) The VES Handbook of Visual Effects: Industry Standard VFX Practices and Procedures. 3rd ed. New York: Routledge, (2) pp. 106-118.

Stump, D. Ollstein, M. Reisner, D. and Wall, W. T. (2021) ‘Acquisition/Shooting, in Okun, J.A. and Zwerman, S. (eds.) The VES Handbook of Visual Effects: Industry Standard VFX Practices and Procedures. 3rd ed. New York: Routledge, (2) pp. 188-212.

Wronkowski, G. (no date) Modern Evening Street HDRI • Poly Haven, Poly Haven. Available at: https://polyhaven.com/a/modern_evening_street (Accessed: 7 October 2025).

[selinasharmapractice1]

Pixel Science – Theory and Technical Research

What is Pixel Science in VFX – and why does it matter?

While art and creativity are often considered a separate thing to technology, the two are 'closely related' (J. and P. Setabundhu, 2018, p.1). This is especially the case with visual effects, where pixel science is just as important as skill and design.  In film and TV, visual effects and special effects are used in production. While they are both effects, there are distinct differences. Michael Fink defines visual effects (VFX) as 'any imagery created, altered, or enhanced […] that cannot be accomplished during live-action shooting (2021, p.1). Meanwhile, special effects 'can be done while a live-action scene is being captured' (ibid., p.2).

VFX help to enhance storytelling in film. This is because it can be used to create what is impractical to recreate on a live set; it can create shots that would be dangerous in real life; and it allows us to create things that do not exist (Fink, 2021, p.3). When watching films that use VFX, viewers often critique them for using "too much CGI". Just typing 'marvel cgi' into google will bring up 'Marvel's CGI is insultingly bad', Reddit thread, as the first search result (AWuTangName, 2022). This is most likely due to the VFX being poorly executed (see Figure 1), as Fink suggests that visual effects should not be obvious if 'done well' (2021, p.3). He explains further that they 'further the story being told' to 'immerse [viewers]’ (ibid.).

Figure 1 - Fenn, A. (2024) Marvel VFX Insider Explains Controversial CGI Decline in the MCU, MovieWeb. Available at: https://movieweb.com/marvel-vfx-insider-cgi-decline-mcu/ (Accessed: 10 October 2025).

There are multiple reasons why VFX can look disconnected from a film. Scott Squires suggests that one of the most important aspects for 'successful' visual effects is the design (2021, p.22). This is true, as a poorly designed visual effect will completely break the immersion for the audience. However, what is often overlooked is the understanding of pixel science. This is a technical understanding of the techniques that effect the look of digital imagery outputs displayed on screens. While the designing aspect is so important, the efforts will be wasted if the technical outputs are ignored. Therefore, knowledge of the pixel science techniques behind visual effects is essential for seamless integration into digital worlds. These techniques should be utilised where necessary to improve the look and believability of visual effects. As VFX artists, the techniques should be chosen carefully and strategically as they will impact the 'final results' and the 'difficulty getting there' (Squires, 2021, p.34).

Techniques that affect Pixel Science

When making photoreal imagery in VFX, the 'realism will be adjusted to make the images cinematic' (Squires, 2021, p.31). Each individual shot will have different problems to solve in order to get the desired outcome, therefore the approach and techniques will differ to suit the requirements of the shot. Squires suggests that we should 'balance the best technique for each shot with the efficiencies of sequence techniques' (2021, p.32). It is imperative for the successful VFX artists to know what techniques are necessary for a shot's requirements as a good framework. However, Squires highlights that 'if you can do something simple that provides the effect, then it probably does not matter that it is technically correct' (2021, p.32). This idea is applicable if what you are doing is a faster workflow. However, if what you are doing increases time spent on projects, such as render time being increased unnecessarily, then technical practices should not be ignored. As J. and P. Setabundhu suggest: 'creativity always works in tandem with technology and skills' (2018, p.4)

One challenge that Karen Goukelas highlights is matching the outdoor and indoor lighting to make the visual effects fit seamlessly into a shot (2021, p.118). Lighting is an incredibly important characteristic to achieve photorealism, as we observe light in our everyday life. When lighting is not true to life, such as over-exposure or lack of shadows, this can break immersion. This is what can often make AI art recognisable as it is 'a-contextual' (Motoarca, 2024, p.515), as it is a simulacrum and a system responding to commands (ibid., pp.506-507) without live human experience. The lighting will often be unrealistic and nonsensical, making the generated work uncanny.  Light can also affect the "shot noise", which is the 'noise in the blacks of the images' (Stump, Ollstein, Reisner and Wall, 2021, p.190). Noise can be tackled digitally with sampling (Autodesk). Goukelas suggests the method of using a chrome ball and a matte ball to see how the materials and objects react to light (2021, p.118). She also suggests the use of HDRIs, which are 'high dynamic range imaging', experimenting with the exposure (ibid.). She argues that this technique is better than the simple chrome ball approach as it 'greatly improves the ability to recreate photorealism and accurately light visual effects elements in a scene' (ibid.).

Output resolution is also necessary to understand when creating visuals that will be shown on screens. These can range from 144p to 8K, but the most common for film and TV today are high definition (HD) that is 1920x1080 pixels, and ultra-high definition (UHD) that is 3840x2160 pixels. These make a drastic difference to the way visuals will look, as the lowest resolution will be blurred while the highest resolution will capture even the smallest detail. Additionally, frame rate can affect resolution, as the 'maximum resolution capability of a camera usually gets reduced at higher frame rates' (Stump, Ollstein, Reisner and Wall, 2021, p.207)

Frame rate is an important factor, as it 'affects the portrayal of motion' in its 'smoothness and sharpness' (ibid.). Standard frame rates in film and TV are 24fps, 25fps and 30fps. Lower frame rates have artifacts that appear, such as motion blur, judder, strobing and flicker (ibid., p.208). This can provide a more cinematic effect. Meanwhile, higher frame rates reduce these artifacts, displaying smooth and sharp motion (ibid.). This can make shots appear more photoreal. The digital camera itself has many attributes that each affect the output of an image as well, such as its focal length, focal distances, f-stop, T-stop, and depth of field (ibid. p.207). 

Colour space is the spectrum of colours seen by the human eye (Stump, Ollstein, Reisner and Wall, 2021, p.194). The subset of those colours that can be represented by a given colour space’s mathematical encoding is its “ gamut" (ibid., p.195). The colours represented on screens are defined by red, green and blue (RGB). When working between mediums and technologies, colour spaces can be lost. To retain colour information, we can use outputs like RAW. The RAW output uses one-third of the storage that RGB values would (ibid., p.201). Therefore, it can be worth using the RAW output for colour space for optimisation.

Evaluation

Core techniques that can affect the photorealism of an image include; lighting; resolution; frame rate and camera; and colour space. These are factors that I will be conscious about when aiming to create photoreal visuals. With this informed research, I will be applying it to my practical CG work.

Bibliography

Autodesk (no date). Removing Noise – Arnold for Maya. Available at: https://help.autodesk.com/view/ARNOL/ENU/?guid=arnold_for_maya_tutorials_am_Removing_Noise_html (Accessed: 10 October 2025).

AWuTangName (2023) ‘Marvel CGI is insultingly bad’, r/marvelstudios. Available at: https://www.reddit.com/r/marvelstudios/comments/10s8v51/marvel_cgi_is_insultingly_bad/ (Accessed: 9 October 2025).

Fenn, A. (2024) Marvel VFX Insider Explains Controversial CGI Decline in the MCU, MovieWeb. Available at: https://movieweb.com/marvel-vfx-insider-cgi-decline-mcu/ (Accessed: 10 October 2025)

Fink, M. (2021) ‘Introduction’, in Okun, J.A. and Zwerman, S. (eds.) The VES Handbook of Visual Effects: Industry Standard VFX Practices and Procedures. 3rd ed. New York: Routledge, (1) pp. 1–3.

Goulekas, K. (2021) 'Acquisition/Shooting', in Okun, J. A. and Zwerman, S. (eds.) The VES Handbook of Visual Effects: Industry Standard VFX Practices and Procedures. 3rd ed. New York: Routledge, (2) pp. 106-118.

Motoarca, I. R. (2023–2024) ‘AI, Copyright, and Pseudo Art’, Yale Journal of Law and Technology, 26, pp. 430–526.

Nanosys (2021) ‘What is a Pixel? An Introduction to Displays’ [video]. Available at: https://www.youtube.com/watch?v=a8hfUlPvta0 (Accessed: 9 October 2025).

Setabundhu, J. and Setabundhu, P. (2018) ‘Cathy Berberian in the Sea of the Uncanny: How Science Inspires Our Art’, 2018 International Conference on Engineering, Applied Sciences, and Technology (ICEAST), Phuket, Thailand, 4–7 July 2018. Piscataway, NJ: IEEE, pp. 1–4. doi: 10.1109/ICEAST.2018.8434427.

Squires, S. (2021) ‘Pre-Production/Preparation’, in Okun, J.A. and Zwerman, S. (eds.) The VES Handbook of Visual Effects: Industry Standard VFX Practices and Procedures. 3rd ed. New York: Routledge, (2) pp. 5-31.

Squires, S. (2021) ‘Acquisition/Shooting, in Okun, J.A. and Zwerman, S. (eds.) The VES Handbook of Visual Effects: Industry Standard VFX Practices and Procedures. 3rd ed. New York: Routledge, (2) pp. 61-71.

Stump, D. Ollstein, M. Reisner, D. and Wall, W. T. (2021) ‘Acquisition/Shooting, in Okun, J.A. and Zwerman, S. (eds.) The VES Handbook of Visual Effects: Industry Standard VFX Practices and Procedures. 3rd ed. New York: Routledge, (2) pp. 188-212.