As my final blog post, I thought I would reflect on my initial blog post and how I have grown over the semester. As I have stated multiple times throughout the semester, going into this subject, I had never really used AI out of fear of the unknown. Now, I feel much more comfortable using AI. I still feel guilty using it for written assignments, but for creative work I find it to be a helpful companion.

In my first blog post I said I wanted to learn to “weaponize (AI) to enhance my work”. Whilst I don’t know if I did this exactly, I have definitely found uses of AI I will take beyond the classroom. Firstly, I love generative AI SFX. Yes, they require quite a bit of mixing to make them sound authentic, but it is so convenient to generate SFX rather than seeking out copyright free SFX on YouTube or recording my own. ElevenLabs also generates multiple SFX from the same prompt, giving more options to be mixed together, creating a richer soundscape. I also haven’t had to re-prompt ElevenLabs many times at all, making it efficient to use. I also have been using ChatGPT more since taking this subject. I find it most helpful for brainstorming, even if it is ideas of what not to do or just as a jumping off point.

I have been surprised how much nuanced conversation there is around AI, especially art which is my thesis topic. Coming into this semester, I would have flat out said AI generated art is not art. However, I have challenged my thinking throughout the semester, and now I really don’t know where I stand with the topic. I think it is a case-by-case basis given the breadth in application of AI in art, and differing degrees of its involvement. I really don’t think there is a definitive answer of AI being or not being art.

It has been really cool to see how the entire class has embraced AI over the semester. This growth is evident in the final projects produced. I really enjoyed Ally and Lily’s music video, as the use of an AI song not only was a creative use of AI, but a relevant topic discussed earlier in the semester. I also really enjoyed Chuling, Ethan and Saffron’s short film. The way they edited AI footage into the piece was really impressive, and their soundscape was rich and well mixed.

Overall, I have really enjoyed this semester and feel I have taken away a lot of valuable skills both in using generative AI and critical evaluation of AI.

 

Appendix: Manifesto Statements

AI generations are like raw footage – mix them, edit them, cut them up!

When in a creative rut, asking AI for advice can show you what you don’t want, fueling inspiration

Using generative AI is a skill – practice, practice, PRACTICE!

Retrotech presents: Retrotech FM!

Retrotech FM is the station for the grooviest hits, hottest products and freshest interviews. Strap yourselves in folks for the listening experience of your life!

 

Created by Kallista, Sophie & Verona

 

Reference List:

ElevenLabs. (2024). Text to Speech & AI Voice Generator. ElevenLabs. [online] Available at: elevenlabs.io/?utm=check&pscd=tr…m9iZXJ0aG9vazU3MDc [Accessed 14 Oct. 2024].

OpenAI (2022). ChatGPT. [Online Large Language Model]. Available at: chatgpt.com/.

Pixabay (2024). Pixabay. [Online] Available at: pixabay.com/sound-effects/search/analog%20window/.

Runwayml.com. (2024). Runway. [online] Available at: app.runwayml.com/video-tools/team…tivity/dashboard [Accessed 14 Oct. 2024].

Apple (2024). Siri. [Online Assistant].

Suno (n.a). Suno. [Online Music Generation]. Available at: suno.com/

Given we submitted our major assignment this week, I thought it would be fitting to do a reflection on the production process, especially since assignment 5 is no longer a long form reflection for media studios.

Working on our audio piece has been one of the best experiences I have had working on a major project in a media studio. Firstly, it was fun to have a group that felt just as passionate about the project as I did. Collaboration felt genuine throughout the process, and we took time to bounce ideas off each other to improve our concept. Usually when creative work is delegated, it can feel disjointed, however this project didn’t feel that way. Even when we worked independently, it all felt under the same vision.

I really enjoyed the production process of creating an audio piece. The bulkiest part of the production process was post-production, compared to video which is usually production. There was a lot more freedom in post-production than with video, as we weren’t just limited to what we could see in a shot or just beyond the frame. After working on this project, I feel like I have a pretty good handle on reverb, EQ, pitch and working with adobe audition in general.

During assignment 2, out of frustration of generative AI not producing what I wanted it to, I often felt the need to pick up a camera to create footage. By comparison, I hardly had this issue when generating audio for this assignment. There was only one instance I felt the need to record my own audio of ambient outdoor noise. Something I did find limiting in the post production phase, however, is if we put an AI generated soundscape into audition, it would have different left and right surround channels, which made it harder to mix. I ended up mono channelling all our AI generated sound effects so they could be panned and mixed more effectively.

An opportunity for the future would be more intention in the order I build a soundscape. When editing, I went in chronological order when building the piece. This made it harder to mix levels later in making dialogue and ambient sounds balance across the entire piece. In future, I will get the levels right of the focus audio for the whole piece, and then build from there, one layer of audio at a time.

Over the weekend I saw a friend who is also studying media. He specialises in editing, and after mentioning I am taking an AI and media subject, he mentioned AI is used in editing software. Over this semester we have spoken about and put into practice AI for creative media. I thought it would be interesting to investigate how AI assists the more practical side of production, particularly editing, as this is the phase of assignment 4 my team is up to. Although, we are creating an audio project and therefore are not using video editing software.

For video editing, I primarily use Final Cut Pro X for editing my media projects. My friend explained FCPX uses machine learning for frame interpolation to get footage to run at a higher frame rate. After some research, I found FCPX also uses frame interpolation for footage that has been slowed down (Seymour 2024). Whilst I don’t adjust my frame rate very often at all in post-production, I do use slow-mo rather frequently, so I decided to test the capabilities of machine learning slow-mo in FCPX. I slowed some footage down to 25%. Whilst it didn’t look as smooth as footage shot at a higher frame rate and then slowed down, it certainly looks better than footage slowed down without frame interpolation. As someone who is often shooting on iPhone, and therefore has limited frame rate options, I will definitely use this feature in the future.

Another feature I found is AI, and have coincidentally used before, in FCPX is Enhance Light and Colour. This is a colour grading feature which analyses and enhances footage when applied (Seymour 2024). I tried to use this for a previous assignment in another class. For my other class we shot a promotional video. I forgot to do a white balance correction before shooting a particular shoot resulting in my footage looking like figure 1:

 

 

Figure 2 seems to be too pink in tint and the exposure is too low. Whilst the one I did myself, figure 3, is not perfect, I do think I did a better job. However, I can appreciate when this technology hopefully advances enough to correct footage to a satisfactory level, it will be a very beneficial. Using AI in the more practical side of production marries to the concept of AI being a tool rather than something taking over work, which has been discussed all semester. Despite its downfalls, having AI in user friendly editors like FCPX makes achieving higher level outputs possible whilst shooting on iPhone.

 

References

Seymour M (2024) AI: the battleground for Final Cut Pro vs. Premiere Pro vs. Resolve, fx guide website, accessed 13 October 2024. https://www.fxguide.com/quicktakes/ai-the-battleground-for-fcp-vs-premiere-pro-vs-resolve/

Our original concept for assignment 4 was to produce a retro, 60s, space age inspired AI products infomercial. Unfortunately, the TV studios were booked too far into the future for us to complete the assignment with a realistic timeline. As a result, we had to pivot to either adapt the original or create a new concept. As a team we reflected on our project and developed a new concept for an audio piece, as producing it in the time frame would be more logistically viable. We also found ourselves bouncing ideas around rapidly for the new idea.

After this, I started thinking about the concept of pivoting in relation to AI. As I have gone on throughout the semester my ‘aha’ moment has been that generative AI is best used for very specific purposes, such as sound effects. I am very much a planner and don’t often pivot creatively whilst in the production phase of a project, unless there is a better idea or logistical issue. This explains why when using a tool as unpredictable as generative AI, I prefer using it for very specific purposes where I can exercise more control to fit the pre-planned vision. However, maybe generative AI is just as useful for when you have very broad ideas, especially without a pre-conceived conclusion. Perhaps generative AI could serve as a jumping off point, with each new image or video inspiring the next, all contributing to a developing narrative. Being more open to pivoting, potentially redeveloping a whole concept, when working with AI could lead to unexpected and fresh outcomes.

Throughout this semester I have been rather rigid in my approach to generative AI. As some of my classmates said in their presentations in week 9, we need to embrace AI’s weirdness. I wish I had done this earlier in the semester, especially for assignment 2.

For assignment 4, our use of generative AI will mostly be for sound effects and dialogue, which compared to visual mediums, is more difficult to get creative with. With that said, as I enter the generative AI process for assignment 4, I’m going to be more open to the weirdness of generative AI. Rather than re-prompting for certain outcomes, I will consider the possibilities of what has been produced.

In class this week we did our assignment 3 presentations. It was fascinating to see many of my peers share similar perspectives and learnings from the course so far. Many people realised AI is best used as a tool rather than a crutch to do everything. There was also a common thread amongst many of the thesis concepts, focusing on the relationship between AI and creativity.

My favourite presentation, unsurprisingly, was Leah’s. Her work throughout this semester continues to be insightful, creative, and spark deeper thought. In her presentation she highlighted her four key learnings from the semester: awareness, patience, exploration and flexibility. The idea of having patience with AI was brought up by several presentations and is an insight I will take with me into assignment 4. Going into this course I assumed AI would be able to do everything I wanted it to do straight away. It took until seeing the assignment three presentations I realised patience is a key factor in using AI, as it requires trial and error to get a desired result.

In Leah’s presentation she also described how she achieved consistency in her characters for assignment 2. This was something I was really impressed by and confused how she managed to achieve. She created an AI image of the characters in Leonardo.AI and then put those images into Runway. This technique is something I will keep in mind, especially for assignment 4.

In our second session for the week, we concluded class by discussing what advice or resources we may be useful for future students of our subject. I believe a prompting guide would be helpful, as different AI softwares require different prompting structures. For example, runway prefers much more succinct straightforward prompting compared to Leonardo.AI which likes long descriptive sentences. I would advise future students to lower their expectations of current AI technologies. They are not as perfect as the media makes them out to be. They are incredible don’t get me wrong, however there is a lot of trial and error involved with generative AI. Additionally, at the start of the semester, I felt a lot of shame in using AI to help with my assignments, which is why I never used it. It felt like cheating. When AI tools are treated as a tool, their power gets transferred from the machine to the user, making it feel like a helping hand rather than an answer sheet to a test.

In class this week we watched everyone’s assignment 2 projects. It was exciting to see how everyone took on the prompt and used AI tools. A common thread amongst the pieces was the incorporation of mythic or supernatural creatures, possibly as AI can generate images usually requiring an VFX artist. Another common format was more experimental or poetry based films.

My favourite film was Tu Phi’s as it had a clear and well executed plot, and the AI generated footage perfectly conveyed the story. I was particularly impressed by the shots of the Axolotl. I have no idea how Tu Phi was able to create such ambitious, consistent and detailed shots of the creature with few hallucinations.

I was also impressed by Leah’s piece. She was able to create two consistent looking characters for the duration of the piece, which is difficult to do with an AI tool. She also managed to create shots with complex camera work, such as a shot that emulates the effect of a camera zooming in and tracking backwards simultaneously.

Seyyid’s piece stood out for its engaging plot which he somehow crammed into two minutes. It was also the only project that used the anime pre-set style, which gave the whole piece a consistent aesthetic. The aesthetic also felt less uncanny-valley-polished-AI- emulating-life-like-figures due to it being animated, which was refreshing.

I think the most successful pieces were able to balance an engaging narrative with consistent and detailed visuals. One of the hardest parts of creating assignment 2 was achieving continuity between shots. I find this to be a fascinating concept, as in film there are continuity personnel on set to ensure continuity. In the case of AI, continuity is no longer about actors’ hair being a consistent length, or props being in the same place, but is about a character having the same appearance characteristics between shots. This seems to make aspects, like the consistent placement of a prop, to be irrelevant in the creation of AI footage, as it is hard enough to get the basics consistent. In the aspect of continuity, AI generated footage seems to be harder to work with and yield poorer results than traditional filming methods.

In class this week we attend the This Hideous Replica exhibit. Some pieces really stood out to me. My favourite piece was a series of models around a table:

Whilst all four model’s meaning was rather on the nose, I enjoyed some other the more nuanced aspects of the pieces. The piece in the orange jumpsuit didn’t have a physical body inside it like the other three pieces, emphasising how sucked into the VR goggles the figure is. The orange jumpsuit also reminded me of a prisoner uniform, highlighting the figure’s entrapment. The figure on the far right also seemed relatively smaller, making it seem like it is maybe a child or a teenage stuck in the laptop. The four figures sitting in this arrangement appear like a family with all of them stuck in technology and faceless, unable to talk to each other.

Another piece I enjoyed was Shadow Tresses (no author). On first viewing the piece was very eerie, with the dark lighting of the room, unsettling audio track and creepy appearance of the puppet against the screen. After viewing the puppet from behind, it is quite simply put together, with a few mechanisms spinning and a light rotating to create shadows. This piece reminded me of our week 1 class discussions on coding. Coding can create the illusion of complex and sometimes scary concepts, such as AI. However, if you peel back the curtain and things simpler than they seem, making the beast less intimidating.

The exhibit was lit by a series of lights like in figure 3 below:

These fixtures were made of various materials and existing prints and designs, such as money notes. In the darker parts of the sculpture, there were small authentication stickers that read “original”. These pieces remind me of discussions I have had in previous weeks about AI and art. The fixtures are made up of pre-existing works and materials which are remixed to create the piece, like how AI is trained on existing works, and then produces something ‘new’ based on prompts. Both are old works reconfigured to create something new and unique.

References:

(n.d) [sculpture], RMIT Gallery, Melbourne.

Luscombe L (2024) Shadow Tresses [puppet], RMIT Gallery, Melbourne.

Debris Facility PTY LTD (2024) PlastiCorpUS [light fixture], RMIT Gallery, Melbourne.

In class this week we covered a reading by Andreas Ervik (2023). In the reading, Ervik raises a theory by William J.T. Mitchell (1986, cited in Ervik 2023:44); there is a “human capacity to recognize an image as an image”. For example, a photo of a person is photo which captures a person in a moment in time. It is not a person. It is a photo. Ervik, of course, relates this to AI and the ability to identify a generated image. Taking the same example, an AI image of person, is not a person, but an image of a person who does not exist.

Something I am realising is how so many AI media theories are linked to much older media theories. When discussing this theory in class my mind immediately went to René Magritte’s 1929 The Treachery of Images, colloquially known as Ceci n’est pas une pipe.

Despite being nearly 100 years old, the image is regularly used to discuss media theory, as it is not a pipe, but merely an image of one. In high school we discussed this in relation to representations, with every piece of media painting different people and groups in specific and different ways.

Similarly to Ervik’s discussion, this theory can be applied to AI generated images. The subject of AI generated content is an image of what that AI software thinks that subject is.

Similarly to how Magritte’s, Mitchell’s and Ervik’s theories all seem to build on each other, even if they are not directly connected, Ervik also points out “that image generators turn other image-making technologies into their content” (2023:44). If AI gets to a point where the content it learns from is also generated by AI, would this remove all human aspects of AI content? At least at the moment human generated content is being used to train AI. If AI is trained by AI content could that lead to a weird version of telephone where prompts no longer give what we ask for? For example, could the AI depiction of a loaf of bread gradually change over time after being influenced by more and more AI generated images, possibly with hallucinations, causing it to no longer resemble what humans would see as a loaf of bread?

 

References

Ervik A (2023) ‘Generative AI and the Collective Imaginary: The Technology-Guided Social Imagination in AI-Imagenesis’, IMAGE, volume (37):42–57.

Magritte R (1929) The Treachery of Images [painting], Wikipedia website, accessed 1 September 2024. https://en.wikipedia.org/wiki/The_Treachery_of_Images

 

In class this week we watched everyone’s mini documentaries from assignment 1. I was quite impressed by a lot of the pieces and took away ideas for future assignments. For starters, some people had quite creative ideas for topics, such as google translate and Apple carplay. There was a lot of effective uses of AI generated content. I struggled with this aspect when creating my assignment, so it was comforting to see it is very possible to seamlessly incorporate AI content and it is a skill I can learn.

I spoke to one of my classmates, Verona, about how she was able to generate content for her piece. She took pre-existing images and gave AI generators commands to add camera work or expand on the images turning them into a video. When I created my piece, I asked the AI video generators cold to create content, explaining why mine may not have turned out as good as my classmates. Verona’s technique is something I will try in future.

The standout pieces in the class to me included Bonnie’s as she included an interview, really making her video feel like a documentary. No-one else did this making her documentary standout even more. I thought Ethan had really well shot b-roll in his piece. Every shot was focused, and the framing was well balanced. However, my favourite piece was Leah’s. It truly was not only an original take on the prompt, but a unique take on a documentary, as she seemed to draw from art and experimental films. She created a piece of art and made AI generated content seamlessly blend into the project. I was in awe of her work. My piece did not incorporate AI generated content very seamlessly so I will definitely take inspiration for future projects.

Across the board there were two main aspects I thought the class could generally improve on. Firstly, sound mixing and sound layering. Quite a few pieces weren’t mixed properly between voiceovers and soundtrack. Some pieces also would have benefited from putting music in the background to cushion any background noise in their voiceover. Something else I think could have been improved on was the framing of shots. Many piece to camera shots were off centre, which whilst I know it’s hard when you’re filming yourself, this can be fixed in post. Whilst my sound mixing was not perfect, I do think relatively, my framing and sound mixing was quite good.