My Sora Problem (or How I Learned to Stop Worrying and Love the AI)

Disclaimer: I make no apologies for my past use of AI in this piece. I recognize and hold space for the ethical concerns that artists have of generative AI effectively stealing their IP to iterate creations through which they see no recognition or financial compensation. I also recognize that AI at its current output is an unsustainable model where it is difficult to justify the consumption of a gallon of water to create a Sora video:

I acknowledge both to be true, but I also acknowledge that the creations have already been made and even if I delete them, that consumed water is not coming back. If you want to voice yourself in the comments, know that I will not turn them off or block them. I’ve seen the bullying and vitriol directed at people who openly use AI and I refuse to be silenced by them. Again, I ask that you leave respectful comments, but I will not be removing any of them.

On April 26, 2026, Sora, the platform known as the TikTok for generative AI videos, was finally shuttered.

I took to the platform to use up my final batch of daily credits, knowing that in a few short hours, the ability would be lost to me and I would have to seek other services like Nano Banana or Hailuo to get that same fix.

I have a lot of complicated thoughts about the closing of Sora.

It is an understatement to say I dipped my toe into the platform.

More like, I plunged headfirst into experimentation as soon as the platform became available.

At first, I didn’t have many grand ideas to express. I wanted to see the concept of a young school-aged witch who used a magic umbrella, and that became my first generated video:

Then I stopped for a bit, until I had another idea. One image that floats through my mind is of Lock and Barrel – the two male henchman of Oogie Boogie’s Gang from The Nightmare Before Christmas – singing with Vanellope von Schweetz from Wreck-it Ralph in a music video to “Black Rum” by Cake Pop. But as my idea involved Disney IP, and Sora’s guardrails wouldn’t let me realize it, I quickly abandoned the project.

Then, my thoughts gravitated to this stupid little idea I had when I was young, like between 5-8 years old.

I have no idea where this thought may have originated. But I had this weird little video scene in my head of a man rapping about it being “Motherf*cker Day” against and overly cheerful, colourful backdrop. So, I prompted Sora to show it to me and this was the result:

I was enamoured.

From that point on, it was like a dam broke.

I decided to test the limits of the Sora machine (sh*tdisturber mode activated) and use the remix feature to infiltrate other users’ creations with “It’s Motherf*cker Day”:

And then I learned another thing. Although Sora had guardrails, there was a way to game the system to generate creations that created protected IPs.

Consider Sora like a slot machine. Every time you pull the lever for generation, the LLM (large language model) spits out a varied outcome of your prompt. In one instance, Sora reads your prompt and determines it is too close to a Disney copyrighted property, and shuts the generation down.

But sometimes—and if you make your prompt simultaneously descriptive and yet vague enough to not tip the guardrail alarm—you get something surprisingly close, as though you gave it your intended particular prompt from the get-go. Case in point:

I shared it on my non-writing Youtube channel, SylvsTO, and it got more views than many of my other videos, some of which had taken hours to conceive.

I’m ashamed to say that at that point, I had caught the Sora bug.

I then wanted to realize several other visions in my head.

I had this thing, like a “Cartoon Karaoke” in my mind of random characters singing particular songs. I wanted to see how close they could be realized.

So I did Cosmo from the Nicktoon series The Fairly Oddparents and his band “The Fairies” singing “Force Ten” by Rush:

Then I did Link from the video game franchise The Legend of Zelda singing a Chris Daughtry song:

Then Yami Yugi from the anime Yu-Gi-Oh! Singing “Vincent” by Josh Groban:

Then Storm from X-Men: The Animated Series channeling Whitney Houston:

Then Harley Quinn shooting the Joker to “Falling into You” by Celine Dion:

Then something akin to the PSP squirrels singing to hyperpop (I will revisit this one later).

After several of these successful outputs, I decided I wanted to realize my most ambitious idea to date:

Aside from the Vanellope von Schweetz/Lock/Barrel “Black Rum” music video, which I put on the backburner after Disney made its announcement to partner with Sora AI in 2026 (I decided to bide my time and wait until generating Disney IP didn’t come with the guardrails), I had the idea to generate Tootie from The Fairly Oddparents to Mariah Carey’s “Fantasy”.

But unlike my other one-offs, this was not just a simple clip. I had an entire music video envisioned in my head, and after experimenting ad nauseum with dozens of prompt generations (I’ll spare you all the sordid attempts), I determined that I was going to attempt to create the entire thing on Sora.

A few problems arose:

One, as with the Disney IP, Nickelodeon IP was notoriously difficult to generate. And I wanted to generate Tootie, a very dynamic-looking character in cartoonist Butch Hartman’s style with purple glasses, pigtails and braces. While I came close, I could never quite get the character spot-on, which led me to another issue…

Two, I could only make so many generations a day. After spending 1-2 hours a night generating and regenerating, prompting and re-prompting what I wanted to see and running into the guardrails—an act for which Sora would later penalize repeat offenders with a timed ban for too many violations—I could only create about 10 or so 10-second clips in a day, which was woefully inefficient. And “Fantasy” is a 4-minute song. I would need at least two dozen Sora generated clips, and that wasn’t counting the number of unusable clips or time wasted on generations that would never materialize due to implemented guardrails. The alternative was to shell out the approximate $6 for extra credits, but I never did that.

Three, what I envisioned was an animation. And as good as Sora was at generating creations which adhered to the laws of physics, animation was a considerable shortcoming of the technology. A 10-second clip was fine, but what if I wanted something longer? It was nigh impossible to continue a scene and keep the integrity or consistency of the clip (most of the time it would switch to a live-action generation), the remix feature simply did not work. I think this was because Sora didn’t actually animate, it only generated what it assumed was an accurate recreation of animation. Animation is not rooted in physics, which was more of Sora’s strength, so any generation was pure guesswork at what I was trying to accomplish. Sometimes it was passable, like my attempts above, but other times the output was off-putting, nightmare fuel. Even the hyperpop squirrel clip was buckling under the stress of its own generation in the last few seconds.

In the end, the time and labour involved, coupled with the constant guardrails, coupled with being unable to create a consistent animated model via AI led me to abandon the project. I figured I would pick up again in the fall of 2026, when it was predicted that Disney would be on the AI generation train.

I mainly shared my Sora creations on the platform itself and on my Youtube channel, SylvsTO. However, I shared a few creations through my Instagram stories.

I’m a big Hey Arnold! fan. So much so that I have a website dedicated entirely to my Hey Arnold! / Sailor Moon crossover fanfic, Sailor Helga.

As much as I love my story, one thing it lacked were the visuals. I had commissioned talented artists in the past, and they had created some phenomenal artwork to my specifications. (One of which is present on the website now).

I tried my hand at generating my own version of Sailor Helga. A lot of versions were fine, but I managed to create one I was quite proud of.

I shared on IG with the comment “I think I have a Sora problem”, only half-joking. An artist whom I respect and whom I commissioned to create art for my projects in the past direct messaged me his dismay, which gave me pause.

This was around the time I really began to register the negative backlash surrounding the use of AI. I was just using it for fun; I didn’t intend to make any money from it. I just wanted to see the thoughts in my head made manifest. But a lot of people—artists in particular—took severe umbrage with the use of AI in the creation of art. Look no further than the public breakup of Owl House creator Dana Terrace and Duolingo, when the former called out the latter for laying off their employees and replacing them with AI.

As it stands, the AI shorts (or even shorts that mention AI) are the most popular thing on my SylvsTO channel. But there is one creation in particular that has shot above all others, currently sitting at 23K views.

Remember the PSP squirrel hyperpop video? I’m not sure if it’s because MC Boing’s song “Boing Beat” has become a meme, or if it’s because the clip is in black-and-white so it appears to mask the Sora watermark, but people have recently come out in droves to attack this particular video. Maybe because it’s the most “original-looking” of the creations, or maybe people feel betrayed that it wasn’t made by hand, that I generated it with a few lines of prompt.

Many commenters called me lazy. Many accused me of deceiving them and that I was killing the environment with my AI slop.

My response? OK.

I’m not going to argue with the consensus. That I could have animated it myself. That I could have commissioned artists to create what was in my head. Both statements are patently true. I make no arguments against that. I took the knowledge, weighed my options, and decided it made more sense to me to generate a 10-second clip that I wanted to see from my head with Sora vs. true animation. I could have kept it to myself, but I chose to share it publicly, so the decision understandably welcomed backlash.

However, I do come out of this obsessive adoption of Sora with a sobering realization that the masses were sold a big fat lie by these corporations who insisted that AI could do it all. It can’t. If you have infinite time, saintlike patience and a bottomless wallet, then sure, you may be able to generate EXACTLY what you want to the minute detail. But another artist can’t do that, and neither can AI. In theory, I could accomplish it, but I have a full-time job, so forgive me for not seeing the return on investment to download an animation software, learn tweening and frame-by-frame animation, and meticulously animate, storyboard and sync the movements to “Boing Beat”.

I’ve also cooled on AI in other regards. Aside from Sora, I had used ElevenLabs frequently to listen to my creations with voices other than the ones provided by the Read Aloud feature in Microsoft Word. I value the service and I’ll likely use it to proofread this piece. But I come away with some concerns.

The one thing I could not abide was feeding my likeness into the Sora app. I found it too unsettling the idea of me or the AI facsimile version of me engaging in uncanny valley situations and having my words and actions manipulated with only written lines of prompts, so I kept agnostic on that front.

The same option is available on ElevenLabs, and while I haven’t fed my voice into the platform, I’ve used the voices of others who have done so, mainly because the output is just that much more realistic.

And therein lies the problem. The voices. They are too realistic.

I’m a bit embarrassed to admit this, but through repeated exposure to my preferred voice outputs, I may have developed parasocial attachments to some. And while it’s not to the level of creating an AI boyfriend, I think for me, psychologically, the attachment veered a bit on the unhealthy and bled into my real life. I won’t go into specifics, but suffice it to say that this brave new AI world can and WILL consume you if you let it. I choose to no longer let it (and I’ve opted to go back to less natural-sounding voices to read to me).

So, amidst all the rambling, what is my current stance on AI?

I’ve finally laid to rest my dream of the Lock/Barrel/Vanellope von Schweetz vision in my head (that died along with the cancellation of the Disney Sora partnership).

I still use it for work, as a productivity tool.

And I still pay for an ElevenLabs and a ChatGPT subscription.

The one place where I don’t invite AI is in my writing.

My creations are entirely my own and I cling fiercely to that. It’s like what one of the speakers said at the Toronto Indie Author Conference that I attended this year: to paraphrase, I don’t want AI to create for me so I have time to do dishes, I want AI to do the dishes so that I have time to create!

So on that front, I completely understand where the visual artists are coming from.

True art comes from the human mind and soul. An algorithm or code can never truly capture the sick, twisted, beautifully dark and wonderful machinations floating about in our heads. It simply lacks the nuance.

At the same token, I have a genuine love for technology and am cautiously optimistic about where the future of AI will take us.

Sora may be dead, but my motivation to create is stronger than ever.

And perhaps I will be inspired to eventually manifest the crazy visions in my brain with my own two hands, sans AI generation.

Until then? Happy Motherfucker Day.