Labels

AI09 : Sora, So far

When OpenAI finally dropped Sora 2, I did what any self-respecting motion designer would do: immediately started forcing Sam Altman to do things and turning iconic Western moments into Bollywood sequences. Because if you're going to prove proficiency with a new tool, you might as well have fun with it.

The Exponential Speed Problem

AI models are growing at an exponential speed. By the time I finish writing about Runway experiments from September, Sora 2 has already dropped and made half those tools feel ancient. The credit crunch chronicles are still relevant, sure, but now there's a new sheriff in town with capabilities that make last month's "close enough" look quaint.

In the AI video space, you don't get to be precious about documenting your process chronologically—you jump to what's current or risk documenting obsolete workflows.

So yes, I'm skipping ahead. The other AI slop will get its documentation later. Right now, I need to show everyone my proficiency with Sora 2 before Sora 3 makes this entire post irrelevant.

Forcing Sam Altman to Do Things

First experiment with Sora 2: making Sam Altman do a thing or two. Call it revenge for all the times AI tools ignored my prompts. Call it testing the boundaries of what's possible. Call it whatever you want—I'm calling it a successful proof of concept that Sora 2's character consistency and motion quality are genuinely impressive.


The technical leap from Sora 1 to Sora 2 is substantial. Better temporal coherence, more believable physics, fewer of those telltale AI glitches that scream "this was generated." Though let's be honest—we're all still guilty of producing some grade-A AI slop when we push these tools too hard.

My Trick to Creativity: Make Everything Bollywood

Here's the thing I've discovered works consistently across every AI video model: make everything as if it was in Bollywood.


Another trick I use, is make the situation japanese, that is also funny in its own ways. pic.twitter.com/xjq5yIJdZY

Want to make something interesting? Add Indian aesthetics. Want to make something funny? Make the situation Japanese—that's also funny in its own ways—but making things Indian is my signature move.

It's not just about novelty. There's something specific about how Bollywood visual language translates through AI generation. The dramatic lighting, the saturated colors, the unabashed theatricality—these elements survive the AI translation process better than subtlety ever could.

Another example of making things Indian. #CharlieChaplin pic.twitter.com/qB831SvrDK

— Cai (@shashrvacai) December 1, 2025

Even Daft Punk Can Belong to India

This became my favorite experiment: taking iconic Western cultural moments and filtering them through Indian visual language. Daft Punk in traditional Indian attire and settings. Charlie Chaplin navigating Mumbai streets. The aesthetic colonization reversed through algorithmic generation.

Cultural remixing through AI isn't appropriation—it's demonstrating how visual languages can be fluid, playful, and generative in ways traditional media never allowed.

The results aren't always perfect. Sometimes you get beautiful cultural fusion. Sometimes you get confused AI trying to reconcile contradictory training data. Both outcomes are interesting.

Moodeng's Future (Like Any Other Famous Child)

Generated a sequence about Moodeng—the pygmy hippo who became an unlikely internet celebrity. The prompt: show the dark trajectory of child fame, but make it a hippo.


Sora 2 delivered something uncomfortably prescient. The way fame corrupts innocence, packaged in 10 seconds of AI-generated video featuring a hippo's descent from adorable to... well, something sadder.

AI video tools excel at capturing emotional arcs we understand intuitively but struggle to articulate—the tool fills in the gaps with training data from thousands of similar narratives.

Recently Found Footage of the Queen

The British Royal Family through an Indian lens. "Recently found footage" as a framing device lets you play with anachronism—make something feel historical that never existed.


Revenge against the Queen's land had to be taken, and only one knight is fitting to do so. The postcolonial subtext delivered via AI video generation, because apparently that's where we are now in 2025.

Revenge against the Queen's land had to be taken, and only one knight is fitting to do so. pic.twitter.com/PrsvABktR3

— Cai (@shashrvacai) December 1, 2025

The Halloween Party Invite That Never Was

This was the invite for my Halloween party this year. Except it wasn't. The party didn't happen. The invite existed only as a Sora 2 generation—a perfect example of how AI video lets you create documentation of events that never occurred.

We're entering an era where the line between "footage of things that happened" and "generated content that could have happened" becomes functionally meaningless for most contexts.

The implications are fascinating and terrifying in equal measure, but right now I'm focused on the creative possibilities rather than the existential dread.

The Jesus Phase

A bit of Jesus phase didn't hurt anyone? Taking religious iconography and running it through Indian visual filters creates something that's simultaneously reverent and irreverent—the kind of cultural mashup that traditional media would agonize over but AI generation does without hesitation.

The results walk a fine line. Sometimes it's genuinely beautiful syncretism. Sometimes it's bordering on offensive to everyone involved. Navigating this territory requires cultural awareness the AI doesn't have—that's where human curation matters.

Could I Make a Fake Influencer?

The question lingering throughout these experiments: I wonder if I could make a fake influencer?

Not in a scam way. In a "what if I generated an entire persona through consistent AI video" way. Sora 2's character consistency makes this genuinely feasible now. Create a character, develop their aesthetic, generate their content—all without a real person ever existing.

The technology for creating entirely synthetic influencers has arrived—the only question is whether the uncanny valley is shallow enough for audiences to accept it.

Honestly? We're almost there. Another generation or two of improvements and you won't be able to reliably tell. The ethics of this are complicated. The creative possibilities are undeniable.

What Proficiency Actually Means

Here's what I've learned racing to demonstrate proficiency with Sora 2 before the next model drops:

Proficiency isn't about mastering every feature. It's about discovering the weird edges where the tool does something unexpected and leaning into that. It's about finding your signature approach—for me, cultural remixing through Indian aesthetics—and using it consistently enough that it becomes recognizable as your work despite being AI-generated.

In the age of exponential AI growth, proficiency means adapting fast, documenting faster, and accepting that expertise expires monthly.

The tools will keep improving. The techniques that work today might be obsolete by next week. But the creative instinct—knowing what to make, how to frame it, which cultural juxtapositions land—that's still human territory.

For now.

The Practical Takeaway

If you're working with AI video tools in late 2025:

Jump to the newest model. Document your process, but don't be precious about chronology. Find your signature move (mine's Bollywood everything). Test boundaries. Make Sam Altman do things. Generate footage of events that never happened. Ask uncomfortable questions about fake influencers.

And most importantly: call your own work AI slop when it deserves it. Honesty about quality matters more than pretending everything you generate is revolutionary.






Email Post
Labels:

Comments

Post a Comment

Popular Posts