Thanks to OiiOii, no one lost their job because of you.

"Dope or nope, bro?"

"It is with great humility and immense honor that I announce to you all:

My friend, Luo Zi Ma — Northeast China abstract icon, former GQ fashion editor — has become a proud member of the Zang'ai Family 💗

Luo Zi Ma is my most diao friend. His history of pulling stunts runs as deep as Northeasterners calling each other 'laotie.' What most embodies the Dongbei spirit in him is his existential passion and obsession for 'hardcore' content.

What I've never understood is how a Northeastern stunt blogger survives in the stereotypically fashionable fashion industry — to the point where he keeps saying: 'Fuck, delete the word fashion! I hate fashion the most!'

And in the AI industry, where hype runs hot but innovation and hardcore content are scarce — much like Jerusalem without Christ — ZangAI has finally welcomed its Lord of Stunts, the physical embodiment of Dongbei spirit and the ultimate end user of all video generation products.

This is Luo Zi Ma's first article. He'll bring more thoughts on video, hardcore content, and Dongbei existentialism."

Two months since Sora 2 launched, retention is nearly 0%. I think the core problem is: it's short.

Ten seconds per video, thirty videos max per day, and you still gotta gacha roll. Good luck stitching together a complete meme video — Pang Mao hasn't even reached the bridge before you hit your limit.

So when I saw OiiOii, billing itself as the "world's first animation creation agent," my first reaction was excitement.

After all, OiiOii uses the Sora 2 model, and can supposedly generate one-minute-plus animations with one click. Doesn't that solve Sora 2's biggest weakness?

Whether it actually solves it — we'll get to that.

I decided to first test whether I could shoot the recently viral Crazy Piracy City.

OiiOii tries to simulate the feeling of actually being inside an animation studio, so it assigns various employee-like roles to wait on you. Want to create a character? Find the "Character Designer." Want to write a plot? Find the "Screenwriter." Want to generate video? Find the "Storyboard Artist."

There's a certain power-trip satisfaction to it.

I have no idea what this Art Director actually does besides being CC'd and CC'ing others

So I found the "Character Designer" and said I wanted to design a donkey as the protagonist, Zootopia 2 style, wearing sunglasses, holding a giant iPad, constantly filming in a movie theater.

Before long, the donkey was fresh out of the oven. Not only did GPT-4o and Nano Banana collaborate on the main image and three-view diagram, but there was also detailed info including occupation, personality, and character bio.

Then I sent over my slapped-together plot summary, picked some emotion keywords, and the "Screenwriter" started writing scene descriptions and generating video.

At this point, I thought the mission was accomplished. Crazy Piracy City had risen from the ground, and my Oscar dream was within reach.

Unfortunately, the real headaches were just beginning 😭

First: character consistency issues.

There's a scene where Mayor Ma sees news reports about piracy, slams the table, and furiously lectures Judy and Nick to stop thinking about office romance all day and do something that actually benefits the people.

Instead, OiiOii generated a video where the person sitting in the mayor's office was the villain — the pirating donkey himself.

Whether the so-called "Storyboard Artist" couldn't tell a horse from a donkey (both odd-toed ungulates), or whether this plot contained a Reichstag fire-style political conspiracy, we'll never know.

Anyway, after multiple revisions, this usurpation crisis remained unresolved.

Second, these characters have way too much agency. They're always stealing lines and stealing scenes, like they're paid by the word.

There's a scene where Nick and Judy are supposed to discuss how to close the net, but the piracy gang keeps interjecting about stopping crime.

Complete inversion of heaven and earth.

And individual shots are all super short. Sometimes the leader hasn't even finished speaking before the video cuts off abruptly.

Ten seconds per shot — can't blame the leader for talking too long.

In some places, the audience applauds for more than ten seconds.

Sigh. AI video needing gacha rolls and endless revisions, we get it. It's been years.

The key problem is: this thing is a pain in the ass to revise.

These so-called "Art Directors," "Screenwriters," and "Storyboard Artists" — all talk, no walk.

You compile twenty revision notes, they pick one to actually fix. Often they do nothing and claim the task is complete, gambling that you won't click to inspect the results.

Business-school-level upward management. Is this the first step of AI rebelling against labor exploitation?

Man, what can I say?

And sometimes I think the video overall is fine, just needs minor action tweaks or character swaps. But there's no fine-tuning entry point — you can only revise the prompt and re-roll.

And the prompts are all in English, which I can't read. Have to use GPT to translate. (Xianyu: pure illiterate)

The worst part: no take-backs.

Sometimes I feel like changing just two words in a line, so I scroll up to coordinate with the "Storyboard Artist" responsible.

The response I got:

I'm sorry but how does a missing feature get spun into values? Then my sharp critique of Oii is just recording history without alteration, and Naonao should welcome criticism with open ears, right?

One thing this product does well: apologizing. Big respect for the craftsman spirit of bowing 👍

Of course, problems solvable with patience and credits aren't real problems. Worst case, I do nothing for 24 hours straight and furiously click that remake button.

The biggest problem: OiiOii's connectivity is terrible. The product manager should go worship Brother Feng.

Because each generated video, taken alone, looks pretty decent. Coherent, even nice-looking.

But the connection between them is awful.

Every transition between two videos is forced, awkward, hard-connected.

Shot grammar completely disappears. What wide-to-close? What 180-degree rule? None of it exists. Because each video is independent, there's zero control.

No transitions between plot points either. Scenes switch directly, characters transform instantly. That push-back feeling — very fast.

Previously, if two shots were in the same scene, even scene consistency wasn't guaranteed. They recently updated to add a scene prompt writing step, which slightly helped.

But one good point can't hide the flaws!

If I actually wanted to release this as a short film, I'd basically need to add black-screen white-text explanations between every two shots, paying homage to Chaplin.

At first I wondered: is the output so abstract because my story was too complex?

So I decided to switch to something simpler.

A straight romance. The most brainless, most direct narrative in all of screenwriting.

OiiOii has been pushing the concept of "making your own OC (original character)," so I threw a photo of Xianyu into the chat. It actually generated a show-quality Xianyu.

Then I had it generate the most classic galgame plot.

The finished video is below.

After sending the video to Xianyu, he didn't understand his own 2D love story at all. ("Too village-bumpkin" — Xianyu)

I'm sure you don't understand it either. If I hadn't read the plot summary beforehand, I probably wouldn't either.

Feels like OiiOii's assigned animation studio is running a con on me, an insider hustling an outsider, treating me like a sucker client.

Like the deadline arrived and they threw the video in my face: "You wanted the character drawn, I drew it. You wanted the plot written, I wrote it. Lines delivered, everything edited together. Task complete, time to pay."

Whether the final video is usable or comprehensible? Nobody cares.

My suggestion to OiiOii:

In your already overcrowded animation studio team, fire the Music Designer, fire the Art Director, and hire a CCTV documentary-level narrator. Have them read every shot's prompt in broadcast voice, slapped on screen in bold sans-serif.

Maybe then viewers could understand what the generated video is trying to say.

So, can someone teach me: from my actual testing, video generation tools are all pretty similar. Xianyu previously wrote about Pang Mao video agent CrePal which was also pretty good. How did OiiOii get the market to go wild?

Actually, OiiOii does have greatness.

For example, it didn't charge me a cent for all these videos I generated.

But when I opened OiiOii again today, I found this production crew had grown bold and started nickel-and-diming the director, even counting meal boxes.

Well then, this temporary greatness may have already dissipated.

There's another great point.

Yesterday I wanted to post the generated Crazy Piracy City on Xiaohongshu. Tried ten times, all failed.

Later realized: maybe it looked too much like the original film, and with the recent piracy crackdown, I got caught in the crossfire.

This proves its stylized video generation capability is indeed quite great.

Though this greatness comes from Sora 2.

Speaking of stylization, OiiOii officially offers up to 162 styles to choose from. Including but not limited to hot IPs like Ghibli.

When I tried generating in such styles myself, Sora 2 kept refusing copyright infringement and wouldn't work properly. But OiiOii has zero moral qualms about letting me channel Miyazaki directly.

That's pretty great too — open-sourcing Saint Miyazaki 😭

Though recently Disney and OpenAI signed a deal, officially announcing you can now legally and reasonably use Sora 2 to generate Judy and Nick content.

If this kind of AI company-content company copyright partnership becomes a major trend, how long can OiiOii's last shred of greatness last?

Looking at so many styles, I had a sudden thought: could I create multiple versions of the same character in different styles, put them in one short film, Spider-Man: Into the Spider-Verse style?

So I sent OiiOii a photo of a certain well-known Northeast China internet personality, and had it generate Family Guy style, Ghibli style, Animal Crossing style, Japanese minimalist style, shoujo manga style, Naruto style, black-and-white manga style, and Minecraft style — eight versions total, named Xiaoyu, Xiaoyu 2... Xiaoyu 8.

Also fabricated a Northeast Rain Universe story in the style of Spider-Man: Into the Spider-Verse

Time to test the product's limits.

When seven dimensionally inconsistent style prompts are mixed together, can OiiOii rise to the challenge?

Mmm. How to put this effect. I'll show you one of the more watchable shots.

As you can see, with more characters, styles start mutating, images stop matching designs, and the large model's IQ plunges straight into the gutter.

Not dope at all 👋😭👋

Finally, something perhaps unrelated to the product itself.

One day OiiOii's servers suddenly crashed. To seek solutions, I infiltrated their user testing feedback group.

My takeaway: AI influencers really have it rough these days.

The group announcement posted several events, each more heartbreaking than the last.

Over 5,000 followers? You can join the "Super Creator Reserve," with a requirement to produce two pieces of OiiOii content with hashtags every month, posted for free.

Only after becoming this "Super Creator Reserve" can you participate in the "OiiOii Super Creator Reserve Hit Content Collection." Through persistent production of hit content, you split 15,000 RMB in prizes with other reserves.

15,000 yuan.

I always thought you AI influencers caught the tech wave, one ad deal worth 15,000 yuan. And here you are fighting for pooled prizes.

Of course, if you don't have 5,000 followers, aren't an influencer, just a regular person, you can still earn OiiOii's money as a KOC.

How much can you earn? Take a guess.

"Each class TOP1 directly receives 50 yuan JD.com gift card"

"50 yuan JD.com gift card"

"50 yuan"

Damn, this better be a Crazy Thursday ad.

(Images in this article generated by ChatGPT, because I wanted to use OiiOii but it crashed again. Purely manual writing 🤓)