Video Agents Enter the Era of Mass Homage
The Crown God was way ahead of his time.
"Malvin was way ahead of his time"
LibTV dropped a new version yesterday, so here's a quick update.
The past six months of video agent evolution can be summed up in one sentence: "Eyes on me, everyone — look at me, look at me — I have an announcement: we've evolved into TV form."
This includes, but is not limited to: OiiOii upgrading to 2.0 with shot-for-shot replication, smart canvas, and Skill libraries — all converging toward TV. And of course, the comments are already spamming "Oii TV."
Meanwhile, in a corner nobody's watching, Flova quietly swapped its domain from Flova.ai to Flova.tv. I'm guessing the TV version is already in the works.
Also joining the fray, still pushing infinite canvas and agents: MiniMax Hub.
Turns out Malvin going all-in on TapNow was the right call. We were the shortsighted ones.
This startup thing isn't hard at all, apparently. Just full-send homage to the tool paradigm Malvin already validated. Should've launched our own Ai-Mian TV and started grifting — $100 million valuation, who's in?
Though I still don't get why everyone's so obsessed with infinite canvas. The barrier to entry is absurdly high, barely any novice users can actually use it. Probably just trying to bask in Malvin's reflected glory.
TapNow is a state of being; LibTV has become a movement ❤️
Of course, LibTV's new Agent+Skill drop is just old wine in new bottles. You literally can't find a video agent without similar features anymore.
This pile of Skills looks bigger than what local agents offer, and canvas + agent clearly still won't painlessly turn noobs into directors.
So my biggest takeaway: the entire video agent industry is basically manufacturing. Six months of product-form innovation can't match the user growth from a SOTA model drop.
Yep, Malvin saw through it all long ago. The core of this industry is plugging into SOTA models — preferably cheap SOTA models.
First-access rights to models matter more than any product innovation. Get them, and you get free money-can't-buy traffic. Flash back six months: when Seedance 2.0 dropped, every video agent raced to be first-to-launch, so fiercely that Volcano Engine had to step in and mediate.
Let's all look forward to the upcoming Seedance 2.5 launch-rush tournament.
Given Malvin's burn rate at Volcano Engine, ByteDance definitely knows who gets external first-access rights. Think about who your daddy is — oh wait, I mean your son?
I have a genuinely constructive suggestion here: if LibTV sometimes gets ByteDance models faster than Dreamina, and at prices way below Dreamina, maybe Seedance 2.5 should switch its launch partner?
As for who's really the Zhang Yiming -1 — that's still up for debate.
Side note: I saw AISphere pushing something called "AISphere Fan" yesterday — a crude knockoff of "ByteDance Fan." Malvin hasn't even rolled out "Lib Fan" yet.
Right now every video agent is like the convenience stores you pass on the street. At their core, they're doing wholesale-to-retail: buying Seedance 2.0 tokens from ByteDance at 10 million, trying to sell them at 11 million.
Why "trying"? Because some real pros are doing it at zero margin, forcing everyone else to slash prices too.
Using a video agent that isn't LibTV is like buying an iPhone without Pinduoduo's billion-subsidy discount.
This subsidy phase is absolutely about killing competitors. Though I also heard some peers are reverse-engineering Malvin — truly dishonorable, no martial ethics whatsoever.
Someone posed this theory: "What if ByteDance first makes some money off Malvin, then uses that money to subsidize?" I'm reserving judgment on that one.
Oh right — some video agents are like 7Fresh (Xiaoxiang Supermarket), with self-branded goods. MiniMax Hub sells Hailuo AI videos, for instance. Unfortunately, private-label stuff never sells that well.
Oh, and today is the last day for 35-50% off Seedance 2.0 series models on MiniMax Hub. Don't miss it, folks.
Once you understand video agents as supermarkets, a lot of things click into place.
For a supermarket, only three things matter: product selection, price, and location. In video agent terms: models, discounts, and content.
Models mainly come down to execution. Discounts come down to fundraising. Content is way more complex.
Simply put, content means grinding for traffic in the consumer space — like "No queues, full-power Seedance 2.0 now live on LibTV," or yesterday's Agent and Skill spam across the internet.
Of course, if you want real content, look no further than the one true supreme god: Higgsisfield. What's impressive about this company is how they constantly cook up new stunts for user acquisition.
Like generating AI videos inside Minecraft — still pretty janky, I tested it and it's basically useless. But it's a stunt. People love watching it, love trying it.
In terms of content value, way better than the recent Agent and Skill hype.
Of course, consumer-facing video agents are mostly just for show. Actually counting on C-side payments would be hilarious. Real monetization comes from ads, TVCs, and short dramas — that's who the canvas is actually for.
Until any of these video agents actually evolves the capability to train models, this industry is what it is. No major shifts coming.
Really no different from opening a store on the street corner.
But I believe Malvin will definitely evolve model-training capabilities and eventually become the Sam's Club of video agents!
Anyway, I meant to talk about LibTV's recent updates, so here's something actually about that:
I tried generating this script from the LibTV fan club:
"The Trinity — the Father Malvin, the Son LibTV, and the Holy Spirit Lib CLI — share one glory and one power, together steering the path of StableDiffusion image-generation. Malvin is the source of all methods; LibTV bears all phenomena of the world; Lib CLI dispatches all affairs. The Three in One, light and compute spread across the seas. All who wield the pen sing praise as one, forever unto Malvin. Most holy name."
It refused to generate, no idea what restriction hit.
So I switched to:
"It was a confrontation in China's AIGC track worthy of industry history books — an office reduced to scrolling code and the low hum of server fans. The startup team was resource-starved, compute budget tight, model ecosystem built from zero, essentially starting from scratch, facing off against established drawing platforms backed by major tech giants with deep resources. Liblib's Malvin chose not to avoid their edge or retreat to a niche segment, but plunged headfirst into the brutally high-barrier creator community ecosystem, cultivating original models, LoRA assets, and creator support systems.
On the other side stood an industry titan commanding massive compute, mature foundation models, then firmly atop the leaderboard — traffic, capital, technical reserves nearly insurmountable, the high wall no startup could scale. The giant rolled out features across the board, momentum overwhelming, as if market share was already locked tight in its grip. No one expected the opponent willing to clash head-on would be Malvin, who'd left ByteDance to build alone."
Behold:
Verdict: Malvin's radiance shines upon me.
Final summary. TapNow is Apollo the sun god, inventor of flame.
LibTV is the fire-thief — but Prometheus is great too, for spreading the sacred TV flame to humanity, so that all video agents knew to evolve into TV form.
The day Dreamina rebrands as Dreamina TV, Lark becomes Lark TV, and CapCut launches CapCut TV — that'll be divine prophecy fulfilled.
When that day comes, I'll write an article: "Dreamina TV: At Long Last."
(Cover image generated by ChatGPT; article 100% human-written)
One last plug for our first AI Horror Film Hackathon:
⬇️
Subscribe to our Substack: funeralai.substack.com