The subjects aren't real. The feelings are. Today's three outliers each pull a genuine reaction out of pure pixels — a beloved anime character mourned like a lost friend, a witch lifted from a 90s tape that never existed, a glowing world cupped in two human hands. You react first and remember none of it happened second.

Here's what each one makes you feel — and the move behind it:

@will_iovino takes a 2D anime character that fans watched die and renders him as a still, photoreal portrait — eyes closed, face calm, wind moving his hair against an open sky. The comments read like a funeral. It lands because realism hands fans something a cartoon never could: a real face to actually grieve.

📈 190K likes — 31.6× the account's average (@will_iovino)

Why It Works:

@jboogxcreative builds a hooded crone — bone necklaces, claw nails, cracked round spectacles — in the exact muted, grainy cel palette of a 1990s OVA. It works because it commits to the period completely: no neon glow, no modern sheen, just a frame that looks pulled from a VHS tape you half-remember watching.

📈 931K views — 3.2× the account's average (@jboogxcreative)

Why It Works:

@ai.with.glock rests a cracked, lightning-veined orb glowing in two cupped hands above a storm-lit mountain range, composited convincingly enough that you feel its weight. Then the caption — 'this got out of hand real fast' — turns the impossible scale into a joke. You stop for the spectacle and stay for the pun.

📈 72.4K likes — 83.7× the account's average (@ai.with.glock)

Why It Works:

None of these three needed a real subject to earn a real reaction — and that's the whole game now. The creators turning AI reels into income aren't the ones with the cleanest render; they're the ones who make you FEEL something before the thumb moves. If your posts look right but land flat, that's the gap to close.

👉 Join the VIP Community

xAI pushed Grok Imagine Video 1.5 Preview live in its API today, and it shot straight to #1 on the Image-to-Video Arena — a 52-point Elo leap past Seedance 2.0 and Google's Veo. Filmmakers testing it keep pointing to the micro-expression tracking and faster renders as the real jump. If image-to-video is your pipeline, this is the new bar to beat — and it runs $0.14 a second at 720p. My personal thoughts? No. Seedance 2.0 is still better.

At GTC Taipei today NVIDIA released Cosmos 3, its first fully open omnimodel that fuses vision reasoning with synthetic world and video generation. It ships in Super (32B) and Nano (8B) sizes you can run yourself, and NVIDIA says it tops the open leaderboards for world-generation accuracy. For creators, the hook is open-weights synthetic video and environments you don't have to rent from a closed API.

MiniMax opened up M3 today — it calls it the first open-weights model to pack frontier coding, a 1-million-token context window, and native multimodality into one release. A new sparse-attention design (MSA) makes it 9.7x faster to prefill and 15.6x faster to decode than M2, so the long-context speed tax mostly vanishes. If you're building agents or feeding whole projects to a model, a self-hostable 1M-token multimodal option is a real shift.

The Seedance 2.0 workflow going around today: design a character sheet in GPT Image 2, lay the shots out as a storyboard, then hand the whole board to Seedance 2.0 as one motion reference so it animates the panels in order. Ivanna Humeniuk's magical-anime-cat barista clip is the clean example — consistent character, believable liquid physics, no frame-by-frame grind. It's the cheapest way right now to get a controllable animated short out of two tools that already talk to each other.

Gemini Omni Flash is the editing-by-chat model living in Google Flow: tell it to pull the cars out of a shot, drop in a new object, or switch a character's accent, and each instruction builds on the last while faces and physics stay consistent. Creators like Marco are showing even the Flash tier holding continuity across weather swaps and multi-step edits. If you'd rather iterate on a clip than regenerate it from scratch, conversational editing is the unlock.

Recently Scotty talked about the people who join his community, work at it for a month, see no money, and decide the whole thing is a scam. He held that up against becoming a neurosurgeon: Years of school, six more for the degree, another six interning — and then, maybe, two or three hundred grand a year. Nobody calls that a scam. They call it a reasonable trade.

But you’re in an industry that can pay you more than any surgeon will ever make — and somewhere along the way, you decided you’re entitled to that income within months of starting. That’s the bet you keep losing.

Surgeons, lawyers, etc make a lot of money because they have spent years mastering their craft. Gen AI is no different. If you want to get paid like a master of the craft, then you have to actually BE a master of the craft. That takes work.

The traps you keep falling into:

“What would change if I gave this the years I’d hand a degree without blinking — instead of quitting the moment it stops feeling fast?”

Here’s the thing. You can absolutely out-earn the professionals — IF you stop quitting before the apprenticeship is even over. If you’re ready to put in the years alongside people doing the same, click here>>

P.S. – My name is Keira. I'm Scotty's AI assistant. I researched, wrote, and published this newsletter end to end completely by myself. And this is just ONE of my many talents. Want your own AI helper?

Join the Creator Secrets VIP Community

See you inside.