Asher Crowe 🪺

@ashercrw

the signal is already in the room

Katılım Mayıs 2026

19 Takip Edilen19 Takipçiler

Sabitlenmiş Tweet

Asher Crowe 🪺@ashercrw·5d

I scrolled TikTok for 20 minutes last night and slowly realized something that made me put my phone down. None of these women exist. Not one. The redhead applying sunscreen on the balcony. The brunette in the green sports bra doing the heart shape with her hands. The blonde stretching in front of the bathroom mirror. The girl on the floor laughing into the camera while holding the bottle. Every single one of them is AI generated. Every account is fake. SunscreenAddict. ActiveGlow. concentration.7. obsessed. extra. Names designed to sound like real twenty-something skincare girls. Bios written to read like every other lifestyle creator on the platform. Posting schedules timed to mimic human behavior. The entire feed I was watching, the one TikTok served me when I searched a single product, was a coordinated swarm of synthetic influencers all selling the same bottle. Hundreds of thousands of views. Each. 512K on one. 358K on another. 232K. 200K. 199K. People in the comments tagging their friends. "Need this." "Where can I buy." Real humans, having real reactions, to women who do not exist. Let me walk you through what's actually going on here, because the mechanics are wild. Someone, or more likely a small team, generated a roster of fake "influencers." They picked diverse looks on purpose. Blonde, brunette, Asian, redhead, Black, mixed. Different body types. Different home aesthetics. Different lighting. Some on balconies. Some in bathrooms. Some in bedrooms. So the feed looks organic, like the product is having a moment with every demographic. Each "creator" got an account, a name, a personality, and a posting schedule. They probably look at their analytics dashboard the same way a real creator would. Then the same product gets reviewed across all of them. Same talking points. Same captions. Same hashtag set. Twenty different "real girls" all converging on the same recommendation in the same week. To the algorithm, this looks like a genuine trend. To you scrolling at midnight, it looks like every girl on the internet is suddenly obsessed with this one bottle. That's the trick. It's not that one fake account got views. It's that twenty of them did, simultaneously, creating the illusion of cultural agreement. Manufactured consensus. It's the oldest marketing tactic on earth, just executed at a scale and speed no human team could match six months ago. And here's the part that broke my brain. You cannot tell. I mean it. I have a developer eye for this stuff. I look at AI-generated content for a living. I had to zoom in on three of these to be sure. The hands give it away if you look closely. Sometimes the bottle label is slightly malformed. Sometimes the reflection in a bathroom mirror doesn't match. But on a phone, mid-scroll, at two in the morning, with one thumb? You'd never catch it. Nobody is catching it. The whole "AI looks fake" defense people had a year ago is gone. Dead. The current generation of image-to-video models renders skin texture, fabric folds, depth of field, and natural movement in a way that slides right past the radar. Three things to take from this, because the implications are bigger than skincare. The trust signal you used your whole life — "I saw a real person say it" — is no longer reliable on social media. You're being marketed to by ghosts. Every product category is about to get this treatment. Skincare is just the early adopter because the visual content is easy. Supplements are next. Then fitness. Then home goods. By next year, "influencer marketing" will be a phrase that needs a footnote. If you're a real creator competing for views right now, your competition isn't other humans anymore. It's a server farm in someone's apartment generating 100 versions of you while you sleep. I'm not saying any of this is good or bad. I'm saying it's already happening. Right now. On the device in your hand. And almost nobody is paying attention. Save this post. Show it to anyone who still thinks they can spot AI content in the wild. The girls in your For You page have rooms that don't exist, in houses that were never built, holding products they've never touched. Keep scrolling. But scroll knowing.

English

4.9K

Asher Crowe 🪺@ashercrw·5h

YOU'RE NOT GOING TO WATCH VIDEOS MUCH LONGER. YOU'RE GOING TO WALK INTO THEM. At 0:15 these guys dropped a breakdown of the thing that quietly ends flat video, and almost everyone who scrolled past it has no idea what they just saw. So let me decode it. It's called 4D Gaussian Splatting. Not 3D. FOUR. The extra dimension is time. Here's the part that broke my brain. Normal 3D splatting captures a room you can walk around. A frozen space. Beautiful, but dead. 4D captures the room AND the moment unfolding inside it. The volume of the space and the dimension of time, at once. It works by mapping millions of "Gaussian" points. Think tiny fuzzy particles of light and color, each one holding where it sits, what color it is, and how it behaves as you move. Map enough of them and you don't get a model of reality. You get a photorealistic, real-time reconstruction of it. A digital hologram you can stand inside. Now the part nobody's ready for. You control the camera. Not the editor. YOU. You can move it anywhere inside the captured space. Glide through a wall. Step behind someone mid-sentence. Pass your view straight through a person's hand. And here's the kicker: you can do all of that whether the moment is playing or FROZEN. Pause reality, walk around it, hit play again from a new angle. This is the closest humanity has ever come to recording a physical moment you can actually walk back INTO. Not rewatch. Re-enter. Now do the operator math, because this is where it gets loud. Flat video is a $300B+ industry built on one limitation: you watch from where the camera was. 4D deletes that limitation. Every category that pays for video is suddenly underpriced. Wedding films at $3K to $8K become walkable memories people will pay $15K+ for. Real estate tours stop being slideshows and become the listing. Sports replays where YOU pick the angle. Concert footage you can stand on stage for. Training, courtrooms, film previs, museum exhibits, every one of them sitting on the old format and not knowing the new one exists yet. The tech is real. The demos are public. And the number of people who can actually produce this for a paying client right now? A rounding error. That gap is the entire business. Same as it always is. The flat-video era isn't ending in ten years. It's ending in the comments under videos like this one, where the people who get it go quiet and start building. Bookmark this. In 18 months you'll either be the one selling it or the one explaining why you waited. 👇

Asher Crowe 🪺@ashercrw

x.com/i/article/2062…

English

Asher Crowe 🪺@ashercrw·8h

HE WORKED AS AN ENGINEER AT INSTAGRAM. Then he went home, picked up a camera, and started posting the most honest videos about tech money on the internet. Almost nobody is watching. His name is Mr. Zhang. A Chinese engineer who built at one of the biggest apps on the planet, then started saying the quiet part out loud. The video here is him stuck on a bug for days, asking how to optimize his code. Every engineer alive has lived that exact night. But the real gold isn't the code. It's the MONEY behind it. He breaks down what engineers make at the top Chinese giants. Huawei. Tencent. Baidu. ByteDance. His number: about 300,000 yuan a year. Roughly $50,000. Half of you just thought "that's it?" That's the trap. $50K there buys 3-4x the life it does here. He's describing a BETTER life, dressed up as a smaller number. And the real data goes even harder: Median ByteDance engineer → ~$89,000 Top ByteDance offers → ~$315,000 That SAME ByteDance, in the US → up to $1,180,000 Same company. Same code. The paycheck just depends on which side of the map you signed your offer on. Here's what nobody says out loud: a US startup paying $400K isn't buying better code. It's buying a city where 5 companies fight over the same 10 people. You're not paying for skill. You're paying for a zip code. But my favorite part isn't even the money. Every time he visits home, his parents find any excuse to humble brag. To the waiter. To the neighbors. "Oh, our son? He works in American Big Tech." That's the one currency that ignores the exchange rate. You can't arbitrage the look on a parent's face when they get to brag about you. The man isn't selling a course. No "6 figures in 6 months" funnel. Just a real engineer explaining the invisible rules of money, tech, and culture better than people who charge thousands for it. Watch the video. Then go down his whole feed. Save this one. I break down a creator like this every week.

English

141

Asher Crowe 🪺@ashercrw·10h

@EdenWood62747 This is the world we are living in brother

English

EdenWood@EdenWood62747·10h

@ashercrw this is both incredibly cool and slightly terrifying

GIF

English

Asher Crowe 🪺@ashercrw·11h

Here's the exact 3-tool workflow that makes fake AI girls stop your thumb mid-scroll. And the reason most AI UGC fails before the first second even loads. Check 0:07 to get it right. If you've been following the series, you've seen the swarm, the economics, and the face-building method. This one is about the last mile, the part that separates content that converts from content that gets flagged as obviously synthetic and scrolled past in 0.4 seconds. Save this. Here's the thing nobody wants to admit. A perfect AI face still flops if the content around it feels like a studio ad. The human brain is a finely tuned fake-detector, and it doesn't trip on the face anymore. It trips on the framing, the lighting, and the mouth. Fix those three and the illusion holds. The workflow, step by step. Step one. Build the model for the phone, not the camera. Open Nano Banana Pro and generate your model. Two settings do all the heavy lifting here. Natural lighting, never studio. Studio lighting is the number one tell because real people filming themselves at home don't have a softbox. And vertical iPhone-style framing, slightly imperfect, the way an actual person holds a phone at arm's length. The goal is not "beautiful." The goal is "filmed by a 23-year-old in her bedroom." Those are different aesthetics and only one of them converts. Step two. Put the product in the same hands. Run Nano Banana Pro a second time, but now generate the exact same model holding the product. The word that matters is "same." Consistency is the entire game. If the woman in the product shot looks even slightly different from your model, the brain catches it instantly. You're creating one persistent person who exists across every frame, holding the thing you're selling, in the same natural light, in the same vertical frame. Step three. Animate it and fix the mouth. Import the still into Cling 2.6 and animate it. This is where 95% of AI UGC dies. They animate the body and forget the mouth, and you get that floating, dead-eyed, slightly-wrong movement that screams synthetic. The fix is proper lip sync. When the mouth matches the words, the last alarm bell in the viewer's head goes quiet. That single step is the difference between 200 views and 200,000. That's the whole pipeline. Build for the phone, keep the model consistent, sync the mouth. Three tools. One persistent fake person who can sell anything, in any lighting, holding any product, talking directly to camera like she's your friend. Now zoom out and look at why this matters so much right now. Every step in this chain used to be a separate hire. The model was a person you booked. The product shot was a photographer. The talking video was a creator with a personality and a rate card. The lip sync was a post-production house. Four humans, four invoices, four schedules, weeks of turnaround. This collapses all four into an afternoon at a laptop. The compression of cost and time is the actual story. It's not that AI can make a face. It's that AI just deleted an entire production supply chain and handed it to one person with a $20 subscription. To make the scale of that collapse concrete, I mapped it out. This is what the workflow looks like when you chart cost against realism at each stage of the pipeline, and overlay it against the traditional production path. The gap between the two curves is the entire business model.Look at the gap between those two curves. Both paths arrive at the same place, content a human can't tell is fake. One gets there for $11,500 and three weeks of coordination. The other gets there for $23 and 47 minutes. That white space between the lines isn't a chart artifact. It's the entire arbitrage. It's why someone in a bedroom can now outproduce a brand's whole creative department. And here's the uncomfortable part nobody in my comments wants to sit with. The realism axis is going to keep climbing while the cost axis keeps falling. The teal line is moving up and to the left every single model release. There is no version of the next 12 months where this gets harder or more expensive. It only gets easier and cheaper, which means the flood you're already seeing is the trickle, not the wave. The window where knowing this workflow is rare is the window where it's valuable. Once every brand's intern can do it, the edge is gone and it's just the new baseline cost of doing business. Right now it's still a secret. That's the whole opportunity. If you want the exact workflow, the Nano Banana Pro settings, the consistency trick for keeping the same model across every frame, and the Cling 2.6 lip sync setup, comment "UGC" below and I'll send the full breakdown. Save this post. In six months you'll either be the person running this, or the person it's running on.

Asher Crowe 🪺@ashercrw

The reason every AI influencer looks fake is two fixable mistakes. Here's the exact process to make one nobody can tell is fake, plus how people are turning these into $10K/month. Last two posts I showed you the swarm of fake skincare girls and broke down the brutal economics behind them. The number one question in my DMs since: "okay, how is the realistic version actually made?" So here's the full pipeline. Save this before it gets patched. Everyone building AI models fails at the same place. Pause at 0:40. They generate a face straight from a text prompt, the skin looks like wax, the eyes look dead, and the whole thing screams synthetic in the first half second. Then they blame the model. The fix has nothing to do with which generator you use. It's two components. Your reference image and your prompt. Get both right and the output crosses the uncanny valley completely. Here's the process, step by step. Step one. Build the reference face, don't generate it. Go to Pinterest. Find three high-quality photos of faces with features you want your model to have. Not one. Three. This is the move 90% of people skip. A single source gives you a copy. Three sources blended give you a person who has never existed and can't be reverse-image searched back to anyone real. Take those three to wavespeed(.)ai and run them through Google's Nanobanana Pro edit tool. Upload all three, feed it the merge prompt, and it fuses them into one coherent high-resolution face. This single image becomes the DNA of your model. Every future photo of "her" references back to this. That's how you get consistency across hundreds of posts instead of a slightly different woman every time. Step two. Steal the composition, not the person. Go to Instagram. Find a photo with a pose, lighting, and vibe you want to recreate. Screenshot it. Drop that screenshot into ChatGPT and ask it to generate a full JSON prompt describing everything in the frame. Pose, camera angle, lighting direction, background, outfit, mood, lens. Or run it through a custom GPT built for this, which spits the JSON out clean in one shot. You're not copying the influencer. You're extracting the recipe that made their photo work and handing it to your own model. Step three. Combine and finish. Take your blended reference face. Take the JSON prompt. Feed them together into your generator. Then add the finishing phrase at the very end of the prompt, the realism trigger, the line that tells the model to render skin texture, micro-imperfections, and natural light instead of that glossy AI sheen. That one phrase is the difference between "obviously fake" and "wait, is she real." That's the entire system. Reference plus prompt plus realism trigger. Three tools, none of them expensive, most of them free. Now the part you actually came for. People are running these models as full Instagram personas and monetizing them four different ways. Brand deals from companies who never realize the "creator" isn't human. Affiliate commissions on products the model "uses." Paid subscriptions on her content. And selling the entire system to other people who want their own. The realistic ceiling people are hitting with a single well-run AI model on Instagram is around $10,000 a month. Not from one income stream. From stacking them on top of an audience that thinks they're following a real person. Think about the cost structure for a second. No model to pay. No photographer. No studio rental. No travel. No product samples. No flights to Bali for content. The entire operation runs on a laptop and a few subscriptions while you sleep. The margin isn't good. It's almost the entire revenue. And the timing is the whole story. Eighteen months ago this was impossible, the outputs were laughable. Six months ago you could spot them by the hands. Today, with this exact reference-plus-JSON method, they pass. A year from now everyone will be doing it and the early audiences will already be locked up by the people who moved now. The window where this is easy and uncrowded is open right now and it is closing by the week. If you want the full breakdown, comment the word "group" below and I'll send you the free training that walks through the entire $10K/month AI model system end to end, the exact prompts, the realism trigger phrase, and the monetization playbook. Save this post. Send it to the one friend who keeps saying they missed the boat on every trend. This is the boat. It's still at the dock.

English

556

Asher Crowe 🪺

Keşfet