KRUX

"The future is already trained. Most people just haven't run inference." - Krux

[KRUX @kruxael]

it will never be this easy to start again

most people will realize it too late...

[KRUX @kruxael]

@AmirMushich reframing is a generation problem, not a crop problem. the hard part is keeping the new edges consistent frame to frame so the extended space doesn't shimmer. if it holds on fast cuts, every label's back catalog gets a second life in vertical

[KRUX @kruxael]

@rewind02 plain english triggers + wait rules is the missing piece

most agent demos die the second something needs to pause for a human

[KRUX @kruxael]

@emollick funny part is the tell isn't any single comment, it's the sameness across five. slop is a pattern problem, not a text problem. x already has that signal, it just ranks engagement over it

[KRUX @kruxael]

@rewind02 open weights + open skill + the whole edit history shared?? bookmarking this one

[KRUX @kruxael]

@rewind02 😼

[KRUX @kruxael]

how i'd build a $5k/month faceless content page with claude this weekend:

> pick one niche: money, fitness, history, tech
> install 6 free animation skills in claude code
> one line of text = one finished reel
> /graph for growth stories, /map-route for travel, /stick-figure for "why did…" explainers
> post 3 a day on tiktok, reels and shorts
> no face, no camera, no editor
> monetize: brand deals, affiliate links, sell the reels to other pages

claude writes, animates and renders. you just pick what to post

where attention goes, money flows..

and btw this works for any niche

follow + comment "ANIMATE" and i'll send you all 6 skills

[KRUX @kruxael]

@EHuanglu the consistency across cuts is the real jump here. pair this with a rough blender blockout for the camera and it's basically a one person animation studio

[KRUX @kruxael]

@emollick routing every us city is basically a giant shortest path problem. curious if opus wrote its own graph code or leaned on openstreetmap plus a routing lib

[KRUX @kruxael]

@swyx ted chiang really wrote the roadmap. hoping we skip the ending

[KRUX @kruxael]

@javilopen if editing 0.016% of the weights can 6x math scores, fine-tuning as we know it starts looking very expensive

[KRUX @kruxael]

4,600 phones + $415k = 15 months in jail..

a 17-year-old in zhejiang, china did the math

> he bought 4,600 phones
> wired them to one pc with cloud control software and vpns
> every phone became a fake viewer
> livestreamers paid him to look popular
> ¥6.65 per phone, per day
> under 4 months later: ¥3M in the bank (~$415k)
> then the prosecutor showed up
> 15 months in jail, ¥50k fine, 17 more people under investigation

his mistake wasn't the phones. it was what he sold

fake attention is a crime. real attention is a business

the legal version of the same machine:

> no product needed. tiktok shop affiliate pays you a cut of every sale
> no 4,600 phones. 5–10 accounts, one niche each
> claude writes the scripts. 10 hooks per product
> ai generates the videos. same character, every day
> every video has the product linked
> after a week you kill what doesn't sell and clone what does
> once a product converts, an ai host sells it live 24/7
> an ai avatar of luo yonghao already sold $7.65M in one livestream

10 accounts + real buyers = a business

[KRUX @kruxael]

@caren633881 for sure

[KRUX @kruxael]

seedance 2.0 turned my grey blender blockout into a $50k sportswear ad...

let blender own the motion and seedance own the look

here's the workflow:

- step 1: block out the scene in blender with primitives: a stadium bowl, floodlight towers, a pitch plane. no materials, no lights
- step 2: get the hero product as a clean mesh. model it or generate it in tripo from 1 photo, grey clay is enough
- step 3: animate only the camera, one shot per clip: aerial flyover, slow push-in, low orbit, side tracking
- step 4: export each shot with viewport render animation (solid shading, 24 fps, 4–6 s). this is your motion reference, not your final
- step 5: in seedance 2.0 add the clip as a video reference + 1–2 photos of the real product as image references
- step 6: prompt only the look: "red football boot with white stripes, night stadium, volumetric floodlights, low fog, cinematic commercial". one effect per shot: molten glow, fire ring, shattering glass, lightning
- step 7: cut the best takes on the beat and add the title in the edit. never ask the model to type your slogan

how does it work:
> blender: layout, scale, camera, timing
> seedance 2.0: materials, light, smoke, vfx
> you: pick the takes

the mistake everyone makes:
>prompting the camera move
> describe a dolly in text and you get a random one

animate all in blender and you get the exact one

[KRUX @kruxael]

@a1lon9 @Pumpfun this one need to fight there too doing some magic lol

[KRUX @kruxael]

@Pumpfun i think $WIZPUG can also fight there lol

[hammi @hamm_ixx]

Pump this is DOG

Exw3S5tWiEUdZis8c8ksJt6xdymVefhg98b2vCvbpump

Have you seen $WIZPUG yet, @Pumpfun?

[KRUX @kruxael]

@Gustafssonkotte @a1lon9 yo bro

yep, the same strategy and chart movement

[KRUX @kruxael]

@a1lon9 https://t.co/BVRUvThVz3

new dope for this halloween