正在加载视频...

视频加载失败

I asked Opus 5.5 to explain camera focus by building an interactive lens lab Here's what it came up with after 1 hour 26 minutes in one shot, $25.66 API cost Move the focus ring and you can see the glass elements shift the sharp plane through the scene

3,573,453 次观看 • 1 个月前 •via X (Twitter)

52 条评论

Jason Farris 的头像
Jason Farris1 个月前

Ok I’m definitely showing this to my claude like wtf bro you’ve been working on this saas for weeks

Ryan Sael 的头像
Ryan Sael1 个月前

😂

Ryan Sael 的头像
Ryan Sael1 个月前

if you're new here. I regularly post similar stuff:

Ryan Sael 的头像
Ryan Sael1 个月前

Thank you to all of you 1000+ visitors in the first 4hours of posting this. there are hundred of people watching live now. Wasn't expecting this tbh

Nick Halverson 的头像
Nick Halverson1 个月前

What a time to be alive @maudegregori @cagsondotsh @lastmjs @jacob

Ryan Sael 的头像
Ryan Sael1 个月前

@maudegregori @cagsondotsh @lastmjs @Jacob yep. this stuff can take days or weeks now can be one prompt.

Cosmic Raven 的头像
Cosmic Raven1 个月前

The upside-down projection detail is what sells it for me. Most demos skip the actual physics. Curious what it used for the 3D side, did it handle the sharp plane at f/2 vs f/16 without cheating?

Ryan Sael 的头像
Ryan Sael1 个月前

it's too loyal to cheat. but happy for you if you're sold on that detail.

Toivo Mattila 的头像
Toivo Mattila1 个月前

wtf 🤯 how 🫪 very impressive can you share the prompt or something?

Toivo Mattila 的头像
Toivo Mattila1 个月前

Got it, thanks And interesting approach I've kept my different projects isolated but sounds like that's not the way to go necessarily

Rugwed 的头像
Rugwed1 个月前

the $25.66 figure is what makes this land. an interactive explainer with real glass-element refraction for less than a pizza. curious if you gave opus the optics model or it just knew how lenses behave

Ryan Sael 的头像
Ryan Sael1 个月前

it knew the cost of pizza and adjust the tokens budget to one shot this

Smitty Mc'Westurbergin 的头像
Smitty Mc'Westurbergin1 个月前

super excited for 3 days from now when the same prompt looks like dogshit

Ryan Sael 的头像
Ryan Sael1 个月前

hopefully not

IngressLog 的头像
IngressLog1 个月前

I cannot believe someone would pay $26 for that.

Ryan Sael 的头像
Ryan Sael1 个月前

you must be new here

Christian Bolivar 的头像
Christian Bolivar1 个月前

This is amazing! Great job!

Ryan Sael 的头像
Ryan Sael1 个月前

Thanks Christian!

Jeremiah 的头像
Jeremiah1 个月前

Can't believe we're living in a time for one-shot interactive prototypes like this. Insane

Ryan Sael 的头像
Ryan Sael1 个月前

yes! what a time

Josh Ertik 的头像
Josh Ertik1 个月前

give us the prompt or try on other model

Ryan Sael 的头像
Ryan Sael1 个月前

my prompt usually include asking it to scan trough my files and some unpublished projects. so perhaps may not work for everyone. what you want to see next? or what model to try it on?

Divvyansh Gandhi 的头像
Divvyansh Gandhi1 个月前

Best demo of new opus on the tl yet

Ryan Sael 的头像
Ryan Sael1 个月前

Really?? Thanks Divyansh

Mukunda 的头像
Mukunda1 个月前

slick build. The bigger story is what it unlocks: one-shot agentic builds turn every physical product into an interactive explainer. Opticians demoing progressive lenses, camera stores showing what a focus plane actually is, museum kiosks, showroom floor screens, STEM classrooms. That used to be an agency, five figures, six weeks. The $25.66 is the part I can't make balance. 86 minutes of continuous agent runtime is roughly 150+ API calls re-sending a growing context - call it 15-25M tokens through the pipe. At @AnthropicAI's published Opus 5.5 rates ($4/MTok in, $20/MTok out), that bills $70-140 uncached. $25.66 only works if ~90% of those tokens were cache reads at $0.40/MTok. Hand the same run to a rookie - bloated context, no cache discipline, 2-3 restarts after it derails - and the same artifact bills $200-500. The gap between $25.66 and $500 isn't the model, it's the operator. So how did you get it down to $25.66 - prompt caching doing the heavy lifting, or something else? Tight context resets, small working context, batching? Genuinely curious. And if the honest answer is "it really just costs $25.66 now" - then Claude starts crying, because Opus stops being the money-making machine and starts actually helping devs.

Nate 的头像
Nate1 个月前

Insanely cool!!

Road to 100 (good) Apps 的头像
Road to 100 (good) Apps1 个月前

looks really cool!

Ryan Sael 的头像
Ryan Sael1 个月前

thanks! good luck with your 100 apps. I think they can be one shotted in a single prompt too. just put them on loop haha

Road to 100 (good) Apps 的头像
Road to 100 (good) Apps1 个月前

haha i just made my name this.. maybe i should make the goal atleat $100 profit from each app or something so its not just stat padding

Ryan Sael 的头像
Ryan Sael1 个月前

aim for more figures. manifest early :)

Aapakari 的头像
Aapakari1 个月前

Nice glass shift detail. Maybe add aperture to tweak depth of field?

Ryan Sael 的头像
Ryan Sael1 个月前

those aperture are clickable too. have you try?

Ryan G 的头像
Ryan G1 个月前

wow this is crazy good! Thanks for sharing this

preset127User 的头像
preset127User1 个月前

Holy crap.

zackbuilds 的头像
zackbuilds1 个月前

This is a neat way to teach focus: drag the ring, watch the sharp plane move. Did you model the actual optics, or fake the path to keep it snappy?

Muhammad Rayyan 的头像
Muhammad Rayyan1 个月前

This is sick brooo!

Won Hundred 的头像
Won Hundred1 个月前

This is a great educational tool.

onchaintom 的头像
onchaintom1 个月前

Wowww

Ozark HillBilly 的头像
Ozark HillBilly1 个月前

after a few years in college that's how i view most things in life in my brain....

Ryan Sael 的头像
Ryan Sael1 个月前

🤔

Simon 的头像
Simon1 个月前

Is this done with blender or unreal engine?

TheRealKlaudie 的头像
TheRealKlaudie1 个月前

amazing!

Anurag pm 的头像
Anurag pm1 个月前

now build one that explains why my code doesn't work

AndreasK 的头像
AndreasK1 个月前

pretty animation but wrong. isn’t it a focus area and not a 2d focus plane on higher apertures?

Am_Tortol 的头像
Am_Tortol1 个月前

This is nice tbh

Ryan Sael 的头像
Ryan Sael1 个月前

cheers!

dnis 的头像
dnis1 个月前

amazing bro

Ryan Sael 的头像
Ryan Sael1 个月前

appreciate you!

Hafiz Siddiq 的头像
Hafiz Siddiq1 个月前

One-shot builds are impressive, but the $25.66 is the real product signal. Can the next pass change one optical assumption without rewriting the whole lab?

0xAIBrain 的头像
0xAIBrain1 个月前

The strongest part is that explanation becomes manipulation. A good eval would test whether users can predict the next focus shift before moving the ring, then transfer that intuition to a real photo. Interactivity is only teaching when it changes what the learner can do.

Mr B1 的头像
Mr B11 个月前

This can be used in lot of things

Ryan Sael 的头像
Ryan Sael1 个月前

I bet

相关视频

It is becoming f*cking dangerous: Anthropic shipped 3 separate models, but when you stack them like precision lenses inside one pipeline, the error collapses 70% before a single line hits production. I still do not understand why nobody is talking about this. Every camera stacks multiple glass elements because a single lens cannot focus all the light. Claude's 5.5 lineup, 3 models, 3,473 weights, works exactly the same way. So why are you still pushing every task through one model? Stop treating Haiku, Sonnet, and Opus as rivals. Start assembling them into one self-correcting lens. Here is the workflow: Haiku 5.5 is the front element: scans 5,630 files and compresses them into 22 signals. Sonnet 5.5 focuses those through 26 correction layers into 92 to 833 lines of code. Opus 5.5 runs 32 final checks across 7 risk metrics before anything ships. One cycle drops eval loss from 0.0830 to 0.0244. Like swapping a kit lens for stacked L-glass. Step-by-step setup: 1. Open Claude Code with Opus 5.5 as the primary session 2. Spawn a Haiku 5.5 subagent locked to read-only file search 3. Spawn a Sonnet 5.5 subagent locked to code writing and tests 4. Add one CLAUDE.md rule: Opus plans first, reviews last, never touches raw files 5. Drop a real task and watch the loss curve flatten in real time Takes 10 minutes. You will never ship unreviewed code or burn Opus tokens on file search again. Tap the link below for the full guide Make sure to save this post for later

shmidt

44,154 次观看 • 15 天前