Scott Ashton

@kettanaito Would love to get your thoughts on a zero copy IPC library I'll be releasing soon - has very http/express-ey feel.

I have no idea how ya'll building apps on Electron when IPC has zero type-safety and is an absolute nightmare to use. Community packages that "solve" this problem don't really solve it (unidirectional tRCP !== IPC). This is very, very sad.

@andrew_n_carr Yes 100+ 😂. It's driving me INSANE though how often I have to "kick" cloud Claude to keep going. It's gotten way worse the past week. Continue... Continue... Stop...

Checking all my side projects like

@staysaasy What's the ratio of system problems (e.g. restructuring) to human problems (e.g. firing) when underperformance happens?

People don’t realize how insanely important it is to have always on mechanisms for dealing with underperformance. Say you’re at a 1000 person company. Shipping a product end to end might involve: PM, Design, PMM, Engineering, the exec team, Finance, Pricing & Packaging, Customer Success, Sales. If 1/10 people underperform (very conservatively), every major project you do will have to deal with underperformance. Losing companies just shrug and watch the world go by. Things miss and people blame others teams. Winning companies have real time oversight and detection of issues and fix them before they become too big of a problem. If you want to ask how well you company is set up to win, ask yourself how fast it’d be able to detect and mitigate underperformance on a major project. Great = days Good = weeks NGMI = months

@blazejkustra_ @wcandillon Thanks for finishing this out for me Blazej!

🚀 react-native-wgpu v0.4.0 now runs in the browser too! 🔥 One GPU API powering iOS, Android, Web and visionOS Thank you @wcandillon for this amazing library!

@tjzeldev That would be so awesome. The magic of worklets is working across multiple threads IN THE SAME FILE. What you guys have done is magic. The ergonomics of Web Workers make them almost unusable IMO and worklets could be a true game changer for concurrent programming in Javascript.

@Scottmas1 It's on our road-map but first we want to bring Web-Worker-like API to React Native 🙈

We just made sure that no native specific code is loaded for `react-native-worklets` in web environments. Static size is over 10x smaller. Reanimated is next.

@Ademola_4life That looks reeeal nice Ademola, I've got hand it to you.

This is the last one 🥲 I just love the volumetric lighting so much! Don’t blame me 😂. Volumetric lighting + ao from ssgi + bloom #threejs #webgpu

@swmansion The download link is busted after signing up for the newsletter!

Building a React Native app with gesture-driven animations – like drag-and-drop, sliders, or scroll? 📱 Download our Reanimated 4 Cheat Sheet for plug-and-play API overviews + ready-to-use code snippets. 🚀

@aarongrider @threejs You using react-native-wgpu by @wcandillon?

With SceneKit deprecated, @threejs + React Native is the best tool I am aware of for building immersive 3D experiences into native apps.

@bbssppllvv How do you do this in Swift code?

Surprised no one mentioned it yet: iOS 26 is using HDR in the UI. Not just for photos or video, but for interface elements too. That means some elements can be 'brighter than white'

@vc_broke I don't know where I first found out about your newsletter but this article nails startups SO well. The line is blurring between conviction and BS.

@IRCSS Oh wow, thank you! This looks awesome. Have you had much trouble generating proper albedo maps? My basic experiments with generating them ALWAYS seem to result in albedo maps with lighting baked into them

@Scottmas1 the comfy side is very simple, here is an example workflow I use. there is a more complicated one that also takes a confidence map and packs in the alpha before projecting back in blebder gist.github.com/IRCSS/3a6a7427…

Blender+ SDXL + comfyUI = fully open source texturing workflow: 1. cameras in blender 2. render depth, edge and albedo map 3. In comfyUI use control nets to generate texture from view 5. project back and blend based on confidence (surface normal is a good indicator)

@IRCSS You have a Comfy workflow? This is really cool

generating each view takes only a couple of second on a good gpu, so you can be done with modeling in a couple of mintues

@Juggernaut_AI What was used for that animation? Never heard of runnit

Sherie sent this in and we had to share! Juggernaut Pro Flux + Runnit Scenes = 🔥 ⚡ 🔥 Love seeing what the community creates with our models and tools. #Juggernaut #RunDiffusion #AIart #Gaming #Anime

@CSM_ai You guys freaking need a better marketing team! This is SOTA for sure. This should be all over the freaking internet!

Our new 3D PBR meshes make production-ready content for logos, websites, and game marketing content exponentially easier. 🚀🎨🌐

@RunDiffusion @Juggernaut_AI That's amazing news!

@Scottmas1 @Juggernaut_AI While we always want to support the opensource community so we are currently working on a version for the community we hope to be ready in about 4 weeks if licenses and testing work out as we hope.

The Juggernaut Flux series has arrived on Runnit via @RunDiffusion! Need API access? Fal & Runware have you covered. And don’t sleep on the upcoming community model make sure you have alerts ON! 🔥 🔗 rundiffusion.com/juggernaut-flux #Juggernaut #RunDiffusion #Runnit #AIArt

@NathanielGivens This 100% seems accurate.

Obviously this is not inevitable. But Trumps "peace through surrender" strategy is making World War 3 *more* likely, not less likely.

This is not how you end a war. This is how you start a bigger one.

@ostrisai Am I right in thinking that differential diffusion LORAs work best with character LORAs, not style ones? It seems like it would be hard to generate a good alpha mask on a style

@Scottmas1 Fore this, just the toggle, a trigger word, and the class name. For instance, for this I did a trigger_word of "ohwx woman" and for the dop class I did "woman". Nothing needs to be done to the images/dataset.

Added Differential Output Preservation to AI Toolkit and the ui. It is ~3x slower to train, but preserves class knowledge and seems to produce much better results and prevents overfitting.

@ostrisai Wow that’s awesome!