𝕏i Chen 陈悉

Pen spinner. Rendering Engineer at @2K @31UnionOfficial. Worked on @CallistoTheGame, Nintendo Switch, CoD: WWII & MW. Cyclist, composer, archer. Opinions my own

[𝕏i Chen 陈悉 @hypernewbie]

Alpha version of VEFontCache, my single-header minimalist GPU font rendering library, is up on github!!
https://t.co/t2EeCgx2Vd

#rendering #fonts #gamedev #vertexengine #github #OpenSource

[x.x @xxgameroom]

Eden's Eclipseはこんな感じのシューティングゲームです。STEAMで販売中です！
https://t.co/C8U6erIt41
#shmup #シューティング

[𝕏i Chen 陈悉 @hypernewbie]

Quick demo game to show off engine!

Art = AI python-script procedural.
Gameplay = painstakingly directed AI, step by step over ~11 hours.
Music = entirely human made by me.

Mostly GLM 5.2, Sonnet 4.6. Gemini 3.5 Flash on assets.

#STG #Shmups #BulletHell #GameDev #IndieDev

[𝕏i Chen 陈悉 @hypernewbie]

@sawyerhood On it

[𝕏i Chen 陈悉 @hypernewbie]

@tsuishi I'm using them to detect safe spots in static patterns!

[𝕏i Chen 陈悉 @hypernewbie]

Anyone recognise this one?
#STG #Shmups #BulletHell #GameDev #IndieDev

[𝕏i Chen 陈悉 @hypernewbie]

@SkylerMiao7 Definitely the buggiest open source model, not sure about best but it is pretty good in the 1% situation it isn't spilling CoT everywhere

[𝕏i Chen 陈悉 @hypernewbie]

@SkylerMiao7 And also very very buggy. When switching to / from other models, it completely breaks the adaptive thinking and often you end up with a non thinking model. When you use m3 by itself it's good.

[𝕏i Chen 陈悉 @hypernewbie]

@SkylerMiao7 It's so stupidly slow

[𝕏i Chen 陈悉 @hypernewbie]

@_mohansolo In mixed crlf / lf cases, the AG harness is now so broken the only way the model can edit code is via python calls. Plz fix. And call this beta software like it is while ur at this. It's nowhere near 2.0 this is 0.2.

[𝕏i Chen 陈悉 @hypernewbie]

@SebAaltonen I tried it and it generated some serious amount of total slop that I had to immediately delete. Not impressed at all so far with agentic. Sonnet 4.5 is LEAGUES better for agentic coding.

Very good "ask" or codesearch model and very well behaved model otherwise.

[Sebastian Aaltonen @SebAaltonen]

My "No Graphics API" blog post is live! Please repost :)

https://t.co/uL8HsfIlNV

[𝕏i Chen 陈悉 @hypernewbie]

Vlfs - Single python script vibecoded LFS for your git repo.
Single-version large file download script, designed for prebuilt libraries + game assets. gdrive or R2.

https://t.co/MWqmmlqBFO
https://t.co/4V11knCRNN

#Python #Git #OpenSource #GameDev #IndieDev #DevTools

[𝕏i Chen 陈悉 @hypernewbie]

@_StanGirard is it ACP protocol?

[𝕏i Chen 陈悉 @hypernewbie]

https://t.co/M6ZX7fTCez

@Microsoft This is what people mean when they call you slop. An entire 3 paragraph of obviously AI written word soup describing what is essentially one button to generate a AI commit message from git diff. A useful AI feature actually.

[𝕏i Chen 陈悉 @hypernewbie]

Vrdg - Python script to generate rendergraph from C++ comment declarations.
https://t.co/Eg1FJKZUxv
https://t.co/Jnyr7MOXwI

[𝕏i Chen 陈悉 @hypernewbie]

@jendrikillner would love a visibility boost to inspire someone to write a good HL Vulkan RHI.

[𝕏i Chen 陈悉 @hypernewbie]

Vrhi: AI Slop Vulkan RHI

Many gfx APIs don't get basics right. Transient staging, cached FBOs/PSOs/dsets, RT. Instead they force invented concepts onto a rock.

I wrote 1 using AI so you can write 1 better:
https://t.co/qF8Uij3pjp
https://t.co/xdRvshjaeY

#VulkanAPI #Rendering

[𝕏i Chen 陈悉 @hypernewbie]

Load game assets at DDR2 RAM speeds, v1.04 with sharded file support for fast random access!

Vark: semi-vibecoded archive library & CLI tool using LZAV compression.

https://t.co/3tdhmrE3Nq

#GameDev #IndieDev #gamedevelopment #indiegame #indiedev #OpenSource

[𝕏i Chen 陈悉 @hypernewbie]

Load game assets at DDR2 RAM speeds!!!

Vark: archive library & CLI tool using LZAV compression.

Note: Vark is vibe coded, closely instructed and architected by a SWE (me).

https://t.co/3tdhmrE3Nq

#GameDev #IndieDev #gamedevelopment #indiegame #indiedev #OpenSource

[Omar 🍋 @ocornut]

Dear ImGui now embeds a vector/scalable version of the classic font by default, manually subsetted/optimized so it fits in 18 KB (opt-out possible). Embedding ensures easily and readily usable in all contexts. Should default to new one on most setups. Here's a compare of both: