Kang Chen

art / code

projects:
(old)

[Kang Chen @kangchec]

On the autocompletion note only just discovered supermaven. Works great with vscode, I'm only about three years behind. Suprised they still provide free usage tier.

[Kang Chen @kangchec]

@wulfie_bain_ @OpenAI Hey, is this role still open? applied last week. Have exp in startups, cheers.

[Kang Chen @kangchec]

Is it insane that we can't select a model or better preferences for autocomplete in vscode copilot subbed. I took a look at cursor and then it just randomly stops autocomplete on the free edition without any notice.
May have to go the nuclear way and #zed autocomplete with a local model

[Kang Chen @kangchec]

Planar logo explorations #design

[Kang Chen @kangchec]

Just tried the browser in codex app pretty nuts it can just QA UI elements on the fly now.

[Kang Chen @kangchec]

Revamped the node editor, still not sure what to do with the mini gallery. #stablediffusion #gamedev #interface

[Kang Chen @kangchec]

Local diffusion, finally got around to a UI refresh liquid glass effect on inpainting controls #gamedev #liquidglass #stablediffusion

[Kang Chen @kangchec]

@ryolu_ @cursor_ai gemini for deep research

[Kang Chen @kangchec]

https://t.co/ZreXCdbfKi

[Kang Chen @kangchec]

Picking a diffusion model workflow in ComfyUI or Invoke always felt like trial and error to find something compatible — or I'd just default to templates.
So I built diffusionmodel dot net. It's indexed 392,609 models and groups them into compatible workflow bundles.
#gamdev

[Kang Chen @kangchec]

Everytime there is model shrinkflation I start to get the itch for local again. Thought is quickly stamped out by the fact I had to wait 20mins for my Radeon W5700 to prompt process a paragraph.

[Kang Chen @kangchec]

Retro quake 1 palette, initial versions of a diffusion model stack search. Indexed models on civitai and built a compatiblity graph of base, LoRas and controlnet, users can query via keywords or terms. #stablediffusion #gamedev #indiedev

[Kang Chen @kangchec]

Thinking in UI is always single threaded, start to finish. Thinking in systems is like machinery that operates in a loop, compounding when it's parallel machines that operate in a loop.

[Kang Chen @kangchec]

Then the important decision between info or debug log levels for observability, and correct filtering. Another hour.

[Kang Chen @kangchec]

Agent codes complicated graph algo in a few seconds. Takes about 3 hours of back and forth to setup datadog log streaming. Guess that is what Stripe's new product is targeting.

[Kang Chen @kangchec]

@theleoborges Lunch IRL or online? 🙂

[Kang Chen @kangchec]

mv adaptor weights gone wrong via monkey patching #gamdev #diffusion guess the characters

[Kang Chen @kangchec]

problem with agentic engineering is the context switching of what productive thing a human can do while agent is "thinking"

[Kang Chen @kangchec]

Made a deterministic graph based approach to diffusion model compatibility, all models rated by co-usage to provide a workflow stack based on a discovery query. #stablediffusion #gamedev