SoShal We ain't nexopia.
DE
Deck @decklyndubs
Nov 8, 2026
Public
Sprite sheet test for on prem "software factory." Entirely built on a home server. Image generation for the character concept from qwen image, it was turned into a 3d model by a Chinese model, Claude used blender to rig and adjust the frame, itch free animation pack used, back to blender for capturing all of the different frames and angles. I was spending a lot on Gemini and meshy for game dev. Should be able to do it all local now. Claude is the final judge. Because it's free basically, the one thing you can do it say "build 100 models 100 times for the same character, try it 100 different ways and let me pick the best one." That's the real power - it isn't to replace frontier models, it's to augment them and let them try over and over until it's right. I was spending like $30 a day on meshy throwing away stuff that didn't look good. Now Claude can just get local model instruction to build characters on my machine for 24 hours straight and then have it judge them all and let me make the final picks. Next up - I have the factory designing a game based on another couple projects i have worked on (graphic novel/game concepts.). Been running for about 36 hours now on the game design and building a story in a world I had turned into a graphic novel with gpt. I have to get permission to publish things from the employer but then we are going to sell some games!!!
Post media
DE
Deck @decklyndubs
Nov 8, 2026
Public
RAM/NAND is getting bought for data centers and prices are getting to "unaffordable for normies." I started to build an EPYC server w/ GPU capacity to get my bills down and got around to verifying the capabilities. The idea would be to basically cut all of my subscriptions/token spend except for something like $20/month Claude sub. Here is a report on findings. I ended up only needing 32gb VRAM for most capability (expert offload onto system ram) and checked current models to see what I could do for vision, 3d model gen, rigging, animation etc to build a fairly fully featured "software factory" that I can operate for like $40/month in a cheap fronteir model sub and hydro. The fronteir model is the ultimate quality gate and facilitates amongst coding models that run local (125B param MOE currently - runs at 40tok/s decode w/ a 128k context, much faster for general chat) and then TTS, image generation, and 3d model generation from sprites. Results are really good. Here is map for how to do it, how it works, where the weaknesses are, and how to fill them in. Had a few nay sayers along the way tell me local model usage was too slow blah blah but I managed to put this server together for less than $4kUSD. https://claude.ai/artifact/WB8cq8jcbLRD6MAR9k9kBr
@skrinkles
Kiki @skrinkles
Oct 18, 2026
Public
Flatland is a novella that contemplates the struggle of introducing new ways of thinking and the suppression of new thought by authority figures for their own benefit.
DE
Deck @decklyndubs
Oct 17, 2026
Public
hello world