/tech/ - Soyence and Technology

Local LLM: https://ollama.com/
Local Gen AI: https://comfy.org/

>>34904
Is there a local model that lets me make giga gifs?

>>34905
ComfyUI has video generation templates that use local models but it may not be as good as cloud options out there and it might be way too slow for what it's worth unless you're rich as fuck and have a super fast rig specifically built for local AI. Most 'teens and giggers just use cloud-based options with PixVerse being the most popular option for that

i use jan.ai on my pc. its good if you dont want to fiddle with configs or networking at all

>>34905
change extension to .json, this is a comfyUI workflow, (should prompt you to download the correct models - do note this takes up like 40gb of disk space (preferably install comfyUI on another drive)), I assume you are vram constrained, anyways ltx-2.3 is the best open source ai video gen model, for me on my pici it took like a minute per second (although i had very little ram at the time)

and i guess for local LLMs, the best you can reasonably run on consumer hardware is qwen3.6-27b (at least 16gb of vram), however the one you would probably use is qwen3.6-35b-a3b moe (offloaded to the cpu), there has been a few projects like colibri which take the idea of loading experts into the ram / vram of your gpu from the disk, (also theres some other new snca project that loads the experts from your ram to your gpu), for smaller models that can realistically run, use qwythos-9b-v2-gguf, I guess you can fuck around

>>34917
I have a 9060xt 16gb

LOCAL AI
IN YOUR FUCKING SCULL
IT'S CALLED A BRAIN
YOU FUCKING RETARDED APE
OPEN YOUR SCALP AND TALK TO YOUR LOCAL LLM, BITCH