Ever wondered what it would feel like to learn a language from a native speaker - with no translations, ever? I built a browser extension using local Ollama and Stable Diffusion to mimic what I imagined it might feel like to learn as a child from a native speaker.

#llm #local #selfhosted #opensource #languagelearning
#browserextensions

https://blog.paractmol.nl/the-ollama-browser-extension-i-made-to-teach-myself-dutch/

The Ollama browser extension I made to teach myself Dutch

What would happen if AI became my native language teacher — not translating words, but explaining them through context, images, and exposure, the way a child learns from a parent? That question became the starting point for a small experiment.

Paractmol
@paractmol you know you can achieve MUCH better result with stablediffusion in 2026? there are sdxl and sd3, i hardly imagine they can generate something that ugly with reasonable settings (30 samples 7.5 scale). also there is flux (2 or more idk), it might be even better than sd. if you have 3060 or better, you should have no issues running better newer models. i got better pictures than this goalkeeper in 2023 with sd 1.5 or what was there
@ralen Oh, interesting! Thanks for sharing. I didn't know about it. First, I tried the free API to generate images to test out the concept. Then, I switched to Stable Diffusion and noticed a rapid decline in quality. I didn’t know what to tweak or change.
@paractmol sd 3.5 seems to be the most recent one, also try these flux models. you can use automatic11111111, invokeai or comfy ui. i used invokeai for my uni project and managed to vibecode some http server that received prompt and returned an image with it, it's pretty easy - just wrapping and redirecting
@paractmol Did you learn Dutch?
@ClickyMcTicker it helps to expand the vocabulary. I used it as a supplement.