So more progress on my project. It actually properly waits for the screenshot to finish being made before trying to upload it. Aside from the one image that after two tries just did not want to translate shows more promise. Was using `qwen2.5-vl-7b-instruct` via their api but can easily point to llama.cpp like setup once the new qwen2.5-vl models can be hosted via llama-serve also would probably be faster as well.

(Sorry for leaving cursor on the game during the recording )
(Game is Doki Doki Pretty League for the PlayStation)

The flow is still not AS SMOOTH as the translation service you can already set up with #Retroarch in my opinion. But I feel I can get there with some more work.