Meshcore-open developers implemented image transport over LoRa. Images are downscaled to 512x512px, then encoded by AEIC codec [1] and then split into 163B chunks to fit into LoRa packets.
The codec requires a 800MB LLM which occupies several GB of memory when running. To transport small images over the mesh which will IMO fail in most real life scenarios because good luck delivering multiple large packets via multi-hop paths.
This implementation is about the same size as the whole code base of #Meshy. This feels like an extreme case of "because we can" enabled by subsidized tokens.
