Update: Regarding the issue of Lobster not being able to read images using Minimax M2.5, I’ve figured out a native solution. In the Openclaw.json file, there’s an image model setting. Just set it to an affordable image-reading model, like Claude Haiku 4.6 or GPT55 mini. This way, for text communication, it’ll still use Minimax M2.5, but for image reading, it’ll switch to Haiku 4.6. As for how to set it up, just ask your own Lobster—it’s pretty simple. No need to install anything complicated. Blind Lobster finally regained its vision