If the AI stops mid-generation or the interface hangs without any output, the most likely cause is a browser-side issue. Clear your cache, reload the page, and try submitting the prompt again. Chrome tends to behave better than Safari for long generation sessions.
If you can't log in or get kicked out mid-session, start with the basics: check that cookies are enabled and that no browser extension is blocking requests. Logging out fully and back in usually resets a broken session token.
Sometimes a prompt goes through but nothing changes in the editor. This usually means the diff failed to apply cleanly. Try rephrasing the instruction more specifically — instead of 'fix the layout', describe exactly which element and what the expected behavior is.
The live preview iframe can fail to load if the generated code has a runtime error. Open the browser console (F12) and look for red errors — they point directly to what broke. Refreshing the preview panel alone often isn't enough; you need to address the underlying code issue.
If publishing to the live URL hangs or returns an error, check whether your project has unresolved build errors first. Lovable won't deploy broken code. If the build looks clean but deployment still fails, wait a few minutes and retry — it's occasionally a backend queue issue on their end.
When changes stop appearing in the linked repository, disconnect and reconnect the integration from project settings. Make sure the repo permissions haven't been revoked on the GitHub side — that's the most common reason sync silently breaks.