There's nothing to fix on this end, the old version is exactly the same pipeline. If you use the same prompts as v1.2 you will get the same output. The difference is, a few months ago the provided endpoint swapped the LLM that was used. This wasn't done on purpose, this was due to a massive pricing increase vs. the old LLM. That's where the difference in output comes from, not v1.2 vs. v2.0+. In order to get similar output, you'll need to use your own local LLM or pay for an endpoint that supports it unfortunately.