I'll take this request.
Up for discussion: what should happen when there are multiple SaveImage nodes in the workflow? We don't know which one saved the image. Saving it in the workflow feels overkill. My first idea is to take the last one in the processing order. It won't be correct in some cases but that's not a huge problem. Or I can just take all the SaveImage nodes. Any thoughts?

I'll take this request. Up for discussion: what should happen when there are multiple SaveImage nodes in the workflow? We don't know which one saved the image. Saving it in the workflow feels overkill. My first idea is to take the last one in the processing order. It won't be correct in some cases but that's not a huge problem. Or I can just take all the SaveImage nodes. Any thoughts?

I think generally this is not doable without large change impact on other parts of the code base. To support fully load every output , we need to persist the queue history in a DB, and there will need to be a mechanism to map the generated asset to the task it was generated from.

If we are talking about just putting single asset that has workflow as metadata back to correct output node, the problem becomes simpler, which is just attach the output node id as part of the media metadata, and retrieve that accordingly.