GPT-2 generating irrelevant images when prompted with polishing words
To be specific, the previous version of the GPT-2 refinement model had issues where the same set of prompt words would generate two completely unrelated images. This has been fixed in versions after 774. In addition, fooocus has replaced the SDXL_base model. However, this issue still exists in the current API and we hope it can be merged soon.
GPT-2 generating irrelevant images when prompted with polishing words To be specific, the previous version of the GPT-2 refinement model had issues where the same set of prompt words would generate two completely unrelated images. This has been fixed in versions after 774. In addition, fooocus has replaced the SDXL_base model. However, this issue still exists in the current API and we hope it can be merged soon.