Experimented with multi-turn-image-generation using previous response and image ID, however results are inconsistent and far inferior to the remix feature on Sora.
Curious if anyone has had much luck? Using method linked below produces poor results compared to Sora. Having issues with it losing likeness of scene objects after a few generations on the same response ID (the base image), this does not happen with remix.
Following:
https://platform.openai.com/docs/guides/image-generation#multi-turn-image-generation