And now: A memetic image prompted with text - and shut down by the vision inspection mid-blur during generation, for the f-word appearing in what was to be re-created in joyous quality…So a new session was required with a substitute phrasing (for a minute of typing those words in Photoshop).
Clear error-free wording without any jumble.
Follow-up corrective generations seem to lose quality in the depictions, compared to starting a new session for one image. Perhaps the image going back in for vision understanding is a degrading experience for the AI.