I tried this prompt (and some variations) : ```` This is the end of the afternoon the day before the Halloween night. In this little city, many children wait the night to collect treats. Most of them are already ready. This city is like a typical western European little city today. Globally, this world is similar to our world but countless humanoid races (often fantasy ones) coexist with humans. The scene focus on two girls in this city, both eight years old, Jane and Judy. Jane is a centauress wearing a witch costume. She have a pumpkin-like basket in her hands. Judy is a drider wearing a pirate costume. She have a skull-like basket in the hands. They just met and are chatting at waiting the night. These two girls are the core of the image. They are entirely displayed, nothing hide any part of them and, from the point of view, they are the two closest characters. In the background, there are many other people (of various races), usually in Halloween costumes. However, this is still day time (this is the end of the afternoon). The sky is a sunny, blue and cloudless. ```` All of them failed, producing images UNRELATED to the prompt. This wasn't supposed to be hard or tricky. Apparently, this AI is outdated, trying to follow a now outdated logic. That logic mostly was "For any reference to any drawable thing, this thing must be drawn". It was tainted by now as outdated tweaks. More modern AI deeperly analyze the semantic of the prompt. An other problem is fantasy races. Beside a few exception, such as metmaids and werewolves, most of them where behind dead angles in the training algorithms, preventing them from properly model these creatures. In this prompt, I chose two races most modern AI have no problem to draw. This was an opportunity to check how it evolve since the last time. Visibly, it didn't evolve at all

Changed the prompt to improve results.