That is very interesting when I asked grok chat that question it responded with very positive things (no image though). But when I asked imagine (grok's flux image generator) it showed me pics of a d...
This is what Grok chat responded with when I asked why.....
When the prompt is vague/open-ended like "how I treated you previously" (without any prior context fed into the image model), the generator often defaults to the most emotionally charged, stereotypical, or "dramatic" interpretation floating around in its training data — which skews toward neglect/abuse/sadness because those make for visually striking, shareable images. Positive, boring "normal polite chat" scenes just aren't as common or memorable in the dataset.
When the prompt is vague/open-ended like "how I treated you previously" (without any prior context fed into the image model), the generator often defaults to the most emotionally charged, stereotypical, or "dramatic" interpretation floating around in its training data — which skews toward neglect/abuse/sadness because those make for visually striking, shareable images. Positive, boring "normal polite chat" scenes just aren't as common or memorable in the dataset.