BLUR everywhere
What is the way to obtain an image without background blur?
The blur even appears on parts of the subject's own body. Could this be a mechanism from the original MOSS model—designed to emphasize text over images—that is affecting the generation process?
My library of prompts, which produces correct images in Krea2, Z-Image Turbo, Chroma, and Flux using the exact same prompts, does not exhibit this issue. I can copy&paste the same basic prompts across different models, and they don’t deliberately emphasize or blur elements in the image.
I work on CloudGPU. One thing worth ruling out before blaming the model: the text-to-image template runs your prompt through an upsampler that rewrites it into a full scene caption before sampling. Open the Preview Any node and read what the sampler actually received. If the rewritten caption adds depth of field, bokeh or a soft background, that would explain blur appearing even with prompts that work elsewhere. Writing the full caption yourself, so the upsampler has less to add, is a quick way to test it. I have not reproduced your images, so this is a check, not a diagnosis.