It is early days for me in the field of generative ai. So, I am experimenting on my local machine with RTX 4080 SUPER. Most of these experiments use stock workflows suggested/ provided by the models.
I am neither an artist or student of art. It is difficult to provide references to specific styles. I am mostly driven by what looks appealing as I browse the artwork here. Though, I have tried hands are few LoRAs over past couple of days.
Today, I came across this: https://civitai.com/images/24631665
So, I wanted to try a variation of this locally... and see where I get without the LoRA mentioned here. It so happened that I had a FLUX.1-schnell workflow opened. I pasted the prompt and the image was just not what I expected. Repeated attempts gave similar (disappointing) results... So, tried on SD3.5 Medium. This time I wasn't disappointed.
Prompt
This is the prompt that I started with:
magnificent watercolor painting, high quality image of a beautiful peasant girl in india that wears a scarf around her hair, luminous eyes light her face. her mood is joyful, dynamic camera angle
Images from FLUX.1-schnell
Here are 2 images generated - that don't appear to be watercolor painting - but for (possibly) the background:
Assuming that word "magnificent" in the prompt could be misleading, I removed and tried again. But there was no difference.
Images from SD3.5 Medium
With original prompt, this is what I got:
These are definitely watercolor, but the overall mood is warm, but not necessarily cheerful.
So, changed the prompt to:
magnificent watercolor painting, high quality image of a beautiful peasant girl in india that wears a scarf around her hair, luminous eyes light her face. she is cheerful, dynamic camera angle
...and tried again - to get this:

Conclusion
Even though I have uploaded handful images here, overall analysis is based on about 100 images across both models.
FLUX.1-schnell just didn't follow 2 keywords watercolor and painting in the prompt. The overall mood of the images was definitely joyous.
As additional step, I substituted watercolor with oil. But, this didn't seem to make much difference.
SD3.5 Medium was true to spirit for for both keywords watercolor and painting. The mood in images is warm, but not necessarily joyous.
Perception of mood is quite subjective. But, the expressions here were mostly subdued. (While creating the images above for this article, I got much better results with cheerful than original attempts.)
It is also quite possible that I missed something quite obvious when working with FLUX.1-schnell.
Thoughts?
and/or
Tricks to get better paintings off the stock FLUX.1-schnell model..