How I fake a LOADED brush and real gesture without the model making a literal 'painting'
The problem with gestural abstraction via AI: prompt 'gestural brushstrokes' and the model gives you a PICTURE of brushstrokes, a rendered photo of a painting, flat, with fake canvas texture and a little frame sometimes. I want the GESTURE itself, the energy of a loaded brush dragged fast, not a photo of someone else's painting.
What's getting me closer:
Prompt the PHYSICS, not the object: 'thick wet paint dragged in a single fast arc, paint piling at the end of the stroke, bristle streaks'. Describe what the brush DID, not 'a brushstroke'.
Avoid 'painting', 'canvas', 'artwork', these pull toward 'photo of a framed painting'. I want to be INSIDE the gesture, not looking at a painting of it.
Generate huge and crop IN, so a single gesture fills the frame and there's no 'painting sitting on a wall' framing.
Still not perfect, the model has a strong prior that 'brushstroke' = 'image of a painting'. Prompting the physical ACTION instead of the noun helps most. Describe the verb, not the noun.