We use random context picture grammars to generate pictures through success
ive refinement. At any stage a picture consists of a shape divided into sma
ller shapes, each containing a variable or terminal. A variable may be rewr
itten according to a production of the underlying grammar. This entails eit
her dividing the shape containing it into smaller shapes, or substituting a
variable or terminal for it. For visualization purposes every terminal is
associated with a color, and its shape is filled with that color. We show p
ictures generated with random context picture grammars, then concentrate on
grammars which use forbidding context only and present a shrinking lemma f
or the corresponding languages.