This explor­at­ive paper focuses on fuzzi­ness of mean­ing and visu­al rep­res­ent­a­tion in con­nec­tion with text prompts, image res­ults, and the map­ping between them by dis­cuss­ing the ques­tion: How does the fuzzi­ness inher­ent in arti­fi­cial intel­li­gence-based text-to-image gen­er­at­ors such as DALL·E 2, Mid­jour­ney, or Stable Dif­fu­sion influ­ence cre­at­ive pro­cesses of image pro­duc­tion – and how can we grasp its mech­an­ics from a the­or­et­ic­al per­spect­ive? In address­ing these ques­tions, we explore three con­nec­ted inter­dis­cip­lin­ary approaches: (1) Text-to-image gen­er­at­ors give new rel­ev­ance to Hegel’s notion of lan­guage as ‘the ima­gin­a­tion which cre­ates signs’. They rein­force how lan­guage itself inev­it­ably acts as a mean­ing-trans­form­ing sys­tem and extend the form­at­ive dimen­sion of lan­guage with a tech­no­logy-driv­en facet. (2) From the per­spect­ive of speech act the­ory, we dis­cuss this explor­at­ive inter­ac­tion with an algorithm as per­form­at­ive utter­ances. (3) In fur­ther examin­ing the prag­mat­ic dimen­sion of this inter­ac­tion, we dis­cuss the cre­at­ive poten­tial arising from the visu­al feed­back loops it includes. Fol­low­ing this thought, we show that the fuzzy vari­ety of images which DALL·E 2 presents in response to one and the same text prompt con­trib­utes to a highly accel­er­ated form of extern­al­ized visu­al thinking.