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.