I coined this phrase early in the Dolly text to image phase. I also regularly use the terms aimage and aidias (LLM idea generation).

I have used heavily used all of the public facing text to image large language models: Dall-E, Bing Image Creator, Stable Diffusion (local and online) etc. The best implementation of them all was Microserf’s dolly implementation (Bing Image Creator). This version of Dolly always worked better the UnOpen AI version. The UnOpen AI version is only getting more nerfed as time goes on.

Microserf has now attached their implementation of Dolly to a thing called Designer. Each prompt gives four aimages that you must then download. Below is an interesting bug in this website.

The first image shows the four aimages created from the prompt.

WordPress blocks still don't manage images consistently. Centering and sizing images has been a challenge since the img tag. The present monetization guided UX composed of opaque blocks makes things worse. One spends literally hours fucking with input boxes, failing and learning nothing. 

link to image

The following gallery shows the four aimages that are downloaded.

None of them match. This is much worse than just a linking error. Clearly the prompt is returning many more aimages than we are allowed to access. The prompt produced at least 8 different aimages. Four of which are shown and cannot be downloaded and 4 that are downloadable but not shown.

Where is the ghost in this machine?

Leave a Reply

Your email address will not be published. Required fields are marked *