This topic came to me after I’d played around a bit with ChatGPT’s image-related features. I gave it a photo taken in the village where I live in Sussex, and told it to write a 100-word description of it. It produced the following.
The photo captures a quiet suburban or rural road during what appears to be early spring. The road curves gently to the right, flanked by lush greenery and a variety of flowering trees showing off shades of pink and white blossoms. There are well-manicured bushes and a collection of yellow daffodils bordering a small gravel area in the foreground. The sky is overcast, suggesting a cool, perhaps damp atmosphere. A large house with a reddish-brown roof is visible in the distance, alongside a red-brick house with a dark roof. The scene is peaceful, with no people or vehicles in sight, evoking a serene residential vibe.
Then I asked ChatGPT to generate an image based on the above description. In other words I fed its output back into its input, which it didn’t seem to mind. In fact it obliged me by producing the first of the images you see below. This image features elements from the description, yet it looks quite different from the photo. (Trust me on that. I don’t want to spoil the look of this post by uploading a poor photo.) Would the AI have produced an image closer to the original photo if I’d asked for a longer description? I doubt it. I’ve found it difficult to direct AI image generation with any precision. Sure, you can push it in the right direction, but it does tend to go its own way on the finer points. Repeating the process with other photos, I began to sense a dreamlike quality in their AI counterparts. The AI images combined features from the photos with eerie additions and modifications. In a dream you often know where you are, even though nothing looks as it should. You just know. The same applies with these photos, since I know where I took them. I use the word lucid in my post title because in a lucid dream you recognise your dream as a dream, just as I know that my images are just images.
In this first scene (described above) I’m wandering along a road about 300 metres from my house. This dream road doesn’t have the vibe of my neighbourhood, yet I understand how it came about. In the image we see a prosperous suburban or semi-rural area, with large houses and manicured verges. Many such places exist in the UK, in other parts of Sussex and in counties like Surrey and Cheshire. The dream road has no pavement, so I’d have to dream-walk in the gutter. You must dream-drive here, rather than go on foot. The trees resemble ones I’ve often seen on dinner plates.
To get to this dream pub you’d carry on down the preceding dream road and turn right after half a kilometre. It looks much like the pub of waking life, but more gemütlich. ChatGPT does a better job of recreating a photo if the photo has only one main element (as in this case). The ChatGPT photo description doesn’t provide enough information to place multiple elements correctly. In any case the AI tends to disobey, or garble, precise stage directions. This pub has a dim interior, in contrast to the overlit one of waking life with its garish sports screens. The AI has turned the St George’s flag through ninety degrees. It amuses me to think of this as a disrespectful transformation.
This ancient half-timbered house has grown bigger in my dream, with walls now straight enough to please a surveyor. It reminds me of Suzi Quatro’s story about when her father travelled from the USA to England to visit her. On arrival, she showed him around the Tudor mansion she’d bought. I don’t want to worry you, he whispered with genuine concern, but the walls are crooked. The AI has moved the lichen-covered stones from the churchyard and planted them in the greensward opposite the house.
Of all the images, this one feels closest to my waking life (despite having several components). Locals call this open area Tanyard Field. Nearby Pinchnose Green recalls the pungent smell of tanning that once pervaded the area. The houses’ sizes, colours and placement are pretty much as they should be. I’d feel reassured in my dream at this point.
I turn into the lane leading back to the High Street and see the familiar barriers and cones left by the gas-pipe men. But this dream lane looks like a residential side street in a sizeable town rather than a Sussex village. The AI has pinched the Union flag from a private social club at the far end and hung it above the road.
This dream High Street confirms the urban impression. The real village High Street often has a busy feel, but it doesn’t have these tall buildings.
I always stop and peer through the window of this antique shop. In this dream version we see four-storey buildings opposite, housing other antique shops. This could be Edinburgh. And why not, in a dream?
Ah yes, the garage at the south end of the High Street, where I sometimes buy my newspapers. In contrast to the typically English shopping thoroughfare, it has a Hopperesque quality.
Finally I’ve reached the home straight: the road where I live. Yet it resembles somewhere up north. Lancashire or Yorkshire, I’d say. My dream has retained the parked Audi, a familiar object on the road, along with the daffodils. Seeing as I’d need to walk up a non-existent alleyway on the left to get back home, I assume my house no longer exists. I panic and wake up.
Blackthorn 