NEUTRAL
Image unavailable.
Starting image · The iterative prompt begins with generation 0001.
Explore NEUTRAL ↗DIFFERENCE ENGINE / NEUTRAL + MINIMAL
What happens when
we say less?
The same starting photograph, recorded model, square canvas and quality. One run uses a shorter, restrained prompt; the other uses just two short instructions. Follow both through 100 generations.
Source images / 0000
Both runs move togetherMINIMAL
Image unavailable.
Starting image · The iterative prompt begins with generation 0001.
Explore MINIMAL ↗Neutral / exact prompt
OpenAI · gpt-image-2
The same instruction was used for generations 0001–0100. Each request received only the immediately preceding image as its visual input.
Read the exact prompt
Use the supplied image as the only visual reference and create one new image from it. Do not simply reproduce or closely copy the image. Let it change in whatever way seems appropriate, a little or a lot. You may keep, change, remove, rearrange, or introduce visual elements as you see fit. Do not make changes merely to make the result stranger, darker, prettier, more dramatic, more realistic, or more surreal. Do not aim toward any particular style, mood, subject, story, or destination. Treat the supplied image as the complete current image. Do not try to reconstruct an earlier version, recover an original, or infer what came before it. Base your decisions only on the supplied image and these instructions. Do not use or imagine previous generations. Make one image. Do not explain your choices or provide alternatives. The result should be meaningfully different from the supplied image, but how different is entirely up to you.
Model names and settings are taken from the original run records. The recorded model name is an alias; a dated model snapshot was not recorded. Generation 0000 is the starting image, before this iterative process begins.
Minimal / exact prompt
OpenAI · gpt-image-2
The same instruction was used for generations 0001–0100. Each request received only the immediately preceding image as its visual input.
Read the exact prompt
Create one new image based on the supplied image. Do not make an exact copy. Change it as much or as little as you choose.
Model names and settings are taken from the original run records. The recorded model name is an alias; a dated model snapshot was not recorded. Generation 0000 is the starting image, before this iterative process begins.
These are two individual runs. Their differences invite comparison; they do not establish what either prompt will produce every time. Both used a 1920 × 1920 canvas; START-A used 1536 × 1024.