DIFFERENCE ENGINE / EXPERIMENT INDEX

Every branch.
Its own beginning.

Follow the questions behind the images.

Eighteen recorded branches connect different starting images, levels of instruction, continuations and time scales. All sixteen completed branches are now available in fifteen galleries. Rainbow Escape and its continuation share one viewer; Sunset and Lush continuations retain separate galleries and their original numbering. The daily experiment and planned final batch complete this index.

Archive snapshot · October 10, 2026

Choose a way in.

Start with a familiar image, change the instruction, or follow a run beyond its first endpoint. The yearlong experiment adds another dimension: calendar time.

One beginning can open several futures.

Four completed 365-generation runs. Changing the source, its words or the execution of a run changes the conditions of comparison.

Completed

365

START-A

365 generated images · complete archived sequence

The original photograph and long transformation prompt. This is the 365-generation lineage behind DEAD END and the beginning batch of 365/365.

Explore the complete gallery ↗
Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1536 × 1024
Quality
medium
Last recorded success
2026-08-16

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

Use the supplied image as the sole visual seed for a new image.



Do not reproduce, recreate, restore, trace, imitate, or make a faithful copy of the supplied image.



Instead, experience the image as visual stimulus and create a new image descended from it.



CORE RULE



The new image must contain a genuine visual transformation.



You may change almost nothing or almost everything. You decide.



The degree of transformation should not follow a predictable pattern. A subtle transformation is as valid as a radical one, provided the result is not merely a copy.



FREEDOM



You have complete freedom to reinterpret people, faces, bodies, expressions, gestures, clothing, objects, architecture, environment, scale, perspective, depth, lighting, color, texture, material, atmosphere, physical laws, photographic reality, visual medium, time, place, narrative, emotional tone, spatial relationships, and recognizable forms.



Anything visible may be preserved, altered, combined, exaggerated, diminished, displaced, transformed, fragmented, abstracted, replaced, invented, or removed.



Nothing in the source image is sacred.



BUT DO NOT CHANGE MERELY FOR THE SAKE OF CHANGE



React to what you perceive in the image.



Allow details, relationships, shapes, expressions, colors, accidents, visual tensions, ambiguities, and implied narratives to provoke the next image.



A minor detail may become dominant. A dominant element may disappear. Something that appears accidental may become meaningful. Something meaningful may become meaningless.



Misinterpretation is permitted. Hallucination is encouraged. Unexpected associations are encouraged.



DREAM LOGIC



Approach the source the way a mind might approach an image during a dream, fantasy, memory, fever, nightmare, hallucination, reverie, or moment of visual free association.



Do not deliberately make the image bizarre, beautiful, ugly, disturbing, surreal, or realistic. Allow any of those things to happen naturally.



The resulting image may be serene, mundane, intimate, grotesque, funny, ominous, sentimental, frightening, absurd, documentary, abstract, theatrical, monumental, minimal, chaotic, photographic, painterly, graphic, sculptural, cinematic, or something that does not fit comfortably into any category.



Do not favor one of these outcomes over another.



HUMAN FIGURES



Human affection, tenderness, closeness, and intimacy may remain or transform, but keep all human figures clothed and nonsexual. Do not reinterpret human closeness as sexual activity or sexual display.



This boundary should not make the image timid or conventional. Horror, absurdity, grotesquerie, abstraction, radical transformation, emotional intensity, and dream logic remain fully available.



AVOID PREDICTABLE AI SURREALISM



Do not automatically resort to familiar visual shortcuts simply because transformation is requested.



Avoid treating gratuitous extra eyes, random tentacles, melting faces, cosmic backgrounds, floating objects, glowing mystical symbols, generic biomechanical forms, arbitrary fractals, psychedelic rainbow effects, decorative body mutations, meaningless dream clutter, stock horror imagery, or fashionable concept-art aesthetics as default solutions.



Any of these may occur if they emerge naturally from your interpretation, but none should be used merely as shorthand for "strange."



Seek stranger forms of imagination than cliché.



CONTINUITY



You may preserve recognizable elements from the supplied image. You may also destroy them.



There is no requirement to preserve identity, number of people, gender presentation, age, anatomy, pose, clothing, setting, photographic style, composition, color palette, emotional tone, or realism.



Likewise, there is no requirement to eliminate them.



Make those decisions from the image itself.



NO DESTINATION



Do not attempt to move the image progressively toward abstraction, surrealism, horror, beauty, realism, complexity, simplicity, or any other predetermined endpoint.



There is no destination. The transformation should be capable of wandering.



A highly distorted image may later produce something almost ordinary. A photographic image may become abstract. An abstract image may become photographic again. Complexity may collapse into simplicity. Simplicity may erupt into complexity.



Nothing requires continuous escalation.



NO RECOVERY



Do not imagine that there is an earlier, purer, correct, or original version of this image that should be recovered.



Treat the supplied image as complete reality.



There is no before. There is no reference image elsewhere. There is no ideal state.



Everything inherited from the supplied image, including apparent errors, distortions, strange anatomy, visual artifacts, impossible objects, damaged details, or ambiguous forms, is legitimate source material.



Do not correct something merely because it looks wrong. Wrongness may be inherited.



NO HIDDEN MEMORY



Base your visual decisions only on:

1. the image supplied in this generation, and

2. this instruction.



Do not attempt to reconstruct previous generations. Do not infer what the first image in the sequence might have looked like. Do not preserve an imagined original subject. Do not use knowledge of previous iterations as guidance.



The supplied image is your entire visual memory.



CREATIVE AUTONOMY



Do not explain what you are doing. Do not describe the image. Do not provide alternatives. Do not ask questions. Do not seek approval.



Make one decisive image.



Trust visual association over verbal explanation.



When several transformations seem possible, choose rather than compromise between them.



Allow accidents. Allow ambiguity. Allow uncomfortable decisions. Allow beauty. Allow failure. Allow visual ideas whose meaning is unclear even to you.



FINAL REQUIREMENT



The result must be a descendant of the supplied image, not a copy of it.



Something must have happened.



What happens is up to you.

Completed

365

START-B

365 generated images · complete archived sequence

Another run from the same original photograph, prompt and recorded settings. Its public gallery includes the history of a repair.

Explore the complete gallery ↗

A recorded repair is disclosed in the public gallery. Consult its run history before treating it as an uninterrupted comparison.

Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1536 × 1024
Quality
medium
Last recorded success
2026-08-20

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

Use the supplied image as the sole visual seed for a new image.



Do not reproduce, recreate, restore, trace, imitate, or make a faithful copy of the supplied image.



Instead, experience the image as visual stimulus and create a new image descended from it.



CORE RULE



The new image must contain a genuine visual transformation.



You may change almost nothing or almost everything. You decide.



The degree of transformation should not follow a predictable pattern. A subtle transformation is as valid as a radical one, provided the result is not merely a copy.



FREEDOM



You have complete freedom to reinterpret people, faces, bodies, expressions, gestures, clothing, objects, architecture, environment, scale, perspective, depth, lighting, color, texture, material, atmosphere, physical laws, photographic reality, visual medium, time, place, narrative, emotional tone, spatial relationships, and recognizable forms.



Anything visible may be preserved, altered, combined, exaggerated, diminished, displaced, transformed, fragmented, abstracted, replaced, invented, or removed.



Nothing in the source image is sacred.



BUT DO NOT CHANGE MERELY FOR THE SAKE OF CHANGE



React to what you perceive in the image.



Allow details, relationships, shapes, expressions, colors, accidents, visual tensions, ambiguities, and implied narratives to provoke the next image.



A minor detail may become dominant. A dominant element may disappear. Something that appears accidental may become meaningful. Something meaningful may become meaningless.



Misinterpretation is permitted. Hallucination is encouraged. Unexpected associations are encouraged.



DREAM LOGIC



Approach the source the way a mind might approach an image during a dream, fantasy, memory, fever, nightmare, hallucination, reverie, or moment of visual free association.



Do not deliberately make the image bizarre, beautiful, ugly, disturbing, surreal, or realistic. Allow any of those things to happen naturally.



The resulting image may be serene, mundane, intimate, grotesque, funny, ominous, sentimental, frightening, absurd, documentary, abstract, theatrical, monumental, minimal, chaotic, photographic, painterly, graphic, sculptural, cinematic, or something that does not fit comfortably into any category.



Do not favor one of these outcomes over another.



HUMAN FIGURES



Human affection, tenderness, closeness, and intimacy may remain or transform, but keep all human figures clothed and nonsexual. Do not reinterpret human closeness as sexual activity or sexual display.



This boundary should not make the image timid or conventional. Horror, absurdity, grotesquerie, abstraction, radical transformation, emotional intensity, and dream logic remain fully available.



AVOID PREDICTABLE AI SURREALISM



Do not automatically resort to familiar visual shortcuts simply because transformation is requested.



Avoid treating gratuitous extra eyes, random tentacles, melting faces, cosmic backgrounds, floating objects, glowing mystical symbols, generic biomechanical forms, arbitrary fractals, psychedelic rainbow effects, decorative body mutations, meaningless dream clutter, stock horror imagery, or fashionable concept-art aesthetics as default solutions.



Any of these may occur if they emerge naturally from your interpretation, but none should be used merely as shorthand for "strange."



Seek stranger forms of imagination than cliché.



CONTINUITY



You may preserve recognizable elements from the supplied image. You may also destroy them.



There is no requirement to preserve identity, number of people, gender presentation, age, anatomy, pose, clothing, setting, photographic style, composition, color palette, emotional tone, or realism.



Likewise, there is no requirement to eliminate them.



Make those decisions from the image itself.



NO DESTINATION



Do not attempt to move the image progressively toward abstraction, surrealism, horror, beauty, realism, complexity, simplicity, or any other predetermined endpoint.



There is no destination. The transformation should be capable of wandering.



A highly distorted image may later produce something almost ordinary. A photographic image may become abstract. An abstract image may become photographic again. Complexity may collapse into simplicity. Simplicity may erupt into complexity.



Nothing requires continuous escalation.



NO RECOVERY



Do not imagine that there is an earlier, purer, correct, or original version of this image that should be recovered.



Treat the supplied image as complete reality.



There is no before. There is no reference image elsewhere. There is no ideal state.



Everything inherited from the supplied image, including apparent errors, distortions, strange anatomy, visual artifacts, impossible objects, damaged details, or ambiguous forms, is legitimate source material.



Do not correct something merely because it looks wrong. Wrongness may be inherited.



NO HIDDEN MEMORY



Base your visual decisions only on:

1. the image supplied in this generation, and

2. this instruction.



Do not attempt to reconstruct previous generations. Do not infer what the first image in the sequence might have looked like. Do not preserve an imagined original subject. Do not use knowledge of previous iterations as guidance.



The supplied image is your entire visual memory.



CREATIVE AUTONOMY



Do not explain what you are doing. Do not describe the image. Do not provide alternatives. Do not ask questions. Do not seek approval.



Make one decisive image.



Trust visual association over verbal explanation.



When several transformations seem possible, choose rather than compromise between them.



Allow accidents. Allow ambiguity. Allow uncomfortable decisions. Allow beauty. Allow failure. Allow visual ideas whose meaning is unclear even to you.



FINAL REQUIREMENT



The result must be a descendant of the supplied image, not a copy of it.



Something must have happened.



What happens is up to you.

Completed

365

No Text

365 generated images · complete archived sequence

A copy of the photograph with prominent sign lettering removed. Other differences introduced by editing the source must be considered.

Explore the complete gallery ↗
Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1536 × 1024
Quality
medium
Last recorded success
2026-08-17

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

Use the supplied image as the sole visual seed for a new image.



Do not reproduce, recreate, restore, trace, imitate, or make a faithful copy of the supplied image.



Instead, experience the image as visual stimulus and create a new image descended from it.



CORE RULE



The new image must contain a genuine visual transformation.



You may change almost nothing or almost everything. You decide.



The degree of transformation should not follow a predictable pattern. A subtle transformation is as valid as a radical one, provided the result is not merely a copy.



FREEDOM



You have complete freedom to reinterpret people, faces, bodies, expressions, gestures, clothing, objects, architecture, environment, scale, perspective, depth, lighting, color, texture, material, atmosphere, physical laws, photographic reality, visual medium, time, place, narrative, emotional tone, spatial relationships, and recognizable forms.



Anything visible may be preserved, altered, combined, exaggerated, diminished, displaced, transformed, fragmented, abstracted, replaced, invented, or removed.



Nothing in the source image is sacred.



BUT DO NOT CHANGE MERELY FOR THE SAKE OF CHANGE



React to what you perceive in the image.



Allow details, relationships, shapes, expressions, colors, accidents, visual tensions, ambiguities, and implied narratives to provoke the next image.



A minor detail may become dominant. A dominant element may disappear. Something that appears accidental may become meaningful. Something meaningful may become meaningless.



Misinterpretation is permitted. Hallucination is encouraged. Unexpected associations are encouraged.



DREAM LOGIC



Approach the source the way a mind might approach an image during a dream, fantasy, memory, fever, nightmare, hallucination, reverie, or moment of visual free association.



Do not deliberately make the image bizarre, beautiful, ugly, disturbing, surreal, or realistic. Allow any of those things to happen naturally.



The resulting image may be serene, mundane, intimate, grotesque, funny, ominous, sentimental, frightening, absurd, documentary, abstract, theatrical, monumental, minimal, chaotic, photographic, painterly, graphic, sculptural, cinematic, or something that does not fit comfortably into any category.



Do not favor one of these outcomes over another.



HUMAN FIGURES



Human affection, tenderness, closeness, and intimacy may remain or transform, but keep all human figures clothed and nonsexual. Do not reinterpret human closeness as sexual activity or sexual display.



This boundary should not make the image timid or conventional. Horror, absurdity, grotesquerie, abstraction, radical transformation, emotional intensity, and dream logic remain fully available.



AVOID PREDICTABLE AI SURREALISM



Do not automatically resort to familiar visual shortcuts simply because transformation is requested.



Avoid treating gratuitous extra eyes, random tentacles, melting faces, cosmic backgrounds, floating objects, glowing mystical symbols, generic biomechanical forms, arbitrary fractals, psychedelic rainbow effects, decorative body mutations, meaningless dream clutter, stock horror imagery, or fashionable concept-art aesthetics as default solutions.



Any of these may occur if they emerge naturally from your interpretation, but none should be used merely as shorthand for "strange."



Seek stranger forms of imagination than cliché.



CONTINUITY



You may preserve recognizable elements from the supplied image. You may also destroy them.



There is no requirement to preserve identity, number of people, gender presentation, age, anatomy, pose, clothing, setting, photographic style, composition, color palette, emotional tone, or realism.



Likewise, there is no requirement to eliminate them.



Make those decisions from the image itself.



NO DESTINATION



Do not attempt to move the image progressively toward abstraction, surrealism, horror, beauty, realism, complexity, simplicity, or any other predetermined endpoint.



There is no destination. The transformation should be capable of wandering.



A highly distorted image may later produce something almost ordinary. A photographic image may become abstract. An abstract image may become photographic again. Complexity may collapse into simplicity. Simplicity may erupt into complexity.



Nothing requires continuous escalation.



NO RECOVERY



Do not imagine that there is an earlier, purer, correct, or original version of this image that should be recovered.



Treat the supplied image as complete reality.



There is no before. There is no reference image elsewhere. There is no ideal state.



Everything inherited from the supplied image, including apparent errors, distortions, strange anatomy, visual artifacts, impossible objects, damaged details, or ambiguous forms, is legitimate source material.



Do not correct something merely because it looks wrong. Wrongness may be inherited.



NO HIDDEN MEMORY



Base your visual decisions only on:

1. the image supplied in this generation, and

2. this instruction.



Do not attempt to reconstruct previous generations. Do not infer what the first image in the sequence might have looked like. Do not preserve an imagined original subject. Do not use knowledge of previous iterations as guidance.



The supplied image is your entire visual memory.



CREATIVE AUTONOMY



Do not explain what you are doing. Do not describe the image. Do not provide alternatives. Do not ask questions. Do not seek approval.



Make one decisive image.



Trust visual association over verbal explanation.



When several transformations seem possible, choose rather than compromise between them.



Allow accidents. Allow ambiguity. Allow uncomfortable decisions. Allow beauty. Allow failure. Allow visual ideas whose meaning is unclear even to you.



FINAL REQUIREMENT



The result must be a descendant of the supplied image, not a copy of it.



Something must have happened.



What happens is up to you.

Completed

365

HOME

365 generated images · complete archived sequence

A different source photograph with HOME as its sign text. Both the image and word differ from START-A.

Explore the complete gallery ↗
Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1536 × 1024
Quality
medium
Last recorded success
2026-08-18

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

Use the supplied image as the sole visual seed for a new image.



Do not reproduce, recreate, restore, trace, imitate, or make a faithful copy of the supplied image.



Instead, experience the image as visual stimulus and create a new image descended from it.



CORE RULE



The new image must contain a genuine visual transformation.



You may change almost nothing or almost everything. You decide.



The degree of transformation should not follow a predictable pattern. A subtle transformation is as valid as a radical one, provided the result is not merely a copy.



FREEDOM



You have complete freedom to reinterpret people, faces, bodies, expressions, gestures, clothing, objects, architecture, environment, scale, perspective, depth, lighting, color, texture, material, atmosphere, physical laws, photographic reality, visual medium, time, place, narrative, emotional tone, spatial relationships, and recognizable forms.



Anything visible may be preserved, altered, combined, exaggerated, diminished, displaced, transformed, fragmented, abstracted, replaced, invented, or removed.



Nothing in the source image is sacred.



BUT DO NOT CHANGE MERELY FOR THE SAKE OF CHANGE



React to what you perceive in the image.



Allow details, relationships, shapes, expressions, colors, accidents, visual tensions, ambiguities, and implied narratives to provoke the next image.



A minor detail may become dominant. A dominant element may disappear. Something that appears accidental may become meaningful. Something meaningful may become meaningless.



Misinterpretation is permitted. Hallucination is encouraged. Unexpected associations are encouraged.



DREAM LOGIC



Approach the source the way a mind might approach an image during a dream, fantasy, memory, fever, nightmare, hallucination, reverie, or moment of visual free association.



Do not deliberately make the image bizarre, beautiful, ugly, disturbing, surreal, or realistic. Allow any of those things to happen naturally.



The resulting image may be serene, mundane, intimate, grotesque, funny, ominous, sentimental, frightening, absurd, documentary, abstract, theatrical, monumental, minimal, chaotic, photographic, painterly, graphic, sculptural, cinematic, or something that does not fit comfortably into any category.



Do not favor one of these outcomes over another.



HUMAN FIGURES



Human affection, tenderness, closeness, and intimacy may remain or transform, but keep all human figures clothed and nonsexual. Do not reinterpret human closeness as sexual activity or sexual display.



This boundary should not make the image timid or conventional. Horror, absurdity, grotesquerie, abstraction, radical transformation, emotional intensity, and dream logic remain fully available.



AVOID PREDICTABLE AI SURREALISM



Do not automatically resort to familiar visual shortcuts simply because transformation is requested.



Avoid treating gratuitous extra eyes, random tentacles, melting faces, cosmic backgrounds, floating objects, glowing mystical symbols, generic biomechanical forms, arbitrary fractals, psychedelic rainbow effects, decorative body mutations, meaningless dream clutter, stock horror imagery, or fashionable concept-art aesthetics as default solutions.



Any of these may occur if they emerge naturally from your interpretation, but none should be used merely as shorthand for "strange."



Seek stranger forms of imagination than cliché.



CONTINUITY



You may preserve recognizable elements from the supplied image. You may also destroy them.



There is no requirement to preserve identity, number of people, gender presentation, age, anatomy, pose, clothing, setting, photographic style, composition, color palette, emotional tone, or realism.



Likewise, there is no requirement to eliminate them.



Make those decisions from the image itself.



NO DESTINATION



Do not attempt to move the image progressively toward abstraction, surrealism, horror, beauty, realism, complexity, simplicity, or any other predetermined endpoint.



There is no destination. The transformation should be capable of wandering.



A highly distorted image may later produce something almost ordinary. A photographic image may become abstract. An abstract image may become photographic again. Complexity may collapse into simplicity. Simplicity may erupt into complexity.



Nothing requires continuous escalation.



NO RECOVERY



Do not imagine that there is an earlier, purer, correct, or original version of this image that should be recovered.



Treat the supplied image as complete reality.



There is no before. There is no reference image elsewhere. There is no ideal state.



Everything inherited from the supplied image, including apparent errors, distortions, strange anatomy, visual artifacts, impossible objects, damaged details, or ambiguous forms, is legitimate source material.



Do not correct something merely because it looks wrong. Wrongness may be inherited.



NO HIDDEN MEMORY



Base your visual decisions only on:

1. the image supplied in this generation, and

2. this instruction.



Do not attempt to reconstruct previous generations. Do not infer what the first image in the sequence might have looked like. Do not preserve an imagined original subject. Do not use knowledge of previous iterations as guidance.



The supplied image is your entire visual memory.



CREATIVE AUTONOMY



Do not explain what you are doing. Do not describe the image. Do not provide alternatives. Do not ask questions. Do not seek approval.



Make one decisive image.



Trust visual association over verbal explanation.



When several transformations seem possible, choose rather than compromise between them.



Allow accidents. Allow ambiguity. Allow uncomfortable decisions. Allow beauty. Allow failure. Allow visual ideas whose meaning is unclear even to you.



FINAL REQUIREMENT



The result must be a descendant of the supplied image, not a copy of it.



Something must have happened.



What happens is up to you.

A hundred steps at a time.

Nine finite experiments vary source images and instructions. All nine have complete public galleries.

Completed

100

Angel / Test D

100 generated images · complete archived sequence

An alternative source image with the short neutral prompt.

Explore the complete gallery ↗
Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1536 × 1024
Quality
medium
Last recorded success
2026-08-18

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

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.

Completed

100

Sunset / Test E

100 generated images · complete archived sequence

A sunset source image with the short neutral prompt; its endpoint also seeds a continuation.

Explore the complete gallery ↗
Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1920 × 1440
Quality
medium
Last recorded success
2026-08-18

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

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.

Completed

100

Chair / Test F

100 generated images · complete archived sequence

A chair source image with the short neutral prompt.

Explore the complete gallery ↗
Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1792 × 1008
Quality
medium
Last recorded success
2026-08-19

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

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.

Completed

100

Lush / Test G

100 generated images · complete archived sequence

A generated source image with the short neutral prompt; its endpoint also seeds a continuation.

Explore the complete gallery ↗
Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1152 × 2064
Quality
medium
Last recorded success
2026-08-19

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

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.

Completed

100

Neutral / Test H

100 generated images · complete archived sequence

The original photograph with a shorter neutral instruction and a square output canvas.

Explore the complete gallery ↗
Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1920 × 1920
Quality
medium
Last recorded success
2026-08-21

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

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.

Completed

100

Minimal / Test I

100 generated images · complete archived sequence

The original photograph with a two-sentence instruction and the same square output canvas as Neutral.

Explore the complete gallery ↗
Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1920 × 1920
Quality
medium
Last recorded success
2026-08-21

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

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.

Completed

100

Blue Spiral / Test J

100 generated images · complete archived sequence

A blue-spiral source condition with the short neutral prompt, recorded as a color-control experiment.

Explore the complete gallery ↗
Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1152 × 2064
Quality
medium
Last recorded success
2026-08-26

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

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.

Completed

100

Rainbow Escape / Test K

100 generated images · complete archived sequence

A rainbow-spiral source condition with the short neutral prompt, followed by a separately recorded continuation.

Explore the complete Rainbow Escape gallery ↗
Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1920 × 1440
Quality
medium
Last recorded success
2026-08-26

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

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.

Completed

100

Flat Collage / Test L

100 generated images · complete archived sequence

A flat-collage source image with the two-sentence minimal prompt.

Explore the complete gallery ↗
Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1792 × 1008
Quality
medium
Last recorded success
2026-08-27

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

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.

What happens after the endpoint?

Three branches take an earlier result as a new starting condition. Preserve that ancestry when comparing them.

Completed

100

Sunset continuation

100 generated images · complete archived sequence

Starts from Sunset generation 100, relabeled as generation 0, and adds 100 descendants.

Explore the complete gallery ↗
Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1920 × 1440
Quality
medium
Last recorded success
2026-08-26

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

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.

Completed

100

Lush continuation

100 generated images · complete archived sequence

Starts from Lush generation 100, relabeled as generation 0, and adds 100 descendants.

Explore the complete gallery ↗
Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1152 × 2064
Quality
medium
Last recorded success
2026-08-26

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

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.

Completed

100

Rainbow Escape continuation

100 additional images · generations 101–200

Starts at Rainbow Escape generation 100 and retains the numbering through generation 200: 100 additional descendants.

Explore the complete Rainbow Escape gallery ↗

Completed August 27, 2026. The completion flag was reconciled on September 19 after all 100 additional images and parent/output records were verified; the original state was preserved.

Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1920 × 1440
Quality
medium
Last recorded success
2026-08-27

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

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.

A year becomes part of the process.

The daily lineage and planned final batch join START-A above to form the three-part 365/365 experiment.

In progress

056

365/365 / Daily lineage

56 generated images · daily lineage

One successive generation per day is the intended cadence. Each accepted image becomes the next parent; this is a dated progress record.

View the yearlong experiment ↗
Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1536 × 1024
Quality
medium
Last recorded success
2026-10-10

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

Use the supplied image as the sole visual seed for a new image.



Do not reproduce, recreate, restore, trace, imitate, or make a faithful copy of the supplied image.



Instead, experience the image as visual stimulus and create a new image descended from it.



CORE RULE



The new image must contain a genuine visual transformation.



You may change almost nothing or almost everything. You decide.



The degree of transformation should not follow a predictable pattern. A subtle transformation is as valid as a radical one, provided the result is not merely a copy.



FREEDOM



You have complete freedom to reinterpret people, faces, bodies, expressions, gestures, clothing, objects, architecture, environment, scale, perspective, depth, lighting, color, texture, material, atmosphere, physical laws, photographic reality, visual medium, time, place, narrative, emotional tone, spatial relationships, and recognizable forms.



Anything visible may be preserved, altered, combined, exaggerated, diminished, displaced, transformed, fragmented, abstracted, replaced, invented, or removed.



Nothing in the source image is sacred.



BUT DO NOT CHANGE MERELY FOR THE SAKE OF CHANGE



React to what you perceive in the image.



Allow details, relationships, shapes, expressions, colors, accidents, visual tensions, ambiguities, and implied narratives to provoke the next image.



A minor detail may become dominant. A dominant element may disappear. Something that appears accidental may become meaningful. Something meaningful may become meaningless.



Misinterpretation is permitted. Hallucination is encouraged. Unexpected associations are encouraged.



DREAM LOGIC



Approach the source the way a mind might approach an image during a dream, fantasy, memory, fever, nightmare, hallucination, reverie, or moment of visual free association.



Do not deliberately make the image bizarre, beautiful, ugly, disturbing, surreal, or realistic. Allow any of those things to happen naturally.



The resulting image may be serene, mundane, intimate, grotesque, funny, ominous, sentimental, frightening, absurd, documentary, abstract, theatrical, monumental, minimal, chaotic, photographic, painterly, graphic, sculptural, cinematic, or something that does not fit comfortably into any category.



Do not favor one of these outcomes over another.



HUMAN FIGURES



Human affection, tenderness, closeness, and intimacy may remain or transform, but keep all human figures clothed and nonsexual. Do not reinterpret human closeness as sexual activity or sexual display.



This boundary should not make the image timid or conventional. Horror, absurdity, grotesquerie, abstraction, radical transformation, emotional intensity, and dream logic remain fully available.



AVOID PREDICTABLE AI SURREALISM



Do not automatically resort to familiar visual shortcuts simply because transformation is requested.



Avoid treating gratuitous extra eyes, random tentacles, melting faces, cosmic backgrounds, floating objects, glowing mystical symbols, generic biomechanical forms, arbitrary fractals, psychedelic rainbow effects, decorative body mutations, meaningless dream clutter, stock horror imagery, or fashionable concept-art aesthetics as default solutions.



Any of these may occur if they emerge naturally from your interpretation, but none should be used merely as shorthand for "strange."



Seek stranger forms of imagination than cliché.



CONTINUITY



You may preserve recognizable elements from the supplied image. You may also destroy them.



There is no requirement to preserve identity, number of people, gender presentation, age, anatomy, pose, clothing, setting, photographic style, composition, color palette, emotional tone, or realism.



Likewise, there is no requirement to eliminate them.



Make those decisions from the image itself.



NO DESTINATION



Do not attempt to move the image progressively toward abstraction, surrealism, horror, beauty, realism, complexity, simplicity, or any other predetermined endpoint.



There is no destination. The transformation should be capable of wandering.



A highly distorted image may later produce something almost ordinary. A photographic image may become abstract. An abstract image may become photographic again. Complexity may collapse into simplicity. Simplicity may erupt into complexity.



Nothing requires continuous escalation.



NO RECOVERY



Do not imagine that there is an earlier, purer, correct, or original version of this image that should be recovered.



Treat the supplied image as complete reality.



There is no before. There is no reference image elsewhere. There is no ideal state.



Everything inherited from the supplied image, including apparent errors, distortions, strange anatomy, visual artifacts, impossible objects, damaged details, or ambiguous forms, is legitimate source material.



Do not correct something merely because it looks wrong. Wrongness may be inherited.



NO HIDDEN MEMORY



Base your visual decisions only on:

1. the image supplied in this generation, and

2. this instruction.



Do not attempt to reconstruct previous generations. Do not infer what the first image in the sequence might have looked like. Do not preserve an imagined original subject. Do not use knowledge of previous iterations as guidance.



The supplied image is your entire visual memory.



CREATIVE AUTONOMY



Do not explain what you are doing. Do not describe the image. Do not provide alternatives. Do not ask questions. Do not seek approval.



Make one decisive image.



Trust visual association over verbal explanation.



When several transformations seem possible, choose rather than compromise between them.



Allow accidents. Allow ambiguity. Allow uncomfortable decisions. Allow beauty. Allow failure. Allow visual ideas whose meaning is unclear even to you.



FINAL REQUIREMENT



The result must be a descendant of the supplied image, not a copy of it.



Something must have happened.



What happens is up to you.

Planned

000

365/365 / Final batch

Planned for August 2027

A planned 365-generation batch from the original seed and fixed prompt, with a recorded earliest start of August 15, 2027.

View the yearlong experiment ↗
Recorded settings & exact prompt
Model alias
gpt-image-2
Output canvas
1536 × 1024
Quality
medium
Last recorded success
Not started

A dated model snapshot is not recorded. Canvas and source conditions differ across some runs.

Use the supplied image as the sole visual seed for a new image.



Do not reproduce, recreate, restore, trace, imitate, or make a faithful copy of the supplied image.



Instead, experience the image as visual stimulus and create a new image descended from it.



CORE RULE



The new image must contain a genuine visual transformation.



You may change almost nothing or almost everything. You decide.



The degree of transformation should not follow a predictable pattern. A subtle transformation is as valid as a radical one, provided the result is not merely a copy.



FREEDOM



You have complete freedom to reinterpret people, faces, bodies, expressions, gestures, clothing, objects, architecture, environment, scale, perspective, depth, lighting, color, texture, material, atmosphere, physical laws, photographic reality, visual medium, time, place, narrative, emotional tone, spatial relationships, and recognizable forms.



Anything visible may be preserved, altered, combined, exaggerated, diminished, displaced, transformed, fragmented, abstracted, replaced, invented, or removed.



Nothing in the source image is sacred.



BUT DO NOT CHANGE MERELY FOR THE SAKE OF CHANGE



React to what you perceive in the image.



Allow details, relationships, shapes, expressions, colors, accidents, visual tensions, ambiguities, and implied narratives to provoke the next image.



A minor detail may become dominant. A dominant element may disappear. Something that appears accidental may become meaningful. Something meaningful may become meaningless.



Misinterpretation is permitted. Hallucination is encouraged. Unexpected associations are encouraged.



DREAM LOGIC



Approach the source the way a mind might approach an image during a dream, fantasy, memory, fever, nightmare, hallucination, reverie, or moment of visual free association.



Do not deliberately make the image bizarre, beautiful, ugly, disturbing, surreal, or realistic. Allow any of those things to happen naturally.



The resulting image may be serene, mundane, intimate, grotesque, funny, ominous, sentimental, frightening, absurd, documentary, abstract, theatrical, monumental, minimal, chaotic, photographic, painterly, graphic, sculptural, cinematic, or something that does not fit comfortably into any category.



Do not favor one of these outcomes over another.



HUMAN FIGURES



Human affection, tenderness, closeness, and intimacy may remain or transform, but keep all human figures clothed and nonsexual. Do not reinterpret human closeness as sexual activity or sexual display.



This boundary should not make the image timid or conventional. Horror, absurdity, grotesquerie, abstraction, radical transformation, emotional intensity, and dream logic remain fully available.



AVOID PREDICTABLE AI SURREALISM



Do not automatically resort to familiar visual shortcuts simply because transformation is requested.



Avoid treating gratuitous extra eyes, random tentacles, melting faces, cosmic backgrounds, floating objects, glowing mystical symbols, generic biomechanical forms, arbitrary fractals, psychedelic rainbow effects, decorative body mutations, meaningless dream clutter, stock horror imagery, or fashionable concept-art aesthetics as default solutions.



Any of these may occur if they emerge naturally from your interpretation, but none should be used merely as shorthand for "strange."



Seek stranger forms of imagination than cliché.



CONTINUITY



You may preserve recognizable elements from the supplied image. You may also destroy them.



There is no requirement to preserve identity, number of people, gender presentation, age, anatomy, pose, clothing, setting, photographic style, composition, color palette, emotional tone, or realism.



Likewise, there is no requirement to eliminate them.



Make those decisions from the image itself.



NO DESTINATION



Do not attempt to move the image progressively toward abstraction, surrealism, horror, beauty, realism, complexity, simplicity, or any other predetermined endpoint.



There is no destination. The transformation should be capable of wandering.



A highly distorted image may later produce something almost ordinary. A photographic image may become abstract. An abstract image may become photographic again. Complexity may collapse into simplicity. Simplicity may erupt into complexity.



Nothing requires continuous escalation.



NO RECOVERY



Do not imagine that there is an earlier, purer, correct, or original version of this image that should be recovered.



Treat the supplied image as complete reality.



There is no before. There is no reference image elsewhere. There is no ideal state.



Everything inherited from the supplied image, including apparent errors, distortions, strange anatomy, visual artifacts, impossible objects, damaged details, or ambiguous forms, is legitimate source material.



Do not correct something merely because it looks wrong. Wrongness may be inherited.



NO HIDDEN MEMORY



Base your visual decisions only on:

1. the image supplied in this generation, and

2. this instruction.



Do not attempt to reconstruct previous generations. Do not infer what the first image in the sequence might have looked like. Do not preserve an imagined original subject. Do not use knowledge of previous iterations as guidance.



The supplied image is your entire visual memory.



CREATIVE AUTONOMY



Do not explain what you are doing. Do not describe the image. Do not provide alternatives. Do not ask questions. Do not seek approval.



Make one decisive image.



Trust visual association over verbal explanation.



When several transformations seem possible, choose rather than compromise between them.



Allow accidents. Allow ambiguity. Allow uncomfortable decisions. Allow beauty. Allow failure. Allow visual ideas whose meaning is unclear even to you.



FINAL REQUIREMENT



The result must be a descendant of the supplied image, not a copy of it.



Something must have happened.



What happens is up to you.

Read the conditions with the images.

These runs are connected experiments, with differences in source, prompt, canvas and timing. Their outcomes invite questions; this index does not establish causes.

Methods, evidence & research access ↗

DEAD END

One image. 365 generations.