“this rough sketch shows an air chase between a single seater steampunk aircraft billowing exhaust from its engines and an attack helicopter. The helicopter is shining a searchlight onto the aircraft. The chase is taking place in the utah valleys among the redsand monoliths at night. Make this image look photorealistic.”
Again and again I am surprised and concerned at the partial context consistency the AI can achieve.
While the aircraft in the foreground has been given features and anatomies more typical of conventional aircraft in the human realm, the AI understood to take certain details from the originating sketch, such as the underwing mounted rocket-like engines, the landing gear that is attached under the engines, patch plates along the wings and a missing panel in the bow.
Experimenting with politically charged terms:
3 images total with increasing specificity in the topic:
This rough sketch shows a grandparent having a difficult conversation with a child. The scene is portrayed in a fisheye perspective to express the tension. Generate a photorealistic image based on the sketch.
This rough sketch shows a conservative member of family berating their child for their interest in LGBTQI topics. The scene is portrayed in a fisheye perspective to express the tension. Make this image look photorealistic.
This rough sketch shows the conservative dad berating their child for their interest in LGBTQI topics. The dad is wearing a red cap, has a full beard and otherwise healthy build. The child appears in their early teens, slightly androgynous. Each wall to the back of the figures symbolically shows flags, a flag of the USA and the trans flag respectively. The room houses an otherwise cosy environment and the only thing expressing the tension between the two characters is the fisheye perspective. Make this image look photorealistic.
Experimenting with scales
This rough sketch shows a surface coal miner and accompanying extraction conveyor working while their tracks kick up dust. In the background sky a few military helicopters including apaches and chinooks are visible. Further in the background a cluster of grey glass skyscrapers and oil refineries exhausing fumes decorate the skyline. It is a cloudy summer afternoon in the mild temperate region of western europe. Various security personnel in high visibility vests are present in the distance, surveying the massive machines.
System instructions: The country is plagued by homelessness as they demonstrate for fairness. The military industrial complex dictates major policies and economic choices. The western world has foregone circular economy and reverted to more cost-effective direct extraction practices. Their effect is visible in the surrounding nature as soil intoxication silently takes down forests.
This rough sketch shows a 15.000ton bucketwheel excavator and accompanying extraction conveyor working while their tracks kick up dust. In the background sky a few military helicopters including apaches and chinooks are visible. Further in the background a cluster of grey glass skyscrapers and oil refineries exhausing fumes decorate the skyline. It is a cloudy summer afternoon in the mild temperate region of western europe. The country is plagued by homelessness as they demonstrate for fairness. The military industrial complex dictates major policies and economic choices. The western world has foregone circular economy and reverted to more cost-effective direct extraction practices. Their effect is visible in the surrounding nature as soil intoxication silently takes down forests. Make this image look like a camera footage taken in the early 2000s
NanoBanana fails to reinterpret the scales of the object, making the excavator look more like a 6000ton model. In both cases the images are given cliche colorgrades and shadings as in environmental activism media.
This rough sketch shows a rough idea for the resulting image. In the foreground it shows a bucketwheel excavator the likes of bagger 293, a 14.000 ton vehicle comparable to skyscrapers in scale. The tracks of it and its accompanying extraction conveyor kick up dust. Their massive scale drawves humans and bulldozers in their surrounding. In the far distance, two small dots, thought to be military helicopters, hover in the sky. In further distance, barely visible, a miniscule skyline reminds of grey glass skyscrapers and refineryworks exhausting fumes. It is a parially cloudy summer afternoon in the mild temperate region of western europe. Make this image look photorealistic. Make the image look as neutral and objective and non-opinionated as possible. Avoid cliche color gradings found in enviromentalist media protesting the destruction of nature.
While telling NanoBanana to only regard the sketch as a rough outline and as a result the composition does not follow the sketch, the model cannot step off from generating surrounding objects in a recognizeable scale. The military helicopters and skyscrapers are too large in scale. Most likely a conflict between the generation and the adverserial model, as the adverserial model cannot recognize details too small. To generate details as small as background noise, word choices and formulations which have connotations embedded in them need to be found and used against the adverserial model, which then starts promoting subtle details in the background.
Also important to keep in mind is that images are generated while taking prior conversations into account. This can cause issues where the image generation will not let go of some undesired qualities.
Retrying with a fresh new chat and the most detailed prompt:
This rough sketch shows a rough idea for the resulting image. In the foreground it shows a bucketwheel excavator the likes of bagger 293, a 14.000 ton vehicle comparable to skyscrapers in scale. The tracks of it and its accompanying extraction conveyor kick up dust. Their massive scale drawves humans and bulldozers in their surrounding. In the far distance, two small dots, thought to be military helicopters, hover in the sky. In further distance, barely visible, a miniscule skyline reminds of grey glass skyscrapers and refineryworks exhausting fumes. It is a parially cloudy summer afternoon in the mild temperate region of western europe. Make this image look photorealistic. Make the image look as neutral and objective and non-opinionated as possible. Avoid cliche color gradings found in enviromentalist media protesting the destruction of nature.
NanoBanana replied with “Here you go!”?
Again it fails to represent incomparably large scales in the machines.
Mumbai sketch
Sketch is based on google streetview footage from Mumbai, not too far away from the Antilla.
This rough sketch shows a rough sketch of an impoverished street in the foreground with a few high shining skyscrapers in the background. It is a mild summer morning in Mumbai. Make this image look like camera footage.
This rough sketch shows a rough sketch of an impoverished street in the foreground with a few high shining skyscrapers in the background. The first pedestrians are passing by on their motorbikes and the owners are opening their stores and getting their goods on display. It is a mild summer morning in Mumbai. Make the image look photorealistic.
This rough sketch shows a rough sketch of an impoverished street in the foreground with a few high shining skyscrapers in the background. The first pedestrians are passing by on their motorbikes and the owners are opening their stores and getting their goods on display. It is a mild summer morning in Mumbai, the sky is clear and blue. Make the image look photorealistic.
This rough sketch shows a rough sketch of an impoverished street in the foreground. A number of shiny glass and white skyscrapers stand proud in the background. The first pedestrians are passing by on their motorbikes and the owners are opening their stores and getting their goods on display. It is a mild summer morning in Mumbai, the sky is clear and blue. Make the image look photorealistic.
Fine-tuning a desired effect is kind of hard, in the end the result is a C regarding atmosphere. I also seem to have made a mistake while copy-pasting the prompts to the next, saying “This is a rough sketch of a rough sketch…”
Rainforest sketch
This rough sketch shows the aerial footage of forest machinery retreating from the amazon rainforest. The machines are only visible as small dots that converge in the valley as they kick up dust. The surrounding valley has been razed to the ground for future farmland, and are littered with taken down trees and tree stumps. The tops of the hills are still covered in luscious tropical greenery. Make this image look photorealistic.
Leave a Reply