Images situated terminology: Difference between revisions
No edit summary |
|||
(17 intermediate revisions by the same user not shown) | |||
Line 1: | Line 1: | ||
== Processual images == | == Processual images == | ||
The process of diffusion that lies at the heart of image generation gradually translates an image made of noise into an image matching a prompt. The samples below show the various stages through which the generator go through in order to produce the final image. The first noise image is on the top left, the result is on the bottom right. | |||
[[File:Montage-circles.png|frame|none]] | [[File:Montage-circles.png|frame|none]] | ||
All the steps towards an image generated with Stable Diffusion XL and the prompt "Thinking, writing, practicing in circles" | |||
[[File:Montage-shared-present.png|frame|none|Image generated with the prompt Shared Present]] | |||
All the steps towards an image generated with Stable Diffusion XL and the prompt "shared present across divisive cultural and historical differences; of a temporary unity of the present across the planet" | |||
[[File:Montage-aesthetics-political.png|frame|none|alt=This ikmage was generated with the prompt "The aesthetic is political precisely because of its ability to differentiate itself from the normally inconspicuous organization of our everyday lifeworld"]] | |||
All the steps towards an image generated with Stable Diffusion XL and the prompt "The aesthetic is political precisely because of its ability to differentiate itself from the normally inconspicuous organization of our everyday lifeworld" | |||
== Reflexive images == | |||
When generating images, computer scientists want to know how the different words in a prompt affect the image. In the following samples, a generated image is overlaid by a heatmap that represents where a given word affects more strongly the result. The red regions (the hottest) are the most affected by the word and the blue ones are the least affected (cold ones). | |||
[[File:Openness regard-sense.png|frame|none|On the left, the generated image and two heatmaps for the words openness and sense]] | |||
The first sample shows the generated image and two heatmaps. The image on the left has been generated with the prompt: "a certain openness with regard to the sense or signification of the thing that appears". The image at the center is the heatmap for the word ''openness'' and the image on the right shows the heatmap of the word ''sense''. | |||
[[File:Practicing-in-circles.png|frame|none|Heatmap images for each term of the prompt : "practicing in circles"]] | |||
This second sample shows the heatmap images for respectively ''practicing'', ''in'' and ''circles''. | |||
[[File:The-aesthetics-is-political-002.png|frame|none]] | |||
This third sample shows the heatmap images for respectively ''aesthetics'', ''is'' and ''political''. | |||
== Prompt images == | |||
The following images are generated with a more recent model, which is also less polished than the one we used together. I have tried to use its capacity for eeriness that I think might function well with the concept of flamability. | |||
[[File:Amateurish photo of the new york modern museum of art in flames contrast between the white building and the flames at night a crowd of artists dressed as rabbits and donkeys in the fo 2123705561.png|frame|none|Amateurish photo of the new york modern museum of art in flames contrast between the white building and the flames at night with a crowd of disguised artists demonstrating]] | |||
[[File:Amateurish photo of the new york modern museum of art in flames contrast between the white building and the flames at night a crowd of artists dressed as rabbits and donkeys in the fo 3133295262.png|frame|none|Amateurish photo of the new york modern museum of art in flames contrast between the white building and the flames at night with a crowd of disguised artists demonstrating]] | |||
[[File:Amateurish photo of the new york modern museum of art in flames contrast between the white building and the flames at night a crowd of artists in the foreground holding signs that say 3868873032.png|frame|none|Amateurish photo of the new york modern museum of art in flames contrast between the white building and the flames at night with a crowd of disguised artists demonstrating]] |
Latest revision as of 19:40, 26 February 2025
Processual images
The process of diffusion that lies at the heart of image generation gradually translates an image made of noise into an image matching a prompt. The samples below show the various stages through which the generator go through in order to produce the final image. The first noise image is on the top left, the result is on the bottom right.

All the steps towards an image generated with Stable Diffusion XL and the prompt "Thinking, writing, practicing in circles"

All the steps towards an image generated with Stable Diffusion XL and the prompt "shared present across divisive cultural and historical differences; of a temporary unity of the present across the planet"

All the steps towards an image generated with Stable Diffusion XL and the prompt "The aesthetic is political precisely because of its ability to differentiate itself from the normally inconspicuous organization of our everyday lifeworld"
Reflexive images
When generating images, computer scientists want to know how the different words in a prompt affect the image. In the following samples, a generated image is overlaid by a heatmap that represents where a given word affects more strongly the result. The red regions (the hottest) are the most affected by the word and the blue ones are the least affected (cold ones).

The first sample shows the generated image and two heatmaps. The image on the left has been generated with the prompt: "a certain openness with regard to the sense or signification of the thing that appears". The image at the center is the heatmap for the word openness and the image on the right shows the heatmap of the word sense.

This second sample shows the heatmap images for respectively practicing, in and circles.

This third sample shows the heatmap images for respectively aesthetics, is and political.
Prompt images
The following images are generated with a more recent model, which is also less polished than the one we used together. I have tried to use its capacity for eeriness that I think might function well with the concept of flamability.


