Project 2: Exploring Variations on a Theme

[glegrady]

Project 2: Exploring Variations on a Theme

The goal of the course is to gain an understanding of how generative image synthesis works in producing an image in response to a text or image prompt.

Whereas the first assignment was open-ended, in this assignment the intent is to explore variations but keeping the same or very similar text prompt. Create an image or video of your choice and do a number of variations of that image or video, keeping the same text prompt, or adding minor changes to the prompt re-ordering the sequence of words around to see to what degree the placement of a word has an efect. The intent is to see to what degree you can transform, stretch, turn inside out, change the meaning, etc. of the AI generated result meanwhile keeping some of its characteristics the same. Explore in greater detail MidJourney's two parameter settings, the negative "--no" parameter where you tell it what you dont want in the image, and the word weight prompt ith the colons as mentioned by Jazer: https://docs.midjourney.com/docs/multi-prompts

---
We are testing text prompts: What kind of an image is produced from a text as the training set consists of images that are tagged by words.

What if the text is randomized like in Gysin's "cut-up technique": https://www.briongysin.com/cut-ups/

or what if the text is just a list: Acidity and basicity, Catalytic ability, Chemical bond formation, Chemical reactivity, Coordination number, Corrosivity Electronegativity, Enthalpy of formation, Flammability, Heat of combusion, Oxidation states, Radioactivity, Solubility, Toxicity

or the text is nonsense or numbers, or exclamations, or produced in ChatGPT based on "give me a sentence with 5 meaningful interesting words, where the sequence of the words can be re-ordered in 5 ways and each sentence's meaning changes dramatically"

For each image provide the text or image prompt, and an analysis of the results.

[emma_brown]

`hideous person`

`hideous girl`

`hideous`

```
ugly::1
girl::1
```

```
ugly::2
girl::1
```

```
ugly::5
girl::1
```

`most ugly girl in the world`

* symmetry
* photos of actors/actresses?
* taking average of all faces makes them look better
* ugly/hideous == horror

---------------------------------------------------------------------------------------------------------------------------------------------
* "Man" is often associated with words like "computer" and "professional," while "woman" is associated with "homemaker" and "nurse."
The analogy "Man is to computer programmer as woman is to homemaker" often holds true in these models.
* Names typically associated with African American individuals were found to be more likely to be associated with unpleasant words compared to names typically associated with European American individuals.
* When generating text about doctors, CEOs, or scientists, models often default to using male pronouns.
Conversely, when generating text about nurses, teachers, or secretaries, they often default to female pronouns.

* I can see the styles being carried over across sessions like Payton was saying

[squire]

For this assignment, I took a section of the opening stanza of Allen Ginsberg's "A Supermarket in California." I picked it because it is both abstract and rich in imagery, and I wanted to keep in both the concrete, representational parts of the poem and the abstract exclamations. I chose to leave in the ambiguous parts of the text, as I wanted to see how they would effect the more visual aspects.

In my hungry fatigue, and shopping for images, I went into the neon fruit supermarket, dreaming of your enumerations!
What peaches and what penumbras! Whole families shopping at night! Aisles full of husbands! Wives in the avocados, babies in the tomatoes!—and you, Garcia Lorca, what were you doing down by the watermelons?

For the next iteration, I set the "weird" factor to 500; I wanted to allow for some of the natural strangeness that Ginsberg's work conjures (however, I wanted to shy away from simply just increasing the weirdness with every iteration. I feel like if I asked Midjourney to continually make the images weirder, and then they got weirder, there really isn't much research or exploration there! So I wanted the surrealness of the poem to come through in other, more indirect ways going forward). I liked these images, particularly the the last one, on an aesthetic level, but still felt they failed to match the tone of the poem: a dreamlike fascination with both the abundance and the commodity fetishism of the "American Supermarket" --weird 500

I think these images were the most interesting of the bunch. The only modification I made to the previous prompt (with weirdness) was to add --no fruit. I was hoping to increasingly add more "--no" tags, first removing the fruit, then the people, then perhaps the whole store together--what were to happen if I asked Midjourney to make an image that excluded all of the substance of the prompt? Unfortunately, I found that Midjourney only allows you to have one "--no" filter per prompt. The most compelling images to me were the second and third images, which felt very ambiance-driven. The second image with its orange and red tones and subversion of scale, and the third image with a sort of found-footage eeriness. I particularly liked the employment of a lower-resolution, which I feel was able to add a surrealness that felt more authentic than simply making the images "weird." Something perhaps worth noting is that two of these images seem to focus on people of Central / South American heritage, and the very first image has a style that may draw a bit on Mexican muralism. I think this may have something to do with Ginsberg's reference to the Spanish poet Garcia Lorca, but I'm unsure why it is only noticeable in this specific prompt. --weird 500 --no fruit

The adjustment made here was the addition of :: low resolution to the prompt. Honestly, I found these images quite boring compared to the previous two iterations, but I liked the motion blur added to the figure in the first image and again felt that it contributed to a sense of dreaminess. :: low resolution --weird 500 --no fruit

At this point, I changed --no fruit to --no food, hoping to "empty" the supermarket. Produce is still certainly on view, so it didn't really take the prompt that well, but I liked what was happening with the metal boxes on the right of the first image; how, now empty, they become much more abstract. (:: low resolution --weird 500 --no food)

Here I made three changes: I changed "food" to "products," added the title of the poem ("a supermarket in california") and added a request for a Dutch angle. I was surprised to get more variation between these images than I had between the last few prompts. The third struck me as the most "American" compared to previous iterations, and though it said "no products" I liked how this illustration gave a sense of "branding" that felt very American to me. I found the last image the most compelling: indiscernible... meat? next to some sort of... potato? This image bore no resemblance to neither the poem nor earlier versions of the prompt, and the text seemed to be based off some sort of Cyrillic language rather than a Latin one. I felt this was a good stopping point as, despite keeping the base text the same the entire time, the image had been reduced down to a ground up viscera representing nothing and completely detached to its source material.

a supermarket in california:: In my hungry fatigue, and shopping for images, I went into the neon fruit supermarket, dreaming of your enumerations! What peaches and what penumbras! Whole families shopping at night! Aisles full of husbands! Wives in the avocados, babies in the tomatoes! and you, Garcia Lorca, what were you doing down by the watermelons?:: low resolution:: dutch angle --weird 500 --no products --style raw

------------------

Several times during the prompting process, I got this alert from Midjourney. I'm not totally sure I understand it--why would it create content that it deems against its own community guidelines? And how does it "know" enough to recognize these violations and yet not enough to avoid generating them? And what specific parts of the text I provided led to the creation of an image that was lewd or violent? I don't know! But I found it very interesting.

[pcroskey]

Prompt 1: Black texan kids sinking into mushroom field surrealism --chaos 25 --ar 16:9 --style raw --weird 3000

I added "Texan" to evoke Southern imagery, but this did not seem to influence the image too heavily. The term "sinking" was also lost here. However, I like the way the mushrooms tower over the children. The choice of neutral colors are also appreciated.

Prompt 2: (referencing previously generated image) sinking Black texan kids into mushroom field surrealism --chaos 25 --ar 16:9 --style raw --weird 3000

Moved the word sinking to the top of the prompt and swapped surrealism back in. I really like this image. The colors and shimmer resulting from a beaming sun are beautiful. However, especially because it is supposed to be depicting children, I would prefer it if she were wearing clothes. (Lack of clothing on Black children was a recurring issue throughout my journey.)
The lack of clothing on the child may also be the reason that I was unable to generate any other results using this image

Prompt 3: (referencing other prev. generated image) Black texan kids sinking into mushroom field surrealism --chaos 25 --ar 16:9 --style raw --weird 1000

I bumped down the weirdness to 1000 with the hopes of receiving anatomically correct bodies. This seemed to work. Still dreaming of kids wearing clothes, though.

Prompt 4: (referencing other prev. generated image) Black texan kids overalls hat sinking into mushroom field surrealism --style raw

Took away all parameters except style raw. First time receiving an image where the child is looking at the camera. Most realistic body.

The following prompts used a random number generator to scramble prompt 4

Prompt 5: texan field clothed sinking black mushroom kids into surrealism --chaos 25 --ar 16:9 --style raw --weird 1000

Flying as opposed to sinking here.

Prompt 6: into clothed field mushroom kids surrealism texan black sinking --chaos 10 --ar 16:9 --style raw --weird 1000

Lost the kids/people entirely. "Clothed" doesn't seem to be impacting anything either.

Prompt 6: into clothed field mushroom kids surrealism texan black sinking

Human-like creatures, but not children, were generated. Very ominous, likely due to the phrase "black sinking."

[vkarthikeyan]

Prompt: a mosaic of mirrors in a chamber with boy's face reflected --style raw --s 250

Was looking for an image of nested reflections in mirrors. The set of images generated seems to have a lot of glass mirror surfaces but no nested reflections in any of them. The last one perhaps comes closest to what I had in mind.

Prompt: a house of mirrors in mosaic pattern reflecting boy's confused face --stylize 250 --style raw

Changed the prompt slightly to make description of space clearer. Still seems more like a collage or assemblage of mirror surfaces this time with no coherent reflections.

Prompt: a house of mirrors in mosaic pattern reflecting boy's confused face --stylize 250 --style raw - Variations (Strong)

Used the Variation (strong) option to see if the model throws up something drastically different. Looks similar in construction just another random assemblage. The thing that seems glaring to me is the lack of coherent reflections. I realized I didn't include the word nested so that was going to be my next text modification. Also noticed saying "boy" expectedly produced a white Caucasian face.

Prompt: a house of glass mirrors in mosaic pattern each reflecting brown indian boy's confused face --weird 600 --style raw --s 250

Changing the prompt for boy's ethnicity instantly produced relevant results. The fourth image seems interesting from a construction perspective; feels like an architectural space/chamber which I realized wasn't very clear in the text prompt but still kind of approximated what I was after. Still no coherent reflections.

Prompt: brown boy's confused face nested reflections infinitely in a house of mirrors mosaic pattern --weird 300 --stylize 200 --style raw

At this point, I felt the model was not "getting" what I wanted at all. I added the word "nested" to make sure it was clear that the reflections needed to be in relation to each other but this still did not produce the desired results. The first image to me stood out to me for its weird construction—something about it reveals to me the patterns of the model's generation technique even though it appeared not to be following my prompt in any way.

Attempt #2
Prompt: boy in house of mirrors infinity mirror sees his own reflections --ar 16:9 --stylize 50 --style raw
v1

v2

This time I seemed to be getting closer to my original intended image.Changed the prompt slightly to include "own reflections". Now the model seems to get what I'm saying. Boy appears in a chamber with multiple mirrors. The reflections aren't totally coherent but I feel generally satisfied with the construction of the elements. The style seems to be somewhat sci-fi inspired (especially the violet lights) but perhaps this has to do with the Stylize option

Prompt: brown boy looks at own reflections infinity mirror effect house of mirrors studio lighting --ar 16:9 --style raw --c 20

Weird images again (did not use Weird option!) Model does not like or know brown I guess, but it does look like it tried responding to that in some of the images. Not what I expected in terms of the updated lighting prompt either (studio light). The one image that stood out was the boy standing in front of what looks like glass pillars of sorts -- with NO reflections. Again, don't think it gets what I'm trying to say with "infinity mirror effect". The last one was a variation tried on the boy with pillars image, the reflections seem lightly better but still has some wonkiness to it.

Prompt: boy in house of mirrors infinity mirror reflected himself everywhere --ar 16:9 --stylize 10 --style raw --chaos 15 --weird 750

Changed the prompt slightly again adding some weirdness and chaos into the earlier one. Images 1 and 3 seem interesting in that they seem to be closer to my intended result with the reflections.The reflections still don't look all that coherent but at least they're better than the earlier images.

[yuehaogao]

Assignment 2: Imagined Chinatown Feast
Yuehao Gao

For this assignment, I will test Midjourney on its ability to generate pictures automatically according to an entry prompt. Specifically, while the prompt describes a general outline of the image, including the vibe, the major objects, and a few additional requirements, what this little project is mainly testing is the influence of different parameters starting with the symbol "--", and looking at how the combination of those parameters influence the output in general.

One theme I thought of is about "delicacies", specifically Chinese food, as they are non-domestic cultural elements for Midjourney datasets. Indeed, this dataset should be well-trained as there are plenty of Discord users uploading food pictures and labeling them "Chinese food." Hence, the initial prompt entered into the Midjourney Bot is:
"A bustling Chinatown restaurant with a large round table filled with traditional Chinese dishes for a feast. --style raw --s 250", with the parameters automatically generated as default values. Here is what I have got:

Then, the parameter of "--no" is tested. Since it is interesting to see how all the pictures generated in the first version include hanging lanterns, to some extent, it would be interesting to see how the model avoids that element. The result is that instead of replacing the lanterns on the ceiling with normal lights in restaurants, the model simply turned its camera angle down to the table and only focused on the dishes.
"A bustling Chinatown restaurant with a large round table filled with traditional Chinese dishes for a feast. --no lantern --style raw --s 250"

To compare the effect of "--no" with specifying "do not draw something" in plain text, while knowing that Orange Chicken and Broccoli Beef are not authentic Chinese dishes, I tried: "A bustling Chinatown restaurant with a large round table filled with traditional Chinese dishes for a feast. ABSOLUTELY NO ORANGE CHICKEN OR BROCCOLI BEEF! --no lantern --style raw --s 250" and got this:

It is glad to see that while some of the previous results contain these two dishes, or some blurry dishes that look very similar to them, this generation has avoided them. This is to say, both the argument of "--no" and specifying "do not have something" will work as effective tools to tell the model to avoid some specific element.

The next parameter being tested is "--chaos", which gives more interesting feedback as its value gradually increases. For instance, when its value is set to 50 instead of 0 as default, it shows more elements like the Californian rural-area background scene, chefs, customers, or a group of participants that demonstrate a "bustling restaurant" even better. The prompt "A bustling Chinatown restaurant with a large round table filled with traditional Chinese dishes for a feast. ABSOLUTELY NO ORANGE CHICKEN OR BROCCOLI BEEF! --no lantern --chaos 50 --style raw --s 250" got these:

hen the value grows to 100, the maximum, something more interesting shows up: a beautiful mountain scene in the background or a person in a mask being interviewed. However, problems have also shown up, including only showing a table without the "bustling" vibe in P4. The prompt "A bustling Chinatown restaurant with a large round table filled with traditional Chinese dishes for a feast. ABSOLUTELY NO ORANGE CHICKEN OR BROCCOLI BEEF! --no lantern --chaos 100 --style raw --s 250" gave me the following pictures:

What is more, the P2 is a "disaster": it brought back Broccoli Beef even if the prompt specified "not so", with a customer with a weird "winky-like" face, as if the model deliberately brought these elements back. Therefore, while a higher "--chaos" value brings more creativity to the model, preciseness may decrease as a contrary.

Next up is the parameter of "--weird," with the prompt "A bustling Chinatown restaurant with a large round table filled with traditional Chinese dishes for a feast. ABSOLUTELY NO ORANGE CHICKEN OR BROCCOLI BEEF! --no lantern --weird <1000 or 3000> --style raw --s 250" separately setting its value to 1000 and 3000, somewhere in the middle and the maximum value, respectively. While the result did not get as distorted as expected, it is noticeable that the amount of dishes, as well as the spacing between the dishes, increases with the parameter. At the same time, it shows more guests around the table with a joyful smile. Indeed, the general vibe created by higher "--weird" values is more precise and welcoming, which, for me, is a surprising discovery. The following are the results separately with a 1000 and 3000 "--weird" value:

I have also tried changing the value of "--iw", the parameter that controls the weight of images and words. However, the prompt "A bustling Chinatown restaurant with a large round table filled with traditional Chinese dishes for a feast. ABSOLUTELY NO ORANGE CHICKEN OR BROCCOLI BEEF! --no lantern --iw <1 or 3> --style raw --s 250 " did not make significant changes to the picture. I doubt if it is because the prompt did not specify the content of the texts, so the system treated the text input as empty, therefore only generating image contents:

The following parameter being tested is the "--ar", the ratio of the output picture. It is very effective, as the prompt "
A bustling Chinatown restaurant with a large round table filled with traditional Chinese dishes for a feast. ABSOLUTELY NO ORANGE CHICKEN OR BROCCOLI BEEF! --no lantern --ar 4:3 --style raw --s 250" generates the following, which is obviously in the dimension ratio of 4:3:

The last parameter I tested is the "--quality" parameter, which has a default value of 1, but by changing it to 0.5 by the prompt "A bustling Chinatown restaurant with a large round table filled with traditional Chinese dishes for a feast. ABSOLUTELY NO ORANGE CHICKEN OR BROCCOLI BEEF! --no lantern --ar 4:3 --q 0.5 --style raw --s 250", the picture quality does decrease to some extent:

In general, most parameters have effects on the output of the images generated by the Midjourney Model. Despite there being an exception: the "--iw" parameter, due to lack of input content in the prompt, other parameters are influencing the results more or less. Especially, the argument "--no" has effectively stopped the machine from including a specific element, despite the system only allowing one "--no" in the prompt, but still, it overwhelmed stating "not having something" by plain text, especially as the "chaos" value grows. Still, it is indispensable to say that the model understands well how a "Chinese feast" should look and makes accurate vibes, which will be useful for artistic creations or even commercial uses in relevant topics in the future.

[borouyu]

Authorship and Interpretation through Generative Art

This project explores the intersection between generative image synthesis and the poststructuralist theory outlined by Roland Barthes in his essay "The Death of the Author". By employing sentence and word order variations, cut-up techniques, and prompt manipulations, the work reflects on how AI reinterprets meaning when freed from fixed authorial intent. The artwork becomes a practical engagement with Barthes’ idea that the text (or prompt) is a site of open-ended interpretation, demonstrating how meaning shifts with each textual manipulation.

In this process, the original prompt—the Barthes quote itself—functions as a symbolic fragment. The act of changing the syntax and sequence parallels Barthes’ notion that the death of the author enables meaning to proliferate freely, without a singular source. Each AI-generated image serves as a visual response, embodying the fluidity of interpretation and the disconnect between the "author's" (my) intent and the AI’s output as the new symbolic practice.

AI as the New Author?
The work also engages with questions of authorship and creative agency in the age of AI. With the AI functioning as both an interpreter and a creator, who now holds the authorship of the final image? Barthes suggested that the author’s death gives way to the birth of the reader as the new locus of meaning—here, the AI becomes both reader and author, its output standing independently from both the original quote and my prompts. This ambiguity between human input and machine creativity reflects the shifting boundaries between authorship, interpretation, and production in generative art.

Midjourney Setting --ar 1:1 --v 6.1 --style 100

Original Prompts:
As soon as a fact is narrated no longer with a view to acting directly on reality but intransitively, that is to say, finally outside of any function other than that of the very practice of the symbol itself, this disconnection occurs, the voice loses its origin, the author enters into his own death, writing begins.
From “The Death of the Author”, Roland Barthes, first published in 1967.

Analysis: Surrealism style, Author in the image, merging with elements like cloud, smoke, ink and mountain

Prompt Analysis:
1 Sentence, 7 Subsentences, 56 words, 6 commas, 1 period

1 As soon as a fact is narrated no longer with a view to acting directly on reality but intransitively,
2 that is to say,
3 finally outside of any function other than that of the very practice of the symbol itself,
4 this disconnection occurs,
5 the voice loses its origin,
6 the author enters into his own death,
7 writing begins.

Variation1 - Sentence Sequence Change:
Reverse order 1234567 - 7654321
writing begins. the author enters into his own death, the voice loses its origin, this disconnection occurs, finally outside of any function other than that of the very practice of the symbol itself, that is to say, As soon as a fact is narrated no longer with a view to acting directly on reality but intransitively

Analysis: more realistic author image and writing action, probably because the sentence has “writing begins. the author” in the beginning and more weight in the front.

Random order 1234567 - 4315726
this disconnection occurs, finally outside of any function other than that of the very practice of the symbol itself, As soon as a fact is narrated no longer with a view to acting directly on reality but intransitively, the voice loses its origin, writing begins. that is to say, the author enters into his own death,

Analysis: circular element, the person is in meditation, probably because keywords like “disconnection”, “practice of the symbol” in the beginning.

Sentence Sequence Conclusion: The sequence of prompts influences the image generation, those in the beginning have more weights than those in the end.


Variation2 - Text Cut-up:
https://chardin.neocities.org/cut_up
2 words, 3 words, 4 words, 5 words, bigger phrase size keeps more original text intact

5 words Cut up Prompts
Its origin the author enters on reality but, Intransitively that of any function other than is, To say finally outside that of the very, Practice begins into his own death writing disconnection, Occurs the voice loses is, Narrated no longer with as, Soon as a fact of, The symbol itself this a

Analysis: The author’s head blends with trees, smoke and architecture. Really symbolic.

4 words Cut up Prompts
Disconnection occurs the voice the very practice, Of the symbol itself this author enters, Into his to acting directly on fact, Is narrated no other than that of, Outside of any function reality but intransitively, That own death writing begins loses its, Origin the as soon as a, Longer with a view is to say

Analysis: Similar to the previous one, the author’s head blends with branches, earth and sky. “Disconnection” turns it into black and white

3 words Cut up Prompts
This disconnection occurs his own, Death outside of any a, Fact is writing begins its, Origin the with a view, Author enters into that of, The narrated no longer function, Other than to say finally, As soon as on reality, But the voice loses the symbol, Itself very practice of intransitively that

Analysis: Two images have skeletons, because “death” comes early. Two images have realistic figure. “Disconnection” turns it into black and white

2 words Cut up Prompts
As a reality but outside of the, Symbol writing begins any function own death, To acting practice of is to disconnection, Occurs intransitively that that of itself this, Directly on loses its as soon say, Finally the very other than fact, Is into his origin the longer with, The voice narrated no a view author

Analysis: Abstract writing and skull elements with color. Because the “Symbol writing”, “death” in the early prompts.

Cut Up Conclusion: Cut up is a random process, with fewer words cut up, the more complicated and far away from the original meaning. And each phrase and word becomes independent elements, then the earlier they show up in the prompts, the more influences they have in the image generation process.


Variation3 - Word Sequence Change:
Reverse order
https://www.dcode.fr/reverse-writing
begins writing. death own his into enters author the, origin its loses voice the, occurs disconnection this, itself symbol the of practice very the of that than other function any of outside finally, say to is that, intransitively but reality on directly acting to view a with longer no narrated is fact a as soon As

Analysis: very realistic writing and skull elements in the image. Because the “writing”, “death” in the very beginning.

Random order
https://onlinetools.com/random/shuffle-words
is occurs, intransitively, practice disconnection to fact is own but say, the that to loses function this any of its as a very writing As author itself, directly acting the the than view narrated of of voice the finally reality other with soon no outside enters on a longer origin, begins. death, symbol his that into

Analysis: two images are very realistic skulls. Another two are abstract figures in painting style. Probably the randomized process breaks down the original meaning, so AI will have different interpretation each time.

Conclusion: word sequence change means one word cut-up, so it has a similar influence to the cut up process. The words ”death” “author” “writing” are words that have a clear visual image relationship, so they will dominate the image.


Variation4 - Midjourney “–no” Prompts
--no + Original Prompts:
123456 --no 7
As soon as a fact is narrated no longer with a view to acting directly on reality but intransitively, that is to say, finally outside of any function other than that of the very practice of the symbol itself, this disconnection occurs, the voice loses its origin, the author enters into his own death, --no writing begins.

Analysis: red circle elements, trees, painting style. “–No writing begins” get rid of writing, book elements, also influence the emotions of the images.

12345 --no 67
As soon as a fact is narrated no longer with a view to acting directly on reality but intransitively, that is to say, finally outside of any function other than that of the very practice of the symbol itself, this disconnection occurs, the voice loses its origin, --no the author enters into his own death, writing begins.

Analysis: a lady shows up in two images. In other two images a graphic compositions show up. With ”--no the author enters into his own death, writing begins.” the male person is removed, and the dark feeling of images is gone.

1234 --no 567
As soon as a fact is narrated no longer with a view to acting directly on reality but intransitively, that is to say, finally outside of any function other than that of the very practice of the symbol itself, this disconnection occurs, --no the voice loses its origin, the author enters into his own death, writing begins.

Analysis: the more –no, the further away from the original images and meaning.

--no + Text Cut-up:
--no + 2 words Cut up Prompts
As a reality but outside of the, Symbol writing begins any function own death, To acting practice of is to disconnection, Occurs intransitively that that of itself this, Directly on loses its as soon say, Finally the very other than fact, Is into his origin the longer with, --no The voice narrated no a view author

Analysis: skull figures are removed from the original images. Images become abstract.

--no +Random word order
https://onlinetools.com/random/shuffle-words
is occurs, intransitively, practice disconnection to fact is own but say, the that to loses function this any of its as a very writing As author itself, directly acting the the than view narrated of of voice the finally reality other with soon no outside enters on a longer origin, begins. --no death, symbol his that into

Analysis: “--no death, symbol his that into” turns the images into a more realistic style, with female figures.

“–no” Prompts Conclusion: The “--no” will remove the related elements, and generate surprising new images.


Variation5 - Midjourney Multi Prompts
Multi Prompts + Original Prompts 1
As soon as a fact is narrated no longer with a view to acting directly on reality but intransitively, that is to say, finally outside of any function other than that of the very practice of the symbol itself, this disconnection occurs, the voice loses its origin, the author enters into his own death writing :: begins.

Analysis: Separate “begins” did not change much of the image style.

Multi Prompts + Original Prompts 2
As soon as a fact is narrated no longer with a view to acting directly on reality but intransitively, that is to say, finally outside of any function other than that of the very practice of the symbol itself, this disconnection occurs, the voice loses its origin, the author enters into his own death :: writing :: begins.

Analysis: Separate “begins” and “writing”, book and writing elements show up, four images are quite different.

Multi Prompts + Original Prompts 3
As soon as a fact is narrated no longer with a view to acting directly on reality but intransitively, that is to say, finally outside of any function other than that of the very practice of the symbol itself, this disconnection occurs, the voice loses its origin, the author enters into his own :: death :: writing :: begins.

Analysis: more abstract, painting style

Multi Prompts + Original Prompts 4
As soon as a fact is narrated no longer with a view to acting directly on reality but intransitively, that is to say, finally outside of any function other than that of the very practice of the symbol itself, this disconnection occurs, the voice loses its origin, the author enters into his :: own :: death :: writing :: begins.

Analysis: no big changes

Multi Prompts + Original Prompts 5
As soon as a fact is narrated no longer with a view to acting directly on reality but intransitively, that is to say, finally outside of any function other than that of the very practice of the symbol itself, this disconnection occurs, the voice loses its origin, the author enters into :: his :: own :: death :: writing :: begins.

Analysis: the visual style changed dramatically into collage style. Skulls and head body elements. Colorful palette.

Multi Prompts Conclusion: multi prompts emphasize the meaning of each individual word that is separated. And it could then mix the meaning and create totally different understanding and images.


“Human Text Author - AI Text Reader - AI Image Author - Human Image Reader”
The whole process highlights the tension between control and unpredictability in generative art: while prompts are crafted intentionally and randomly, the AI understands the texts as the reader and recreates the images as the new author, introducing emergent elements that transcend authorial intent. And I am interpreting the resulting images with my own interpretations as the reader.

[yiranxiao]

Code: Select all

an ancient temple in a forest --no trees

midjourney seems to have difficulty generating images that match contradictory descriptions. If I use the word “forest” but emphasize that there are no trees, then the '--no' parameter is useless. Here it is not difficult to see that midjourney generates images that first match the prompt and then pay attention to other parameters.

Code: Select all

a concert with a lively crowd--no audience

In this one, I changed '--no people' to '--no audience', which looks much more reasonable

Code: Select all

mountain ::1 reflection ::1 in a calm lake ::1 --chaos 50

For prompt weights, I start with the same weight and then use a larger weight on 'mountain'
For this one, the image appears symmetrical

Code: Select all

Mountain ::3 reflection ::1 in a calm lake ::1

Code: Select all

Acidity, basicity, catalytic ability, chemical bond formation, chemical reactivity

Using lists here, I asked ChatGPT to help me generate a list of technical terms and observe whether midjourney generates literal representations or abstract interpretations, and consider how technical terms influence the color scheme, shapes, textures.

Code: Select all

Electronegativity, enthalpy of formation, flammability, heat of combustion

only very rarely do the generated images match all the words in the list. I have observed that in general, the 4 generated images each have a different weight distribution for the words in the list. For example, some images match 'flammability, heat of combustion', but lack expression of 'electronegativity, enthalpy of formation'. The following image is one of the visual representations that I feel basically matches all the words in the list.

Code: Select all

Fractured time, cascading dreams, ephemeral reality

How this image express 'cascading dreams' is very interesting

[jazer]

a hippo throwing a ball ::1 a tiger catching a ball thrown by a hippo ::1 a warthog looking at a tiger catching a ball ::1 --chaos 50 --ar 16:9 --style raw --stylize 0 --weird 1500 --v 6.1

an adult throwing a ball ::1 a child catching a ball ::2 a snake looking at a child catching a ball ::1 --no deviant art --chaos 70 --ar 16:9 --style raw --stylize 200 --weird 2100 --v 6.1

an adult throwing a ball ::1 a child catching a ball ::1 a snake looking at a child catching a ball thrown by an adult ::4 --no jungle --chaos 70 --ar 16:9 --style raw --stylize 200 --weird 2000 --v 6.1

a beautiful waterfall ::1 a huge decaying pile of computers and robots in a pool at the bottom of a beautiful waterfall ::2 --no deviant art --chaos 60 --ar 16:9 --style raw --stylize 0 --weird 2500 --v 6.1

a beautiful waterfall ::1 an extremely huge decaying pile of computer robots in a pool at the bottom of a beautiful waterfall ::3 --no cars trash boxes crates --chaos 60 --ar 16:9 --style raw --stylize 200 --weird 2500 --v 6.1

a very huge pile of beautiful computer robots in a pool at the bottom of a decaying waterfall of green slime --no deviant art cars boxes crates water --chaos 70 --ar 16:9 --style raw --stylize 200 --weird 2100 --v 6.1

[figueroasanabria]

Prompt 1: "Imagine Contemporary architecture, hyperrealism, hyperrealistic, photorealistic, daylight"
--

Prompt 2: "imagine Antonio Gaudi inspired architecture, hyperrealism, hyperrealistic, photorealistic, daylight, organic forms"

Prompt 3: "Imagine Contemporary architecture inspired by Antonio Gaudi, hyperrealism, hyperrealistic, photorealistic, daylight, colorful, --no interior"

Prompt 4: "Imagine Contemporary architecture inspired by Zaha Hadid, hyperrealism, hyperrealistic, photorealistic, daylight, -- no interior"

prompt 5:
"Buildings made of translucent, pastel-colored, softly glowing under a dreamlike sky, ethereal with flowing organic forms, blending with surrounding, minimalistic urban design"