wk01 - MidJourney 01

Post Reply
glegrady
Posts: 203
Joined: Wed Sep 22, 2010 12:26 pm

wk01 - MidJourney 01

Post by glegrady » Fri Sep 16, 2022 8:03 am

wk01 - MidJourney 01

Please post 4 to 6 images here with their text query. Following that, do an analysis of the results. For each image discuss the following:

1. Does the image meet your expectations?
2. To what degree does your text query influence the generated image?
3. What is the style of the image, and why do you think it has produced that?
4. Any thoughts about how the visual elements in the image are organized
5. How would you change the query? to achieve what difference?
6. Any other comments?
7. On a scale of 5 - from 5 being GREAT to 1 being LOW your rating of the result
George Legrady
legrady@mat.ucsb.edu

wqiu
Posts: 14
Joined: Sun Oct 04, 2020 12:15 pm

Re: wk1 - MidJourney 1 - 1st Results

Post by wqiu » Sun Oct 02, 2022 6:22 pm

midjourney_1.png
**Prompt: Stroboscopic, Fencing, Sports, Multi-exposure, mild, motion**

1. Does the image meet your expectations?

No really what I was looking for, but it has its own style. I have never seen images created by computer with so organic textures, which gives a feeling of analog film.

2. To what degree does your text query influence the generated image?

The image matches my text description pretty well, except for it kind of correlates the fencing with physical fence, shown as white bars in the image. Things being said, it is still quite different from the image in my mind - Chronophotograph by Étienne-Jules Marey.

3. What is the style of the image, and why do you think it has produced that?

Image looks like a motion blurred film photo. The motion blur should be caused by the keyword “motion”. The silver-grain-like texture might be caused by keyword “stroboscopic and multi-exposure”.

4. Any thoughts about how the visual elements in the image are organized

Too symmetrical than my taste.

5. How would you change the query? to achieve what difference?

I would change “fencing” to “fencer”. I will also change the aspect ratio to 16:9.

6. Any other comments?

I like the middle person being sharp while the other two people being blurred.

7. On a scale of 5 - from 5 being GREAT to 1 being LOW your rating of the result

4/5



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

weihao_chronophotograph_stroboscopic_fencing_sports_multi-expos_eea6ea9f-5daa-4b15-b737-f6b5deab64b4.png
Prompt: chronophotograph, stroboscopic, fencing, sports, multi-expos

1. Does the image meet your expectations?

It surprised me. It is very abstract, though still different from what I imagine.

2. To what degree does your text query influence the generated image?

I misspelled the [multi-exposure] as [multi-expos]. The white bars were created because of [fencing], so does the white man.

3. What is the style of the image, and why do you think it has produced that?

The image is abstract, probably there is ambiguity in the word [fencing] and the misspelled word [multi-expos]. The blurred areas should be caused by [chronophotograph][stroboscopic]

4. Any thoughts about how the visual elements in the image are organized

Image follows rule of third and the subject is surrounded by the box. Great composition.

5. How would you change the query? to achieve what difference?

It is kind of an accident. I would reduce the while lines a little bit by change [fencing] to [fencer]

6. Any other comments?

7. On a scale of 5 - from 5 being GREAT to 1 being LOW your rating of the result

4/5

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

Untitled-2.png
Prompt: **fencers, action scene, win, olympics game, Ilford HP5, sideview
—no ice hockey**

1. Does the image meet your expectations?

Almost met my expectation, but the image was not perfect sideview. The camera was close to one side and further to the other side.



2. To what degree does your text query influence the generated image?

Matched pretty well.

3. What is the style of the image, and why do you think it has produced that?

Black white, motion blur, decisive moment image.

4. Any thoughts about how the visual elements in the image are organized

it looks like typical action scene composition.

5. How would you change the query? to achieve what difference?

I changed [fencers] to [two fencers] to limit the people in the photo to two. It does not work the well. Also, the image doesn’t look silver-grainy of film. I should have add “iso 3200” or “grainy”.

6. Any other comments?

7. On a scale of 5 - from 5 being GREAT to 1 being LOW your rating of the result

4.5/5

--------------------------------------------------------------------------------------------
Untitled-3.png
prompt: **two fencers, action scene, olympics game, Ilford HP5, win, sideview**

**+
ice hockeyBOTMidjourney Bottwo fencers, action scene, olympics game, Ilford HP5, score, sideview**

**+
ice hockey**

1. Does the image meet your expectations?

I was expecting a Black White photo because of the Ilford HP5 (name of a type of Black White film), but it ended up with a colored image. Again, I am surprised by the realistic look of the image and the pose of the right person who is taking off his/her helmet. There was also an accidental copy-paste error and caused repetitive words in the prompt.

2. To what degree does your text query influence the generated image?

It matches pretty well, except for the Ilford HP5. I think it was conflicting with olympics game. The background of city was amazing, probably caused by “action scene”. The pose was affected by the “win” keyword.

3. What is the style of the image, and why do you think it has produced that?

realistic, cinematic image. “action scene” caused it.

4. Any thoughts about how the visual elements in the image are organized

it is a little bit unbalanced, the right is too dark and dense and the left is too bright and empty.

5. How would you change the query? to achieve what difference?

I changed sideview to Side-View and it seems resolved the issue with unbalanced composition.

6. Any other comments?

7. On a scale of 5 - from 5 being GREAT to 1 being LOW your rating of the result

4.5/5



--------------------------------------------------------------------------------------------
Vislab-MAT1_two_fencers_action_scene_olympics_game_Side-View_fl_6632bf70-4a03-4485-80b0-6974c449f3fa.png
**two fencers, action scene, olympics game, Side-View, flash, moment, clear**

1. Does the image meet your expectations?

No, I was looking for a “clear” image but still gave me the motion blur. I was looking for “side-view” but it was not purely side-view.

2. To what degree does your text query influence the generated image?

Once I removed “Ilford HP5” it becomes colorful. The background looks like a flag of certain country, caused by “olympics game”.

3. What is the style of the image, and why do you think it has produced that?

It looks like sports photography. “Moment” “clear” produced this image of a very exciting moment.

4. Any thoughts about how the visual elements in the image are organized

Pretty typical sports photography composition with very interesting background. The middle area of the background fades out to white which is very interesting.

5. How would you change the query? to achieve what difference?

I removed “Olympic Games” to remove the background.

6. Any other comments?

7. On a scale of 5 - from 5 being GREAT to 1 being LOW your rating of the result

4.5/5

--------------------------------------------------------------------------------------------
Untitled-4.png
prompt: DALL·E 2022-09-30 10.25.11 - Stroboscopic, Chronophotograph, fencing, white suits, illford, side-view, series, long, many

1. Does the image meet your expectations?

It is beyond my expectation.

2. To what degree does your text query influence the generated image?

The image matches the description.

3. What is the style of the image, and why do you think it has produced that?

It looks like the color film in the 70s. Like the first image in this link (https://imgur.com/gallery/LActU). The “Stroboscopic” keyword caused it but same thing was not produced in Midjourney. I think the dataset DALLE-2 used must contain the photographs of 70s.



4. Any thoughts about how the visual elements in the image are organized

I like the near-symmetric composition. It reminds me of work of Georges Méliès’s film.

5. How would you change the query? to achieve what difference?

I added Georges Méliès’s name to the query to make the image looks more like his work.

6. Any other comments?

7. On a scale of 5 - from 5 being GREAT to 1 being LOW your rating of the result

5/5

lu_yang
Posts: 9
Joined: Mon Sep 26, 2022 10:23 am

Re: wk1 - MidJourney 1 - 1st Results

Post by lu_yang » Mon Oct 03, 2022 12:35 pm

Image
**Prompt: lebbeus woods drawing, post-apocalyptic building ruins with floating alien structure, experimental architecture, ZAGREB FREE ZONE, heterogeneity, multiplicity, aggregation, heterarchy, architecture as instruments --ar 16:9 --stylize 5000 --chaos 80**

1. Does the image meet your expectations?
No, the style of the image is out of control and different from Lebbeus Woods' drawing style.

2. To what degree does your text query influence the generated image?
The image matches my description text "post-apocalyptic building ruins with floating alien structure". Color and overall tune are restrained by the image prompt
Image

3. What is the style of the image, and why do you think it has produced that?
The style of the image is more organic than expected. It never produces straight lines or topological polygons.

4. Any thoughts about how the visual elements in the image are organized
the image first appears similar to the Walking City by Archigram, which has a monumental structure floating over the city.
Image
First result
Image
After a series of iterations and upscaling, the city on ground level becomes more detailed, and the floating structures are more associated with "alien"

5. How would you change the query? to achieve what difference?
I would reconsider the use of "heterogeneity, multiplicity, aggregation, heterarchy", as they give a rich texture to the image, but do not necessarily match with Lebbeus Woods's drawing.

7. On a scale of 5 - from 5 being GREAT to 1 being LOW your rating of the result
2/5

--------------------------------------------------------------------------------------------
Image
**Prompt: post-apocalyptic building ruins with floating alien structure, experimental architecture, ZAGREB FREE ZONE, heterogeneity, multiplicity, aggregation, heterarchy, architecture as instruments --ar 16:9 --stylize 5000 --chaos 80**

1. Does the image meet your expectations?
Sort of, because I was expecting the multiplicity in the image should be different without prompt "lebbeus woods drawing"

2. To what degree does your text query influence the generated image?
The only difference with the one above is that this one does not have "lebbeus woods drawing" in the prompt. Apparently, this image towards oil painting or photorealistic, while the one above towards more on the drawing, disregards its inaccurate style.

3. What is the style of the image, and why do you think it has produced that?
Organic as usual. I specifically want to show another production a few iterations before the final.
Image
similar images occasionally occur in different prompts and different parameters, the more --style parameters is (the more control from the text), the more it occurs. This is very strange since my text should not lead to an organic style.

5. How would you change the query? to achieve what difference?
Maybe I should improve the text to give a more clear style.

7. On a scale of 5 - from 5 being GREAT to 1 being LOW your rating of the result
2/5

--------------------------------------------------------------------------------------------
Image
**Prompt: lebbeus woods drawing, post-apocalyptic building ruins with floating alien structure, experimental architecture, ZAGREB FREE ZONE, heterogeneity, multiplicity, aggregation, heterarchy, architecture as instruments --ar 16:9 --stylize 5000 --chaos 80**

1. Does the image meet your expectations?
Yes, I expected it will have a sketch style with the prompt "lebbeus woods drawing"

2. To what degree does your text query influence the generated image?
This time I removed the image prompt. It seems the color is affected, but the overall style is still controlled by the prompt "lebbeus woods drawing"

3. What is the style of the image, and why do you think it has produced that?
The texture is organic, it could be due to text "heterogeneity, multiplicity, aggregation"

5. How would you change the query? to achieve what difference?
I want to try different text to change the organic texture.

7. On a scale of 5 - from 5 being GREAT to 1 being LOW your rating of the result
3/5

--------------------------------------------------------------------------------------------
Image
**Prompt: lebbeus woods drawing, post-apocalyptic building ruins with floating alien structure, experimental architecture, ZAGREB FREE ZONE, heterogeneity, multiplicity, aggregation, heterarchy, architecture as instruments --ar 16:9 --stylize 5000 --chaos 80**

1. Does the image meet your expectations?
It is getting closer on the composition of Lebbeus Woods' architecture

2. To what degree does your text query influence the generated image?
I'm using the same text with the same image prompt, but this time I was able to get some random results that create a perfect composition and iterate upon.

6. Any other comments?
One note is I find the results are rather similar to Zdzisław Beksiński's art, such as:
Image
I think Mid Journey is more suitable to create organic structures rather than polygons, I may try Zdzisław Beksiński's style next time.

7. On a scale of 5 - from 5 being GREAT to 1 being LOW your rating of the result
4/5

jiarui_zhu
Posts: 7
Joined: Mon Sep 26, 2022 10:25 am

Re: wk1 - MidJourney 1 - 1st Results

Post by jiarui_zhu » Tue Oct 04, 2022 10:35 am


jkilgore
Posts: 7
Joined: Tue Mar 29, 2022 3:34 pm

Re: wk1 - MidJourney 1 - 1st Results

Post by jkilgore » Tue Oct 04, 2022 2:53 pm

Oops, how do I delete.
Last edited by jkilgore on Tue Oct 04, 2022 4:53 pm, edited 1 time in total.

jkilgore
Posts: 7
Joined: Tue Mar 29, 2022 3:34 pm

Re: wk1 - MidJourney 1 - 1st Results

Post by jkilgore » Tue Oct 04, 2022 4:47 pm

I am on the hunt for pictures of hands. Pictures of hands in the style of a first person point of view, similar to realistic first person shooter video games. Simultaneously, keywords relating to Christian imagery will pop up here and there. I like the idea of video games as theology.
csgo.jpg
Note: The above picture is inspiration for the following prompts. A target to shoot towards.

------------------------------------------------------------
Generated Image 1
------------------------------------------------------------
Vislab-MAT2_hands_in_front_of_a_camera_first_person_unreal_engi_b6637db7-3682-4f88-80d8-b89605be026f.png
prompt: hands in front of a camera first person, unreal engine, first person shooter, photorealistic, 8k

Discussion:
Here is my first somewhat successful attempt. Initially I was just using the keyword first person shooter, but this ended up failing most of the time. Outputs started to somewhat resemble my inspiration picture as I started to use the word "hands" in combination with "first person" and other video game related words.

Aesthetically, the output is quite mediocre. It has the look and perspective of "realistic" first-person video games, but I dislike the textures. The hands are leather-y, but don't look like gloves. There are no fingernails. Quite gross, but in an unappealing way. Further, the setting feels generic.

On the next run, I plan to not rely on "unreal engine", but instead use some specific video game IP that will hopefully push Midjourney to build images with more style.

1/5

------------------------------------------------------------
Generated Image 2
------------------------------------------------------------
Vislab-MAT2_hands_holding_knife_first_person_playing_halo_3_fir_43d51ada-88ab-4f14-94e4-ed1c19aa3f1b.png
prompt: hands holding knife first person, playing halo 3, first person shooter, photorealistic, 8k --quality 2

Discussion:
The generated didn't exactly fit the bill for the original inspiration image. There is only a single finger and a knife placed along the hand. The hand has become more of a surrealist collage of a hand, more than a hand with the ability to grip. Despite it's failures, I like the image. The finger is quite nice; this time it actually has a fingernail and has a texture that is more convincingly an alive hand, as opposed to the mummified hand of the last picture. Further, the perspective is reminiscent of a first person shooter as all the information of note seems to come out of the bottom right corner (notice it's similarity to the composition of the inspiration picture).

It also did quite a good job at picking up on 'Halo 3', generating imagery more reminiscent of the sci-fi style of the franchise. As an aside, writing 'playing {some video game}' rather than {some video game} did a better job at giving first person style compositions.

3/5



------------------------------------------------------------
Part 2: Sticking with One Prompt
------------------------------------------------------------
prompt: hands first person, angels playing counter strike, first person shooter, photorealistic, 8k --quality 2 --ar 16:9

The future images all share the same prompt, but are generated using several applications of the variation parameters. I stuck to a style of image and exhausted its possibilities.

This prompt created a couple key features:
1) Dramatic, golden hour lighting with clouds in the background. Heavenly.
2) Militaristic clothing and helmets
3) Closeups of hands
4) People becoming hands and hands becoming people.
5) Handshakes between people

"Counter Strike" contributed to the militaristic aspects, "angels" contributed to the dramatic lighting, "hands first person" contributed to the hands and mixed with "Counter Strike" to create hand-like people (more on that later).

Notice how the composition and lighting of the shots are derivatives of stereotypical first person shooter advertisements:
csgo_cover.jpg
mw2.PNG

------------------------------------------------------------
Generated Image 3
------------------------------------------------------------
Vislab-MAT2_hands_first_person_angels_playing_counter_strike_fi_f922b7e9-fd66-4d5b-9c04-9fdbcd796ec7.png
Discussion:

If we look back to the features list it out in part 2, we can see that the following image contains (1) dramatic, golden lighting, (3) closeup of hands, and (5) handshakes.

The perspective is quite nice. I enjoy that the two people(ish) are farther away and the hands extend towards the camera.

The hands are great. Obviously hands, but mutilated. It's as if the skin is falling off the bone.

4/5
------------------------------------------------------------
Generated Image 4
------------------------------------------------------------
Vislab-MAT2_hands_first_person_angels_playing_counter_strike_fi_5d403306-ef72-4ba9-a859-e93ec1f74b2a.png
Discussion:

If we look back to the features list it out in part 2, we can see that the following image contains (1) dramatic, golden lighting, and (3) closeup of hands. And wow, in this case, (1) and (3) really combine to create something quite appealing. Again the hand is nice in all the same ways as the last image. The lighting adds a nice glow to the skin, similar to if you were to put a flashlight up to your finger.

This is one of the few variations of the csgo prompt that actually had a convincing first person perspective. It's as if the hand is reaching out from behind the scene.

Interesting note: when I created variations of this image, or ones similar, the images devolve towards branches and trees.

4.5/5

------------------------------------------------------------
Generated Image 5
------------------------------------------------------------
Vislab-MAT2_hands_first_person_angels_playing_counter_strike_fi_e7464df8-3900-44ba-b448-771e57c8db60.png
Discussion:
An example of a hand person.

3.5/5

------------------------------------------------------------
Generated Image 6
------------------------------------------------------------
Vislab-MAT2_hands_first_person_angels_playing_counter_strike_fi_eda0f84b-c354-4861-9d51-ca759c3cf216.png
Discussion:
An example of hands becoming people. The digits become individual people. I love the colors on the figures in this one. Nice clothes too. In general, the figures are great.

4.5/5

------------------------------------------------------------
Generated Image 7
------------------------------------------------------------
Vislab-MAT2_hands_first_person_angels_playing_counter_strike_fi_b3196abd-1317-4a20-854d-bd9dc824b753.png
Discussion:
An example of hands becoming people. The digits have clearly separated and have come triplets. The lighting works well in this one.

4/5

------------------------------------------------------------
Generated Image 8
------------------------------------------------------------
Vislab-MAT2_hands_first_person_angels_playing_counter_strike_fi_a9f20798-dc75-4c3d-9d55-a115687d3d30.png
Discussion:
I liked one's focusing on odd composition of hands. I still am looking for stranger first person perspectives.

4.5/5

jiarui_zhu
Posts: 7
Joined: Mon Sep 26, 2022 10:25 am

Re: wk1 - MidJourney 1 - 1st Results

Post by jiarui_zhu » Wed Oct 05, 2022 1:44 pm

Text query: starry sky viewed on the top of a volcano, photorealistic
Snip20221005_11.png
1. Does the image meet your expectations?
The image does not meet my expectations. I want the image to be photorealistic but the image result it generated is more of a painting style. The stars in the sky are blurry and are not evenly distributed. Also, the red pixels on the volcano, I assume they are lava, are not realistic.

2. To what degree does your text query influence the generated image?
In terms of the content, I think the text query is very successful in generating the image. I want a starry key on top of a volcano, and the resulting image shows exactly the same. However, the style I specified in the text has little influence.

3. What is the style of the image, and why do you think it has produced that?
I do not have a background in art so I can not give an accurate style. I think the style is more like a painting, not a photo, yet it still shows a fair amount of details.

4. Any thoughts about how the visual elements in the image are organized
The starry sky is on the top and the mountains are located in the right bottom of the image. The image is very well balanced, but the stars are not evenly distributed.

5. How would you change the query? to achieve what difference?
I would add “volcano” with lava eruption to add lava in the image to add some dynamic elements.

6. Any other comments?
I feel like it’s very hard to generate photorealistic images that add natural scenes together. At least I haven’t found a good way to do so in Mid Journey.

7. On a scale of 5 - from 5 being GREAT to 1 being LOW your rating of the result
3/5



Text query: starry night with galaxy that stretches to infinity, lava erupts out of a volcano, photorealistic
img2.png
1. Does the image meet your expectations?
The image meets most of my expectations. It contains all the elements I want. Although not very realistic, the image still looks magnificent. My only complaint is that the lava eruption gets blurred with the stars.

2. To what degree does your text query influence the generated image?
The text query heavily influences the generated image. All the elements specified in the text appear in the image.

3. What is the style of the image, and why do you think it has produced that?
Sadly, the style of the image still looks like it’s a piece of painting, not a photo, despite the fact that I specified “photorealistic” in the text.

4. Any thoughts about how the visual elements in the image are organized
The starry sky is located on the top of the image with a huge volcano in the bottom. The lava erupts to the sky and gets blurred with the galaxy that stretches to infinity. There is also lava that flows down the mountain. The elements associated with the volcano look very natural.

5. How would you change the query? to achieve what difference?
The moon is too bright, and it breaks the vibe of the image. I would add the query to remove the moon.

6. Any other comments?
Pretty happy with this generated image. Would print it out and hang it in my bedroom

7. On a scale of 5 - from 5 being GREAT to 1 being LOW your rating of the result
4/5



Text query: starry sky viewed on the top of a volcano, ocean, photorealistic
img2.png
1. Does the image meet your expectations?
The image meets 60% of my expectations. It contains all the elements I want, but the starry sky somehow is too bright.

2. To what degree does your text query influence the generated image?
The text query heavily influences the generated image. All the elements specified in the text appear in the image, but the style is not.

3. What is the style of the image, and why do you think it has produced that?
Again, Sadly, the style of the image still looks like it’s a piece of painting, not a photo, despite the fact that I specified “photorealistic” in the text.

4. Any thoughts about how the visual elements in the image are organized
The image looks like it’s being captured from a distance. I like the way the sky, volcano, and ocean are organized in the image.

5. How would you change the query? to achieve what difference?
I would add “a starry night with thousands of stars” to add more stars to the sky.

6. Any other comments?

7. On a scale of 5 - from 5 being GREAT to 1 being LOW your rating of the result
3.5/5



Text query: a man in red jacket skiing down the mountain in high speed, there are pine trees, photorealistic
skiier.png
1. Does the image meet your expectations?
Mostly. The image does capture the dynamic motion of skiing at high speed, and it contains all the elements I want. However, it’s not super photorealistic, and the ski seems missing. The motion and body position of the skier are not very natural.

2. To what degree does your text query influence the generated image?
Very. The generated image is a realization of the text query.

3. What is the style of the image, and why do you think it has produced that?
The style of the image is like a painting, not a photo. Have no idea why it has produced that.

4. Any thoughts about how the visual elements in the image are organized
This image looks like it is being captured from the back of the skier. The pine trees lie symmetrically on the sides. I like the way the elements are organized.

5. How would you change the query? to achieve what difference?
I would add “steep slope” to the query. The slope in the image looks easy.

6. Any other comments?
There are some other red pixels in the image other than the jacket. It makes the image less photorealistic.

7. On a scale of 5 - from 5 being GREAT to 1 being LOW your rating of the result
4/5

tinghaozhou
Posts: 6
Joined: Mon Sep 26, 2022 10:24 am

Re: wk1 - MidJourney 1 - 1st Results

Post by tinghaozhou » Wed Oct 05, 2022 9:29 pm

The core question I ponder for this class is: how does machine-generated images speculate our planetary futures? My weekly assignments will be centralized around producing images with speculative-fictional literatures as thought experiments on imagining different forms of futures. With that, I hope to consider the relationship between machine-generated images and (not just) the world yet to come (but also) the world of the past tense and the world we inhabit on. Essentially, I'm wondering: is a text-image generator essentially a prediction machine or a memory storage/fabricator?

I start with my experiment with Afrofuturism speculative fiction writer Nnedi Okorafor's short piece, "Mother of Invention." What comes to mind when we think about radical futures of media? Perhaps something metallic, computational, or related to infrastructure, intensified forms of surveillance, variations on portable devices, smart clothing, or other externalizations of artificial intelligence. What makes Okorafor's piece interesting is that she follows a different line of questioning that de-privileges the computational as the only site of future mediation, seeing it as one among many informatic forms that will play a role in shaping media futures. The particular future of media she explores in the short piece has to do with pollen and a grass-saturated and grass-dominated post-oil world. Her short story not only touches upon environmental design and technical invention, but also intellectual property and future forms of colonialism, weather and the anthropocene, sustainable development and motherhood. Pollen here is an informatic form, a carrier of genetic information between plants, carried erratically by the wind. Set in the near future city of New Delta, Nigeria, the story speculates about a form of botanical media future in which we read the relationship/interactions between a smart home Obi-3 and an expectant mother, Anwuli, who's just about to give birth though she has terrible pollen allergies.

This week, I try to use the Midjourney bot to "imagine" the city of botanical futures--the New Delta, Nigeria. The first time Okorafor gives out a detailed description of the city comes with pictorial imaginations: "In the area between New Delta’s low skyscrapers, buildings and homes were carpeted with its world-famous stunning green grass, and the roads were fringed with it, but in this scene the grass was covered with smiley-faced bopping periwinkle flowers. It looked ridiculous, like one of those ancient animations from the early 1900s or a psychedelic drug–induced hallucination."

I started my first prompt with a sentence from this paragraph:
"In the area between New Delta’s low skyscrapers, buildings and homes were carpeted with its world-famous stunning green grass, and the roads were fringed with it, but in this scene the grass was covered with smiley-faced bopping periwinkle flowers."
Vislab-MAT1_In_the_area_between_New_Deltas_low_skyscrapers_buil_12e8777f-bdb8-44c6-939a-2824cb6c4f82.png
The result came back with a not-so-futuristic interpretation, manifested as a contemporary metropolitan city with larger plant coverage. Of course, when reading the prompt, it's clear to us that the description itself is not "instructive" or "direct" enough as there is no sign of particular space and no signal of futurity in it. Meanwhile, I realized it was hard for the bot to fully recognize long-form sentences. Meanwhile, I deliberately left out the second part of the paragraph that actually gives out a kind of "stylistic" description of the imagined city.

Therefore, my second prompt was elaborated not in the form of a sentence but "a cluster of keywords." I also add specific city name and Okorafor's "stylistic" description of the place so that the prompt can be more instructive:
"Post-oil Nigeria, low skyscrapers, buildings and homes carpeted with greenest grass and smiley-faced bopping periwinkle flowers, a psychedelic drug–induced hallucination."
Vislab-MAT1_Post-oil_Nigeria_low_skyscrapers_buildings_and_home_32befdd6-8e95-4183-8c7a-e8ecf5829da0.png
This time, the result came back slightly different in terms of the depicted objects and its style (but not very much)--there are trees that indicate that Nigeria-ness of the image but it doesn't feel very much "futuristic" to me. For me, the stylistic description of "a psychedelic drug-induced hallucination" is not a "strong" stylistic indicator because it doesn't locate a style but only a kind of impression. But it's interesting to look at how the machine understand hallucination and a sort of contemporary impressionist style.

I replaced the "hallucination" with another stylistic description in the paragraph in the third prompt:
"Post-oil Nigeria, low skyscrapers, buildings and homes carpeted with greenest grass and smiley-faced bopping periwinkle flowers, ancient animations from the early 1900s."
Vislab-MAT1_Post-oil_Nigeria_low_skyscrapers_buildings_and_home_1e92a1e0-529c-47d3-b145-7cb21af109e2.png
Apparently, the "ancient animations" part of the prompt is a "strong" stylistic indicator in comparison with the "hallucination" of the second prompt so the result came out with a clear stylistic inclination. With the guidance of animation style, the image also came back with a stronger futuristic form as we can see the skyscrapers built in a traditional Nigerian-style house--a very organic, woody, and porous structure.

I only modified a little bit of the third prompt for my forth prompt--I took out the "ancient" in the stylistic description:
"Post-oil Nigeria, low skyscrapers, buildings and homes carpeted with greenest grass and smiley-faced bopping periwinkle flowers, animations from the early 1900s."
Vislab-MAT1_Post-oil_Nigeria_low_skyscrapers_buildings_and_home_f9c89349-bc86-421a-b02d-5450ea9c73f0.png
Immediately most of the Nigerian-inspired skyscrapers disappeared from the images. What replaced them were a collection of both the modern architectures in contemporary African cities and sketches of traditional African villages. I'm wondering how the machine conceive "ancient" in the prompt.

The final two prompts deal with the issue of "ancient" in the stylistic description but I also replaced the "post-oil Nigeria" with "afrofuturism" aiming to give a stronger stylistic instruction to the bot.
"afrofuturism, low skyscrapers, buildings carpeted with grass and smiley-faced flowers, ancient animations from the early 1900s."
Vislab-MAT1_afrofuturism_low_skyscrapers_buildings_carpeted_wit_1e5abf28-9fa9-4e86-b91d-c3235e4d222d.png
"afrofuturism, low skyscrapers, buildings carpeted with grass and smiley-faced flowers, animations from the early 1900s."
Vislab-MAT1_afrofuturism_low_skyscrapers_buildings_carpeted_wit_8b4414b5-bd2c-4881-b3fc-7dc158fec779.png


BTW, an interesting article on e-flux about DALL-E: https://www.e-flux.com/notes/495428/is-dall-e-a-genius

merttoka
Posts: 21
Joined: Wed Jan 11, 2017 10:42 am

Re: wk1 - MidJourney 1 - 1st Results

Post by merttoka » Mon Oct 10, 2022 1:03 pm

----------------------
You can see these explorations on a Miro board (Week 1) at this link: https://miro.com/app/board/uXjVPQgjXkQ= ... 6714658149
----------------------

As a starting point, I wanted to see how Midjourney prompt parameters affect the output of an image. This website was influential on the parameters I picked. I tried different values of keyword weights, chaos, and stylize commands with a rudimentary prompt: symmetry asymmetry. This prompt was motivated by an exploration revealing how Midjourney perceives higher-level compositional concepts instead of a categorical representation of image subjects.

I also used the same seed parameter for every image generation to keep the image's content as similar as possible and only observe the effect of the parameters on the content. I arbitrarily picked the --seed 4776 in this experiment.

In addition to these parameters, the default image generated with this prompt included many elements with the blue sky, making the image a landscape portrait regardless of the parameters. I wanted to receive more abstract images, so I added --no sky element to suppress the generation of blue sky. This parameter kept the images mostly indoors and/or abstract. And lastly, I added --ar 16:9 for all images to receive non-square images.

As a reference, here is the ground-truth image generated with the basic version of the prompt:
symmetry asymmetry --no sky --ar 16:9 --seed 4776
image.png

Keyword Weights
Midjourney allows modifying the weight of each keyword or phrase with a special syntax (word::#). The range of these weights is arbitrary and can be any number. The only limitation seems to be the sum of all keyword weights needs to be a positive number (that is why I couldn't generate 0,0 image).

First, I wanted to see the effect of modulating keyword weights for each term in my prompt. The result of the following prompt can be viewed as a 2D plot:
symmetry::x asymmetry::y --no sky --ar 16:9 --seed 4776
matrix.png
There are a few things I noticed in this 2D space.
  1. As the weight of the symmetry parameter increases, the image becomes more colorful. This is something unexpected to me.
  2. The standard content of these images is (1) some sort of a human shape (either a face, body, or parts of them), (2) butterflies, (3) crystal-looking structures, and (4) gallery spaces. This is not surprising and potentially reflects the dataset that Midjourney is trained on.
  3. It seems like the symmetry operations are working based on the center of the image. In the bottom right, all images are somewhat symmetric from the center of the picture.
Chaos
Another parameter investigation was about the chaos parameter that ranges from 0 to 100. This parameter adds more noise to the initial noise image from which the variations are generated. A more noisy initial image correlates to a more detailed final result. I picked two images from the previous 2D space and ran different chaos values on them.

symmetry::0 asymmetry::1 --no sky --ar 16:9 --seed 4776 --chaos x
chaos_01.png
symmetry::1 asymmetry::0 --no sky --ar 16:9 --seed 4776 --chaos x
chaos_11.png
In the higher ranges, this parameter seems to keep the image's content and morphs it into something more complex. In the first image above, the pictures in V3 iteratively get more complicated with the introduction of more elements, and when we assign chaos as 100, the text starts showing up. This might be due to the high-frequency nature of fonts.

Stylize

One last parameter tested is the stylize parameter. This parameter seems to be adding more and more content inside the image composition as its value increases.
symmetry::1 asymmetry::1 --no sky --ar 16:9 --seed 4776 --stylize x
stylize.png
One interesting thing was the resurgence of the sky as the primary element in the higher ranges of this parameter. Even though I still have no sky element in the prompt, it doesn't seem to make a difference.

yixuanli539
Posts: 1
Joined: Mon Oct 24, 2022 4:16 pm

Re: wk1 - MidJourney 1 - 1st Results

Post by yixuanli539 » Mon Oct 24, 2022 4:42 pm

This is Maria's work:

This was the prompt for the bird image: Lord, how unutterably disgusting life is! What dirty tricks it plays us, one moment free; the next, this. Here we are among the breadcrumbs and the stained napkins again. That knife is already congealing with grease. Disorder, sordidity and corruption surrounds us. We have been taking into our mouths the bodies of dead birds. (It is a Virginia Wolfe quote.)

And here is a link to the image I spoke about in class:
1.png
The upscaled version is here:
2.png
And here are links to the three other images midjourney produced:

1.
3.png
2.
4.png
3.
5.png
I also found this prompt yielded interesting results: google maps on the surface of the moon (I was impressed by the overlap of the moon 'texture' and the map layout)

Link to image:
6.png

Post Reply