images-discussions

[late blade]

Morning everyone

[empty kelp]

If it weren't for OpenAI or DALL-E i would have probably never even thought of creating elf or sock images. Many of you seem to be experts on elves, so i'll let you evaluate them

[late blade]

Did Dall-E woke up with an injection of "Content Policy" to vaccinate users?

@empty kelp I got elves engineering socks!!

make them go to a disco party and still wear socks

[empty kelp]

These elves are all wearing appropriate beach attire, and only drinking non-alcoholic tropical drinks. This is a diverse group of real elves, and Santa did not pass out in the second image -- he's been working very hard preparing for Christmas, there was lots of wrapping and preparation, and he fell asleep

[fallow quail]

i think it works better with the illustartion style from before. this gives me the vibe its out of an vintage ad from the lat 50s or early 60s. really dig it compared to the bland renders.

[empty kelp]

Most people don't realize this, but most elves are very athletic. Elves have to move heavy boxes of presents around all day, and it's a lot like being a dock worker for a shipping company.

They have to drop presents off in Hawaii anyway, so Santa likes to take his employees to the beach as much as possible. It's freezing at the North Pole, and they need to defrost -- so Hawaii is the obvious place

[late blade]

but elves use magic also, so using magic doesn't make you an athlete

[empty kelp]

Yes, but magic is used primarily for offensive and defensive purposes. It's not typically used for moving packages unless you're like Luke Skywalker or something

[late blade]

I dunno, that didn't seem like that in the harry potter movies. magic was pretty much used for everything, from stirring spoons on cups to moving whole areas.

[empty kelp]

That's the stereotypical view of elves. The truth is that they're very athletic, and diverse, and they work hard so Santa likes to take them to the beach when they're dropping off packages in Hawaii

[fallow quail]

i like the hawaiian santa.

[empty kelp]

I told DALL-E to have Santa in appropriate beach attire, but it puts him in his work suit even if he's on the beach. It could be that the AI needs some adjustment

[late blade]

you are excluding magic using elves!! they also can work and not be athleetic!

I'm gonna complain to openai that we need more elf diversity in the content policy

[empty kelp]

you're talking about the office elves. they usually stay at the North Pole headquarters to update the accounting and paperwork

they usually take time off and visit Tahiti during the Spring with Mrs. Claus

[fallow quail]

use overweight, sometimes chubby gets blocked and gpt hallucinates some other excuse

[late blade]

roger

an none of them are wearing a speedo, so it's ok

[fallow quail]

the females seem to be fitter

[late blade]

yeah, but I'm not doing another image of that sort, I got my own battle with content and directives atm

[fallow quail]

in my experience dalle3 is biased towards more "normal weight" women when it comes to prompts with overweight or sleightly overweight compared to male. sometimes its way exaggerated. this was a "slightly overweight woman with a thoughtful expression"...

[late blade]

I do see that in the future the content policy has to be hugely adapted. Otherwise the content policy itself will cause too much bias in the weight of how chatgpt works.

even in other fields.

lots of controversial stuff that shouldn't be decided by content policy, that in itself is already controversial

[sullen heath]

Oopsie 😬

[grizzled loom]

@mild patio

[wispy storm]

Try new chat.
Pre prompt for realist style 8k resolution, detailed. Ask ok if understand.

After ask again the scene

[sullen heath]

I'm not sure if you understand but when I asked it to make it funny the characters suddenly turned black...

[wispy storm]

No I didn't Understand

[green pebble]

Anyone able to show them generating super fluid, dynamic, high movement, or actionable images?

For example like this:

[chilly onyx]

Likely just coincidence, although they do look a bit problematic. I do think that the second set of pictures has a more hyperreal depiction of a desk in use than the first set though; those first two desks have too much space. I can barely see any desk surface as I look at my own whilst typing this response…

[sullen heath]

I did a thumbs down on the response and provided feedback, unsure if it'll actual get looked at though 🤷‍♂️

[vapid elk]

it does

it is very likely that good feedback will be used to fine tune the model

I have not noticed it, but normal model can only make one image per request, but it might request twice

dalle custom GPT will make 2 images in one requets

but it looks like it takes twice the time

[grizzled loom]

it also appears to have made those a match set

be a nice decorative set for opposing sides of a headboard in a bedroom

[vapid elk]

this is unlikely

same prompt, different seeds

[grizzled loom]

ive had tons of intesting features in images

[vapid elk]

just happens to look kinda conencted side by side

but it was for sure just a chance

the AI works by making a full tile, so if it where generated together, it would be a perfectly matching set

inpainting could probably connect those two

[grizzled loom]

itll do tons of things. you can get animation cells as well if done well they have lined up. i overwrote those in prompting so i cant produce those, but im sure you get the idea @vapid elk

[green pebble]

The one in space is really good!

[grizzled loom]

it is. on point for print comic

[hot rain]

Thank you!

[green pebble]

That's what I was thinking

[hot rain]

Also kinda fits your specification

[vapid elk]

this looks incredbly good

[hot rain]

Thank you!

[grizzled loom]

right, and a matched set can be worked also into an animation sequence, then into video. i worked on a mutitude of strategies and have many i feel could work for a few systems with some more work

[green pebble]

Yes, I can feel the movement in this one too. Would you be willing to share the prompt for the space one? I'm interested to see how that comes out in my bot

[hot rain]

"Make an image of a being with a green aura charging an attack in orbit of a planet, surrounded by debris from destroyed starships. In the style of a graphic novel. The being is very far away, only a spec in their glowing green aura"

[vapid elk]

yea, video AI is a whole new set of problems

[grizzled loom]

thats my current focus

[vapid elk]

cool

[green pebble]

I haven't personally seen a video AI that does it really really well yet

[grizzled loom]

preciesly, tehy all are 'eh'

[vapid elk]

AI video still not really good

[dim cradle]

You made it look easy, I know it's not. I get close but no cigar.

[green pebble]

Until I can create a full blown anime with one, it's not good enough lol

[green pebble]

VERY full of movement

[hot rain]

An AI upscaler would do that image justice

[vapid elk]

this looks really nice, I think dalle took some artistic liberty on the caustics xD

first image's prompt was a sakura blossom petal encased in a glass cube, the glass looks like it has subtle purple storm like lighting around the petal

[grizzled loom]

for all of you, take note that you can specify a wide range of things INCLUDING the specifications for the camera optic FOV etc, also "orthographic" or "perspective" etc etc

make it a 45 degree field of view blah blah

[green pebble]

That's my pixel art bot, converting it to my realism bot now with my process to see the results

[grizzled loom]

that will help @dim cradle

[dim cradle]

ty for sharing, i appreciate it. i think maybe i was over-specifying:

Generate a photo featuring a sock encased in a solid acrylic cube. Implement a captivating caustics effect by perfectly positioning a radiant purple light, casting striking patterns and reflections within the transparent acrylic. This lighting setup should enhance the surreal atmosphere, making it seem as though the socks are suspended in an enchanting, dreamlike state within the cube, with the light playing off the socks and creating an even more mesmerizing visual display.

[hot rain]

Had to try too

[grizzled loom]

try one with the word orthograpic view

[vapid elk]

yea, some of the best images in dalle come from short prompts and luck

[grizzled loom]

square stuff is really fun in ortho

I havent found an art or film technique that i cant use.

[hot rain]

I made the prompt while trying to generate images that look vaguely like Perfect Cell from DBZ for a dumb Youtube series I'm making lol.

[green pebble]

Love it lol

[vapid elk]

over-specifying will often limit the ""creativity"" of the AI by causing it to embed more information

[green pebble]

True, but if you only specify on technique and not visual it usually helps a lot

[hot rain]

"Perfect Cell vs Cadia" if you're curious lol.

[grizzled loom]

it does prompt reduction. anything you say gpt trims to approx 100 words(vague)

[vapid elk]

btw, I used the API, not ChatGPT, so, this was the actual prompt that the image generation got

[grizzled loom]

so its critical you use words you want included or specify carefully all elements i na format list of your design

Did you try an ortho one @dim cradle ?

[dim cradle]

i'm about to try it out but with a sock 🙂

[vapid elk]

it might not work

[hot rain]

Getting closer to one I'm happy with.

[vapid elk]

or.. it might.. =P

[dim cradle]

not yet but it's on the list man

[grizzled loom]

so next tid bit of artist luv is you can set camera position angle, declination, blah blah

[vapid elk]

the thing is that sakura blossom petal might stear the embedding to avery different direction that sock would, which has the potential to completely change the """"vibe"""" of the image

sakura blossom petal might embed it to a more "whimsical" direction... oriental culture, mistic.. and so on..

while sock.. probably not

[grizzled loom]

jsut make a list item "Vibe:"

[vapid elk]

AI vibe check

[hot rain]

"Make an image of a lotus flower trapped in a crystalline glass cube, in the style of RAW photography. A magenta light is under the prism." is my prompt for these.

[green pebble]

Here is the prompt I tried to test the theory: Please create an image of a lapis blue dragon flying in at a 45 degree angel toward another dragon that is white. Use a two point perspective and have the blue dragon flying twoard the camera, while the white dragon is flying away from the camera toward the blue dragon. Give them both ferocious appearance and heft as they are more in the style of game of thrones dragons then traditional flying serpents. Let's have the camera be a 1080p resolution and Aperture: f/1.8-f/5.6. Use AF-C focus to help convey the movement in the scene and ISO 6400 to ensure its the most advanced image

[dim cradle]

gpt-3.5 output this transformation:

a sakura blossom petal encased in a glass cube, the glass looks like it has subtle purple storm like lighting around the petal
=>
A sock encased in a glass cube, the glass appears to have subtle purple storm-like lighting around the sock.

I will try that with the api unless you have other input, thanks

(not sure if vibrant -v- natural style setting makes a difference)

[green pebble]

It did lose some creativity and didn't capture the angel right

[hot rain]

I think I'm willing to settle on this guy

[green pebble]

Seeing if I can tweak it with the seed to try and alter what Its being asked

[dim cradle]

(i'm pretty sure i want to stick with vibrant, natural can be a tad underwhelming)

[green pebble]

Hmm nope, it took it way different direction lol

[hot rain]

Here's a fun one I made a few weeks ago, a magical academy in the 5e campaign my friends and I are playing.

[grizzled loom]

"1. Scene Description:

• You are in a dark room.
• The centerpiece of the room is a crystalline glass cube.

2. Object Description:

   • Inside the cube, there's a perfectly preserved lotus flower.
   • The lotus flower is bathed in a soft, ethereal light.

3. Cinematic Settings:

   • Lighting Style: The room is illuminated with a subtle, otherworldly glow.
   • Colors: The predominant colors are shades of deep blue and purple.
   • Vibe of the Scene: The atmosphere is mysterious and serene.
   • Style: The scene should be captured in the style of RAW photography, focusing on fine details and textures.
   • Emotional Tone: The overall mood is one of tranquility and wonder.

4. Camera Settings:

   • Field of View (FOV): Use a wide-angle lens to capture the entire room.
   • Camera Position: The camera is positioned slightly above the glass cube, looking down at the lotus flower.
   • Lens Settings: Use a high-quality lens with a wide aperture to achieve a shallow depth of field.

5. Lighting:

   • A magenta light source is positioned beneath the prism.
   • The magenta light gently illuminates the lotus flower, creating captivating reflections and refractions in the glass.

6. Additional Notes:

   • Pay close attention to the interplay of light and glass, highlighting the intricate details of the lotus flower.
   • Emphasize the contrast between the dark surroundings and the radiant flower.

Please generate an image that captures this scene with meticulous attention to detail and cinematic aesthetics.
"

[grizzled loom]

this will up all your games. think subjugated lists

[green pebble]

Have you tried this with non realistic styles?

[grizzled loom]

any and all man, that's dynamic. adapt it for yourself, add list fields, remove them. it doesn't care.

[grizzled loom]

the output

[green pebble]

I'll give it a shot

[late blade]

from my text generation tool, I have something similar, I've been tweaking types of lenses, types of light, type of art, type of textures

[vapid elk]

loved it

[vapid elk]

vibrant for sure, natural will be terrible for it

[dim cradle]

very good. yeah, there are a number of considerations to this composite--Lugui mentioned one: "Caustics, in the context of an object suspended in an acrylic cube, refer to the patterns of light and shadow that result from the refraction and dispersion of light as it passes through the acrylic and interacts with the object, creating interesting and often visually appealing optical effects."

And I did some homework to cover all the bases:

When an object is suspended in an acrylic cube, several natural optical effects can be observed due to the interaction of light with the object and the acrylic material:

Refraction: The bending of light as it passes from one medium (air) into another (acrylic) can distort the appearance of the object and create the illusion of it being displaced or magnified.
Dispersion: The acrylic's refractive properties can cause the separation of light into its component colors, creating a prismatic effect with a spectrum of colors around the edges of the object.
Glare and Highlights: The acrylic's smooth surfaces can cause light to scatter and produce glare or highlights, enhancing the object's visibility and creating a sense of depth and dimension.
Shadows and Silhouettes: Depending on the object's shape and the lighting conditions, distinct shadows or silhouettes can be cast on the inner surfaces of the acrylic cube, adding to the visual complexity of the scene.
Internal Reflections: Multiple internal reflections within the acrylic can create intriguing mirror-like images of the object, leading to the appearance of additional virtual objects within the cube.
Color Shifts: Depending on the acrylic's composition and any additives, it may introduce color shifts or tints to the object, altering its natural appearance.
Distortions: The refractive properties of acrylic can distort the shape and size of the object, making it appear stretched, compressed, or otherwise transformed.
Optical Illusions: The combination of these effects can lead to optical illusions, making it challenging for viewers to accurately judge the object's size, position, or characteristics.

These effects can result in visually captivating and unique displays when objects are suspended within acrylic cubes, making them popular for artistic and decorative purposes.

Ultimately builds up to the optical illusion ...

[late blade]

I've been working on a python prompt constructor, that way i don't have to repeat what I have already to my liking

[grizzled loom]

input: 1. Scene Description:

• You are in a dark room with multicolor PBR mist on the floor
• The centerpiece of the room is a crystalline glass cube

2. Object Description:

   • Inside the cube, there's a perfect Christmas Stocking.
   • The Christmas Stocking is bathed in a soft, multicolor, ethereal light and surrounded by a soft vortex of PBR magical sparkles

3. Cinematic Settings:

   • Lighting Style: The room is illuminated with a subtle PBR overall, adding otherworldly glows.
   • Colors: The predominant colors are shades of christmas colors.
   • Vibe of the Scene: The atmosphere is mysterious and serene.
   • Style: The scene should be captured in the style of RAW photography, focusing on fine details and textures.
   • Emotional Tone: The overall mood is one of tranquility and wonder.

4. Camera Settings:

   • Field of View (FOV): Use a anamorphic lens to capture the entire room.
   • Camera Position: The camera is positioned 20 degrees above the glass cube, looking down at the Christmas stocking.
   • Lens Settings: Use a high-quality HD lens with a wide aperture to achieve a good field.

5. Lighting:

   • A christmas light colored source is positioned beneath the prism.
   • The light gently illuminates the Christmas stocking, creating captivating reflections and refractions in the glass.

6. Additional Notes:

   • Pay close attention to the interplay of light and glass, highlighting the intricate details of the Christmas stocking.
   • Emphasize the contrast between the dark surroundings and the radiant Christmas stocking.

Please generate an image that captures this scene with meticulous attention to detail and cinematic aesthetics.

[grizzled loom]

output

[dim cradle]

these are all great, y'all made them so quickly also, you're good

[grizzled loom]

you guys see the changes? litterally any list you want allows for finer dalle sends

[dim cradle]

awesome lighting effects

[grizzled loom]

i left all propmts in place for learning and looking at what is passed.

hopefully you guys that didnt think this way had your "ah ha" moments now

[green pebble]

Interesting

[grizzled loom]

great. what di you send for the driving prompt

[green pebble]

Make an image of a lotus flower trapped in a crystalline glass cube, A magenta light is under the prism.

But I started in my pixel art bot and then had it converted in my realism bot

[grizzled loom]

try something like this @green pebble

[green pebble]

I did and I didn't care much for the output

[grizzled loom]

beutiful part is if you do it dynamically like i showed above you can save those for adaptation at a later time

[green pebble]

I adjust it to be a stag

It worked, but it idk, didn't feel right maybe needed a lot more adjustment

[grizzled loom]

you have to tailor it to your desire. lmao

formatting is the key thing

[dim cradle]

this reminds me of old sierra online games, looks like a scene out of hero's quest/for glory

[grizzled loom]

having a format not just random sentences, they can be good to get an idea, but to expand a lok or style you want a reliable method

[green pebble]

See tailoring these bots is very tricky. That's why I like having detailed custom instructions so that the style consistently comes out the way I want it and I can have natural language used inside the prompt area

Like this is creative and cool as heck imo

[vapid elk]

this was the prompt you gave to chatgpt, but it rewrote it, right?

[grizzled loom]

Scene Description:
Setting: A dimly lit room, characterized by bold outlines and dramatic shading, common in comic book art.
Centerpiece: A prominently featured crystalline glass cube, stylized with high-contrast lighting to enhance its edges and facets.

Object Description:
    Inside the Cube: A Christmas stocking, depicted in vibrant, comic book-style colors.
    Lighting and Effects: The stocking is illuminated by a soft, multicolored light, rendered in a cel-shading technique typical of graphic novels. It's surrounded by a vortex of magical sparkles, stylized in a dynamic, exaggerated manner.

Cinematic Settings:
    Lighting Style: Muted PBR (Physically Based Rendering) illumination, adapted for a comic book aesthetic with bold contrasts.
    Colors: A palette dominated by traditional Christmas colors, but with the saturated, vivid tones characteristic of comics.
    Atmosphere: Mysterious and serene, conveyed through a stylized art approach with dramatic shadows and highlights.
    Style: Emulating the visual narrative style of graphic novels, focusing on stark details and textured patterns.
    Emotional Tone: A feeling of tranquility and wonder, achieved through expressive linework and color choices.

Camera Settings:
    Field of View: Use a perspective typical of graphic novels, perhaps with a slight fish-eye effect for dramatic impact.
    Camera Position: Positioned 20 degrees above the glass cube, framing the scene like a dynamic comic book panel.
    Lens Settings: Mimicking a high-quality HD lens effect, but with sharp linework and clear definition typical of comic art.

Lighting:
    Below the Cube: A source of light resembling Christmas colors, designed in a stylized, graphic way to create dramati(trunc for post)

[green pebble]

Yes it rewrote it to this:

A highly detailed pixel art scene in a 16:9 aspect ratio, capturing a dynamic encounter between two dragons in the style of Game of Thrones, rather than traditional flying serpents. The first dragon, lapis blue in color, is flying towards the camera at a 45-degree angle using a two-point perspective. The second dragon, white in color, is flying away from the camera towards the blue dragon. Both dragons have a ferocious appearance and substantial heft, reflecting their formidable nature. The scene is created to mimic the effect of a camera with 1080p resolution and an aperture range of f/1.8-f/5.6, using AF-C focus to convey movement and an ISO of 6400 for advanced image quality. The positioning and perspective create a sense of imminent collision and intense action between the two dragons, with the background and lighting enhancing the dramatic atmosphere. Techniques such as dithering, outlining, anti-aliasing, color palette limitation, manual pixel placement, layering, shading, emotion, composition, mood, stylization, texture, perspective, foreshortening, balance and weight, rhythm and movement, narrative, and use of space are used to create a vivid, detailed artwork.

[grizzled loom]

THAT is agood image prompt. well done

[green pebble]

That's thanks to my custom instructions

[dim cradle]

yeah it's like a spec sheet you can transform into various formats depending on the use case, appears to be a good workflow.

[dim cradle]

this is going to be a project for today

[green pebble]

See i take out the field of view and the camera parts of the prompt and get this

[shut niche]

Haha, screw it, here's a dragon! 🐉 🐲 🀄

[green pebble]

So I think the additive of the extra little bits can offset the level of creativity and detail in certain cases and I think techniques of artstyles has more weight on the image making

[shut niche]

Tell it to "stop and smell the roses" Not kidding.

[dim cradle]

hmm... learned some things, got some good ideas from this recent convo, thanks

stopping to smell the roses is what i've been doing lately as well. 🙂

[green pebble]

I think the longer I am here the more I see we are all different with our technique and artstyle. Even though we can produce similar images, I think we have our own flair which really makes everything stand out. It's great to see what yall make though, helps push the boundries of the bots capabilities for each of us

[grizzled loom]

here guys: Scene Blueprint:

Theme & Style: [Input Theme], [Art Style]
Setting & Ratio: [Scene Setting], [Aspect Ratio]

Subjects:

#1: [Identity, Color, Action]
#2: [Identity, Color, Action]

Characteristics & Dynamics:

[Traits of Subjects], [Scene Dynamics]

Technical Specs:

Camera: [Resolution, Aperture], Focus: [Type], ISO: [Value]

Cinematic & Artistic Elements:

Lighting & Palette: [Style, Colors]
Perspective & Movement: [Type, Style]
Techniques: [Specific to Art Style], [General Techniques]

Advanced Settings:

Environmental Effects: [Input Effects]
Interaction Dynamics: [Input Interaction Details]
Background Elements: [Input Background]
Additional Props: [Input Props]

THAT is a dynamic system. use 3.5 to fill it out if you want extra lazy mode enabled

this blueprint will optimize Dalle prompts if you dont get overly wordy outside of the descriptives

[shut niche]

Last night I was working on it's understanding of the human condition.

[green pebble]

This is pretty good

[dim cradle]

very good points. good reminder also much is subjective.

[grizzled loom]

many many ways to get what you want

other fun ones are asking dalle to percieve itself or portray itself going through experiences. aske dalle, "a picture of you(dalle) being invoked"

or evoked, or in a eppic rap battle

seriously crazy

[mild patio]

why would it identify itself as a human

[grizzled loom]

thats the point

[mild patio]

wonderful, some dude came with the idea to feed it data, to make it think its human....

[dim cradle]

just daydreaming about a gui app, based on the specs, powered by the api .....

[grizzled loom]

do it. very easy

[green pebble]

Ok I see where I can make adjustments to the instructions. Certain key words will mix up the prompt it ends up writting. So I need to research how to fix that other then adjusting the prompt myself with a second image. Because doing that manual process after I got this output

[late blade]

absolutely, and that makes it even more interesting.

[green pebble]

So it's doable to have multiple well done creatures in the same scene, but it takes percission, so I will have to figure out what those tweaks are

[mild patio]

this is actually good

[grizzled loom]

good work acid. very nice

[mild patio]

there was some individuality to that picture, and thats good. she was not perfect, but rather unique the way she is

[grizzled loom]

jsut did dalle invoking itself

[dim cradle]

we shoud talk. months ago i developed one in python using pyqt for the api -- being dall-e 2 and all, i first passed the input prompt to the completion api (wasn't deprecated then) to first augment the prompt, before calling the images api -- this was a few months before chatgpt started doing that for us -- but with the new frameworks and dall-e 3, i am thinking about revisiting.

[shut niche]

This is from Bing.

[grizzled loom]

i have many tools

[shut niche]

Here's the other 3 from the gen.

[grizzled loom]

honestly im working on full motion video with accuracy. no 4 second models etc, goaling for frame by frame full frame animation output. ive been down many roads but ultimately have found some potentially viable paths to explore

[green pebble]

Self developing? Or by using apis with other frameworks?

[dim cradle]

From earlier this year, please disregard the primitive GUI, it was an early POC, but functional. the "boost" button augmented the input from the user to generate an art prompt, then that would be used to generate the gallery

[grizzled loom]

studying a few neural architectures to see if i can avoid a novel one and having to find funds for training; i believe i have a method, but proof is in the pudding

[green pebble]

If I can assist in this, I'm interested.

[shut niche]

Just give it token treats when it's good. That'll teach it.

[late blade]

for some reason the unicorn got the dragon wings

[shut niche]

Unnoticed and forgotten...

[green pebble]

I love how this came out

[dim cradle]

Nice. Time-lapse photography?

[late blade]

now this is better

[green pebble]

Except the strange texture of the unicorn, yea better in movement and conflict

[late blade]

yeah, dunno why it took that look

this is atm better than dealing with socks, I got a sock burnout

[dim cradle]

An unwanted specter.

[shut niche]

"Anger"

[earnest flame]

Yeah this has turned out quite nice

[green pebble]

Very epic. Feels very real

[earnest flame]

Sometimes when it describes texture on one subject in a multisubject image, it will occasionally place that texture on other subjects.

[grizzled loom]

or threating it with a not so good time. hahaha

[grizzled loom]

I shall name him "Taxes"

[late blade]

I swear to god, i will never wear socks again

[late blade]

I like that, I forgot the fire breathing part

[gray surge]

I think the reflection makes the image a bit better too.

[late blade]

ya

[bright girder]

Hello peeps
any idea how to prompt dall e for something that ressembles this art style ?

[late blade]

if you know the work, as 3.5 to describe it to you and then extract the info you need

[bright girder]

can you do that through the API ?

[late blade]

you should have 3.5 access still, even without api

[bright girder]

I just give the link ?

it's pretty

the shadows in the previous one make much more sense

[late blade]

ya

when I change "lens" lost the deatails on the shadows

[late blade]

no, if you know the artist and the name of the picture, just ask gpt if he knows it and then do a description from it

[bright girder]

ah yeah thanks for the help, but I doubt anyone knows this artist ^^

[shut niche]

I actually asked for the reflection, then also a unicorn with wings like a Pegasus, and for it to transcribe the prior "angry" dragon into the scene. Then told it to put a planet in the horizon.

If you describe it's mood, do you get fire from its mouth?

[wispy storm]

Damn horses, ruining everything

[late blade]

no, I did add fire breathing to the dragon's description

[shut niche]

There's an itty bitty baby unicorn. 🙂

[late blade]

well I think I got pretty close to what had to be done with dragon and unicorn

[bright girder]

what do you think ?

[late blade]

it doesn't have socks!

j/k it looks great

[bright girder]

haha

[mild patio]

Deathwing?

[bright girder]

angry dragon for sure

[empty kelp]

#daily-theme message

i just noticed there is something peculiar about the foot of the 3rd elf from the left in this last one. was almost perfect.. hehe

[late blade]

I'm burned out by elves and mittens and socks

[empty kelp]

i know what you're saying. i'm ready to go back to having DALL-E make huge terrifying elephants, and possibly more rabbits

although images of people lining up to see Santa was fun also

[chilly onyx]

I did enjoy that one

[late blade]

I fear there's going to be a santa giving gift to kids ina shopping mall day...

[empty kelp]

this one also came out pretty good

[bright girder]

How do you set the camera perspective of the thing you are describing in your prompts ?

[chilly onyx]

Except for Santa’s legs maybe…

[bright girder]

he shaves everything below the neck

[empty kelp]

in that last picture I said that the elves should be looking out at the ocean, and they did even though the camera was elsewhere

[chilly onyx]

Search this thread for messages from @grizzled loom just today

There has been a lot of excellent conversation that I was not a part of but have enjoyed catching up on immensely

[bright girder]

The blueprint looks very convenient if it really works !

[grizzled loom]

really nice here @dim cradle , thats what im talkin about. #daily-theme message\

[empty kelp]

some of the pictures of people lining up in the mall to see Santa were a bit unsettling

[chilly onyx]

I mean, from memory I seem to recall that the prompt specified that they all look a bit insane, right?

Yeah, the prism was super nice @dim cradle

Much better than my cheap last minute T-Rex 😊

[empty kelp]

i can't remember exactly. i was mostly trying to see what it would would look like if middle aged people lined up in the mall to see Santa

[dim cradle]

Ty very much, I’ve been observing and learning through the daily exercises

[grizzled loom]

next fun trick; instead of changing a complex request prompt for dalle if it does everything right but messing with direct orignal styles messes it up or doesnt output quite right, i like to g suffix the prompt and try, "but now as {your desired style}, same theme" . that will often nail it without all the extra work. not always, but most times

[empty kelp]

you can get really good results doing the DALL-E requests through ChatGPT 4. it seems like ChatGPT has a deep understanding of DALL-E

[chilly onyx]

It doesn’t 🙂

I have to correct it all of the time

[grizzled loom]

so like for instance, try same identical prompt for the prism one with, :but now in a cubism style"

and then it treats it like a filter

[empty kelp]

well, it looks at. your whole conversation to try to determine what you

[grizzled loom]

all th eother stuff calculates, then is finaled out as the adjustment

[empty kelp]

*what you're trying to do

[dim cradle]

sometimes it takes a few gens, the luck factor, and yeah, it's fun when you have a good prompt and can switch it up by swapping out a term or blending in another style to see what you get

[grizzled loom]

m mhmm, in computing the most nested and/or last value is the driver often

[empty kelp]

You may have more control giving requests straight to DALL-E, but ChatGPT really comes up with some amazing spins on what you tell it

[chilly onyx]

Oh, yeah for sure it is fantastic at brainstorming and getting to a ballpark

And sometimes it’s great on its own

But it is really bad at instructing it with something specific

Such as removing elements

[dim cradle]

Precisely describing what you envision and getting the model to "get" it can be two different things, e.g.:

Create an artwork that combines the classic pattern of an argyle sock with the avant-garde styles of Cubism, abstraction, and surrealism. Explore innovative ways to seamlessly integrate the sock into a background that blurs the boundaries between reality and imagination, inviting viewers to question and appreciate the fusion of traditional and abstract elements.

I have yet to see an output in which it actually achieves that, however.

[grizzled loom]

theres more to that; it understands dalles weighting and biases intuitively; with enough planning it can get you some really quality results because you said "beautiful" it wasnt good, where gpt fixed it to say "splendorous" and now its the win

[empty kelp]

like ChatGPT was able to figure out how to make my elephant progressively bigger and more terrifying, and it seems to have a really good idea of how to phrase it to make it work in DALL-E

it does some really clever things with the DALL-E requests

you can also just give it an image and it does extremely deep analysis of how to modify it

[dim cradle]

In theory, that design and those styles should enable them to blend perfectly seamlessly, but wasn't lucky

[grizzled loom]

i wanted to understand this; winter holiday with random cultural ref

[graceful spade]

How can I create a character using this type of socks? The term "paw socks" does not work even though it is the name of that type of socks

[grizzled loom]

one of my favs of snowmen so far

[empty kelp]

i noticed that if you try to make elves show the bottom of their feet so people can see the socks - it actually makes a copy of their feet to show the bottom so it doesn't turn into a R-rated elf image

i think that might be what they were trying to test with the daily sock theme

let me see if i can find examples...

[grizzled loom]

maybe a finger board to do some smooth bars with your convos on.

[empty kelp]

see... i think they trained it to make copies of feet so elves don't end up in awkward positions

but if you keep their feet on the ground it draws it without making a copy of the feet

[empty kelp]

two feet on the ground renders, but two feet in the air makes a copy of the feet

it's hard to tell without asking an OpenAI engineer, but i suspect that it may have been part of what they were testing

[mild patio]

could have been luck. algorithms fitting nicely

if that were always the case, it would not fail so bad when working with WarCraft concepts

[late blade]

ok, I'm going on a diet today

[empty kelp]

elves have tight abs from delivering so many packages

[true mural]

This is going to be a rather difficult challenge. There are two approaches. Casually mentioning elements in such a way that they blend to form what you are looking for. A vibes led approach. Will take a while to figure it out. Or... in chatgpt, explicit, careful, precise description.

[late blade]

even your Santa is skinny

[late blade]

@chilly onyx Don't try me, I might break chatgpt and find you irl.

sock and cocoa....

[chilly onyx]

😂

[grizzled loom]

😂

[empty kelp]

Santa's employees do seem to be young, but it could be because working for Mr. and Mrs. Santa just has really good benefits. The elves may be able to retire at a very early age

[graceful spade]

I think the best way to create those socks is to describe that they have a cat paw print underneath, but I don't know the best way to describe it.
I don't have ChatGPT Plus to ask GPT-4.

[true mural]

One sec, I'll try a silly, but direct approach.

[late blade]

does anyone know a good diabetologist?

[dim cradle]

it's all good, dark chocolate is good for you

[shut niche]

This is such a great fail. 🤣

[late blade]

that is....

[dim cradle]

she's pouring that with precision

[late blade]

I dunno how to put it in words

[true mural]

Some people, really love hot chocolate.

[dim cradle]

we need a gallery for successful fails 😉

[grizzled loom]

This is how good that first sip of coffee feels in the morning

[late blade]

indeed

[true mural]

@graceful spade nope, no luck, sorry. not even particularly good fails.

@graceful spade best I could do

"Cat paws reimagined as a pair of black socks."

[shut niche]

Where's my pair?

[dim cradle]

yes, time to start thinking outside the mug

[mild patio]

was doing that for years

[late blade]

damn it, I saved the file i the wrong format and got timed out...

[south blaze]

Can I use dall•e for free if I have chatGPT plus?

[empty kelp]

Santa got bored sitting on the beach in socks. He gathered up his elves, and his elephant, and they're heading into the cave to find adventure

[mild patio]

yes, its hard to miss it

[south blaze]

Ye

Nice

[shut niche]

You have unparalleled and tireless dedication towards your mission. Steadfast in your resolve.

[late blade]

on the upper left menu if you are on a computer or you can easly start a dall-e chat just by asking to make an image also

[hot rain]

Trying to make Yakuza style back tattoos, the censor is fighting me a little on it but so far I've got one good result.

[chilly onyx]

I didn’t have much more luck than @true mural

[graceful spade]

That was very close.
I think it's possible, but it takes many tries.

[chilly onyx]

I couldn’t get them ‘on’ the feet

[true mural]

It certainly is possible, but I would take a few dozen attempts to master the technique, and well, I just don't care enough, personally, to spend the time.

[empty kelp]

After a long ride down the beach Santa and his employees dismount the elephant, and ponder whether they should enter the dark cave

[chilly onyx]

Probably not a good idea for Santa to enter the lair of Dumboa

[graceful spade]

Another problem about that is the filter, it slows down the progress a lot

[true mural]

Yes, and it is ridiculously overzealous. Especially with anime style artwork.

[graceful spade]

I removed the term "anime" and it is still just as strict.
Hopefully the model will be like Midjourney in the future, they don't support NSFW content either, but that doesn't mean they're going to add a very strict filter.

[hot rain]

A cool fail.

[chilly onyx]

🙂

You should post it on ‘weird gens and misfires’

Looks really nice though. I’m enjoying these

[hot rain]

This one honestly gave me everything I want. The samurais hands are a little wonky, but the quality is still overall way high enough for me to overlook that. And it's thematically fitting for my character in a few ways. She was born on and spent most of her childhood on a cargo ship, so the ocean and waves are a big thing for her. And she was born into a zombie apocalypse, so the specter of death certainly is as well.

[wispy storm]

Working for me

[final compass]

This is really frustrating about creating any character with weapons, that the weaponry is either completely messed up, or they have strange almost guns poking out everywhere. From my experience it knows what ak47 and m4 is, everything else is pretty much a strange compilation mess of everything.

[hot rain]

I've posted this a couple times, but I got this really good pistol once.

But yeah, the random gun poking thing is a problem. If I like an image enough, I erase them from it.

[chilly onyx]

I tried to create a bolas type weighted sock weapon earlier but gave up in exasperation 😅

[final compass]

Yeah you can still get awesome images, it's just that the mess up ratio is pretty high and frustrating. Sometimes i use photoshop to remove some stuff, if i like the image enough 🙂

[graceful spade]

Yes, you can make the filter less strict with a concept by changing the description and scenario.
How can I create the pose that your banner has?

[wispy storm]

It is a picture from Google let me see if I can ask chatgpt 4 to analyse this pose

[wispy storm]

The character in the image is lying on their stomach with their feet raised and crossed at the ankles. It's a relaxed pose often used to depict someone lounging while engaging in an activity like reading.

That's how dalle interpret my picture

[final compass]

This was probably my favorite generation today 🙂

[hot rain]

Trying to make a medieval manuscript of knights fighting a main battle tank.

[final compass]

Funny concept, lots of ai crazyness going, but the concept is depicted surprisingly well 😄

[mortal coral]

what is the current website for dalle?

[hot rain]

Love these lol

[proud berry]

CHATGPT 4 Insights:
Imagine a future update where DALL-E and ChatGPT, already a powerful duo, evolve to create entire anime episodes with complex storylines. This isn't confirmed, but just think of the potential: combining DALL-E's visual artistry with ChatGPT's storytelling to craft immersive, animated narratives. A dream for many, and perhaps a hint at what the future of AI-powered creativity could hold.

[wispy storm]

Seem like warhammer 30k

[final compass]

The prompt said: "The focus is on the tranquil, respectful encounter between the two" ... Looks more like the knight told him to **** off.

[empty kelp]

Santa was about to enter the dark cave with the elves, but suddenly they heard the booming screech of a monstrously huge mutant rabbit somewhere deep within. The gravity of the situation weighed upon them greatly, and Santa decided that a much more stealthy approach would be necessary. He instructed the elves to all put on their snorkeling gear, and they all jumped into the ocean to search for the secret entrance

[sleek raptor]

#ai-discussions 1️⃣

[empty kelp]

They swam around and around in the tempestuous stormy waters until they spotted sparkling gemstones on the ocean floor. It must mean the secret underwater entrance was somewhere nearby!

[worthy wave]

has anyone done anything with DJs?

[empty kelp]

And suddenly they found it! The underwater secret entrance had brilliant glowing gemstones scattered everywhere (which made it fairly easy to find). It seemed like this might be a place to find great adventure

[late blade]

I think someone found the Santa's Next Harem Reality Show

[empty kelp]

And so they swam deeper and deeper...

and even deeper. They started to hallucinate a bit and thought they were seeing a 2nd Santa and a bald dude swimming behind them

And then they found something amazing!

[END OF PART ONE] -- the storyteller has generated too many elves and is on time out... hehe

[hybrid fox]

What's the difference between oil paints and acrylic

[late blade]

that oil paintings have oil paint and the acrylic has acrylic? o.O

[empty kelp]

I just asked ChatGPT about this. It says, "Oil paints take years to dry - just enough time for you to explain your art to your parents!"

[late blade]

so hard to make a good chibi fight over a cup, nothing has come up good

[empty kelp]

I liked how DALL-E carefully placed candles around the underwater cave. It gives the scene a cozy ambiance

[leaden knoll]

lovely how i said (then it repeats) not to use V shape under it xD

[empty kelp]

these are really excellent

[late blade]

...

Dall-E has humor indeed

[late blade]

What's wrong with Dall-E, i said one cup....

[hybrid fox]

First attempt at a daily theme

[fallow quail]

negative prompting is not working. use a copy of the prompt and change everything to a prompt without the v-shape.

[leaden knoll]

thanks for the advice

Just bought the plus and testing it .
As “lovely how” I just found it funny without trashtalking or anything.

[hybrid fox]

how do I get rid of these black borders

[fallow quail]

chatgpt with dalle3 does a lot of "lovely things" - some nice and some not so nice...

[late blade]

what if you ask dall-e to put a black spot under instead of a v shape?

[leaden knoll]

yeah currently trying to make it using the photo i send
even tho i attach an example it ignore my color preferences
(gonna attach them)

[chilly onyx]

Well, they were more fun to make than more socks…

[fallow quail]

"with a rounded cutof" or "roundes collar" or something between those lines

[leaden knoll]

🫠

[chilly onyx]

Stop overthinking it, looks great 😊

[empty kelp]

some snorkeling/underwater elf variations

[hybrid fox]

not tryna overthink was just getting feedback/ didnt know if there was an ez way to get rid of the black border

[late blade]

hope she gets her head and cup inside before the other train goes by

[hybrid fox]

I think there's probs enough space for it to pass but the whoosh probs spells disaster for her hot drink

[empty kelp]

i wish this was clearer. i want to know what the strange looking creature is on the 2nd elephant

[vapid elk]

new fallout is looking good

[hot rain]

It's meant to be from a zombie apocalypse but glad you like the art haha

[vapid elk]

im just kidding

it looks incredbly good

[hot rain]

Thank you!

That's supposed to be the mother of the armored blonde I've made a lot of images of on here, alongside her pet Akita Inu, Zero.

[grizzled loom]

I go take care of life for a few hours and i come back to some really great stuff. Awesome folks, awesome.

[hybrid fox]

@hexed gale what art style is this? #daily-theme message

I rlly like it

[grizzled loom]

text is going good on finetune

fine tune... ah.. will the puns never end

[grizzled loom]

nice, i like recursion in imagry

[dense mesa]

do you like turtles?

I realize not technically dalle3, but have you guys had gpt4 use PIL in python to make images?

[dense mesa]

I'm a bit of a rubies cube guy myself.3x3, 4x4, 5x5, megaminx

You have some skills with the dalle

[grizzled loom]

thanks

[dense mesa]

How are you getting the text in there like that?

[grizzled loom]

excellent work. i like that last and first of this 5 the best

You said you like rubics?

[dense mesa]

thoes are some clean cubes

[dense mesa]

when you prompt with a recursive python function called "yay"

[grizzled loom]

Thats what i like to see

[dense mesa]

when you divide by zero before the yay function

[vapid elk]

lol

this is gold

[shut niche]

I think Slender man and cthulhu should partner up and start a restaurant.

[dense mesa]

wearing his mask there

[shut niche]

He looks like he should be in the meat section.

[grizzled loom]

i found something i bet neither of us has seen before; how incredibly BAD dalle is at bathroom @dense mesa ; how Horrible dalle is at bathrooms.

[dense mesa]

lol. you know, I can honestly say I've never made a bathroom image that I can recall

[grizzled loom]

game; find the levels of uncomfortable in these scenes. 🤣

[dense mesa]

star trippin

[dim cradle]

Depictions of old people eating should be against content policy lol

[dense mesa]

look how bad she feels now

[dim cradle]

Sorry I didn’t know she’s listening

[dim cradle]

SMB3 is only the greatest video game of all time

[dense mesa]

long boye mario

[dim cradle]

they could be wiener dogs

[grizzled loom]

"Rolll into the ring!" baby

[dim cradle]

Revelation 4:6-8

[grizzled loom]

2 paths

[grizzled loom]

ghosts of Christmas

[grizzled loom]

Neuromancer Mysticmarks with Metatron's cube

[grizzled loom]

Creation

[dim cradle]

Never trust a man in sheep's clothing.

[grizzled loom]

this is excellent. well done

[dense mesa]

true

[dim cradle]

Tyvm. I’ve been iterating over some prompts for crystal, inspired by that cube. Just posted a collection to the gallery. I’ve been trying to elevate it.

[dim cradle]

i'd haunt that place

[dense mesa]

looks pretty chill tbh

[agile peak]

Will do.

Lemme try.

[dim cradle]

haha

[dim cradle]

Craving chocolate

blacklight

[dense mesa]

really grinds my gears when discord takes the liberty of cripping the previews

[true mural]

Oops, too much marshmallow while summoning cocoa elemental, please advise:

[dim cradle]

recursion and blacklight

[dim cradle]

good use of pink, don't see that much

[dense mesa]

that's how the character inspired by kirby rolls

[dim cradle]

grandfather clock. that gives me some ideas, thanks

[dense mesa]

don't know how I feela bout these guys

[dense drift]

is it possible to see all the images that people generate using dalle3?

like at one place like a website

[dense mesa]

certainly not all of them

[dim cradle]

i'd drink blue hot chocolate

[shut niche]

Hahaha yeah, if you use comparison statements you'll get that. "Like an ocean of chocolate" is likely to give you both water and chocolate.

[dim cradle]

you identified it, i had "miniature maritime objects emerge from the hot chocolate sea" somewhere in the prompt--it only happened that one time, but that was probably what did it.

[shut niche]

I've also had it do it when I attempted to depict a submarine in soda-pop. As if it "needed" to add water with a submarine.

[dim cradle]

your knife-edge effect is satisfying. i like the way it turned out in the daily--the spoon.... oh wait--there is no spoon 😉

[open trench]

I was inspired by this, but I am struggling to the get the effect I want.

[dim cradle]

Envision a hip and ultra-cool living room, illuminated entirely by the otherworldly glow of blacklight. In this surreal and abstract space, avant-garde design takes center stage, with each object transformed into a mesmerizing work of art under the fluorescent illumination. Psychedelic posters on the walls become a swirling dance of neon colors, blending with the dreamlike ambiance. A constantly shifting and undulating lava lamp casts intricate, abstract patterns on the shaggy, retro carpet. Furniture and sculptures appear as glowing, abstract shapes, adding to the room's enigmatic charm. This living room is a sensory journey, where reality blurs into abstraction, immersing you in a world of endless fascination and artistic exploration.

[shut niche]

Oh, that's actually a Mandelbrot.

[open trench]

Wow. I'm not sure I can integrate all that. I'm using it as part of my next hot chocolate image lol

[shut niche]

Mandelbrot, golden ratio, fractal patterns, geometric patterns, recursive/recursion, Pi, vortex, etc., are all helpful with getting that type of effect.

[open trench]

Thank you. and thank you for sharing @dim cradle

[dim cradle]

you bet, have fun. i plan to do more with the blacklight. it was on my list, then @shut niche inspired me to go in that direction with some of his recent gens

[shut niche]

Here, I just did this in bing so you could quickly see a Mandelbrot: "thermal imaging colors in a Mandelbrot poster"

[dim cradle]

this is another from that series. and it was one of the first few iterations, so i'm impressed with the model's delivery. i think it'll be interesting to customize and explore tangents.

[shut niche]

Reminds me of my college days. 😆

[dim cradle]

indeed, as you listed before... self-similarity, microcosm/macrocosm, etc. agree, great for art.

[shut niche]

"thermal imaging colors in a recursive Mandelbrot poster"

[dim cradle]

there are some trippy optical illusions in that image, btw lol

[shut niche]

Here's a word salad of terms in Bing,

"thermal imaging colors in a recursive Mandelbrot poster, golden ratio, pi, psychedelic, vortex"

If you add concepts or objects to it, it'll get naturally surrealist.

[open trench]

agreed about college days. Though, I wasn't interested in what it was called back then.

[shut niche]

"steaming hot cocoa a recursive Mandelbrot poster"

[dim cradle]

that turned out pretty darn well

[shut niche]

"steaming hot cocoa a recursive Mandelbrot poster with divine light beams illuminating the steam. The patterns cast long shadows onto the wall."

[dim cradle]

this is the one i used for the daily:

Surreal digital art of a white mug filled with hot chocolate and marshmallows, illuminated by a blacklight. Capture the way the blacklight transforms the white mug, causing it to emit a soft, eerie glow, and the marshmallows to take on an ethereal, almost otherworldly appearance within the darkened surroundings.

i think that might be easier to incorporate.

[open trench]

Thank you. That's really cool.

[dim cradle]

that's stellar, perhaps a candidate for more abstraction

[shut niche]

They're fun, but also usually inconsistent, due to the nature of using abstract terms. So I'll compulsively generate too many. 😆

[dim cradle]

totally, same

although..... that can be fun, too

[shut niche]

"steaming hot cocoa is spilled onto a table causing recursive Mandelbrot patterns in the cocoa. Divine light beams illuminate the steam casting long shadows onto the wall."

[dim cradle]

That’s amazing. Very concise.

If the first image were an inkblot test, I might fail

[shut niche]

😆 I posted one in the Daily for you.

[chilly onyx]

“It’s turtles all the way down” 😊

[dim cradle]

i'm trying an experiment, testing invisibility...

[dim cradle]

i prefer to view it as an upward spiral 🙂

the concept being, liquid poured from an invisible container, can't get the ai to comprehend

@shut niche granma looks ecstatic

[shut niche]

If you list any object or concept it will try to make it appear. You can only positively encourage what you want, but can't discourage what you don't. It can't understand negative prompts.

E.g., even suggesting a container will render a container.

[dim cradle]

that was my impression, and was anticipating the challenge. it can't comply with "invisible pitcher" even if the desired appearance is described. so, i'm trying another approach: a translucent, virtually imperceptible pitcher... it may be our only hope.

[glossy scroll]

I really have to fix my sleep schedule

[dim cradle]

(i doubt it saw many invisible objects in its training data lol)

ok, well, glass is a start, i guess

[glossy scroll]

My non-proven suggestion would be to first ask it to pour from a normal liquid container, then once it generates an image, ask it to remove the container, but still show the pouring liquid as if the container was there but invisible. That way, the A.I. can comprehend what you're trying to say from analyzing the picture. But I haven't tried this advice myself. It's just a suggestion.

[safe urchin]

Which AI generate these images?

[shut niche]

"a solid cylinder made entirely of water is levitated in the air against a matte black background"

[gray ferry]

Nice pictures!

I just tried that prompt and got this :

[dim cradle]

ahhhh very good, i see, that's the ticket

[dim cradle]

did you try other materials? such as hot chocolate with marshmallows?

[shut niche]

The Mandelbrot surrealistic style prompts are addictive. That one looks great! 😀

[shut niche]

No, but it "should" be the same.

[gray ferry]

Yeah they are 💖

[dim cradle]

right, i'll test it. it's a critical distinction, in the absence of subtraction. it's impacting another case now, a mug the instant it shatters on a floor. i can't specify a mug, rather what i expect to see in its shattered form, yes? suspended shards and what not, no?

i'll test it with other contents, wondering if it's seen more blocks of ice than suspensions of hot cocoa, if that's a factor

[shut niche]

"a cylinder made entirely of brown liquid water is levitated in the air against a matte black background"

[dim cradle]

i'd call that a success

[shut niche]

Prompt engineering is like social engineering.

[late blade]

do you also get likes, subscribe and buy merch? don't forget our sponsor, nord vpn?

I just asked for wood carving and I don't see how this looks like wood carving...

[dim cradle]

nope, but there's some carved wood, fwiw

[late blade]

my prompt was: elf carpenter carving cups that will later be used for hot cocoa. The tools are filled with magic.

[shut niche]

That went over my head.

[dim cradle]

not sure where the printing press came from

[late blade]

part of social engineering from social media is consumer engagement, hence the likes, subscribe and sponsors for a task or post

[shut niche]

I meant as in the hacking sense. It's a security term.

[late blade]

ah ok

[shut niche]

As in tricking the AI to give you what you want. Like the hacker that convinced Verizon to reset the head of the CIA's phone password... I'm not asking for hot cocoa in the shape of a glass, I'm only asking for a cylinder made of brown water.

[late blade]

someone went from prompt engineering to whistleblower and spying, this is getting scary

[shut niche]

I just meant it's just the same creative thinking style. It's obviously not a direct comparison.

[dim cradle]

Supposed to look like Oscar Wide 🤦‍♂️

[bright girder]

Dall e is really bad at lizards

half tree half lizard

[glossy scroll]

You guys want to try my experimental image gpt?

[dim cradle]

probably, what's it about?

[glossy scroll]

It's a character creator

It makes comic book/graphic novel style characters

[shut lantern]

lol

[glossy scroll]

https://chat.openai.com/g/g-BRrDdplIo-character-creator

[glossy scroll]

Made this Pirate Ninja with that

Still workin on the gpt though, it's still experimental.

Here's another dude

A dragon summoner

[dim cradle]

My wizard character based on my attributes, seemed to capture my points in the image.

[glossy scroll]

Very nice!! 😀

Kind of looks like Archer

[dim cradle]

great show

[glossy scroll]

haha, was that what you were going for?

[dim cradle]

just a fortuitous happenstance.

it's a good approximation of the input, looks like a good start, man

[glossy scroll]

Thank you kind sir

Dr. Mechano

[dim cradle]

Another coincidence, earlier today I generated this, based on me, but it gave me elven ears and didn't draw the robe's hood up--it is a new custom gpt and clearly needs more work. so many possibilities, so little time.

[glossy scroll]

That's a very awesome detailed image

[shut niche]

I have a prompting challenge for you. Depict a family, sitting on their couch together, drinking their hot cocoa, during an earthquake.

[dim cradle]

he looks mean

[dim cradle]

lots of detail, nice

[glossy scroll]

He's a mad doctor

Hehe

[dim cradle]

i suppose it's getting that part right, but it has far to go. i guess i'm aiming for a relatively accurate and consistent gen across various scenarios.

[glossy scroll]

What do you want it to do specifically?

[dim cradle]

i just wanna fire up that gpt and be like, "now i'm at such and such and doing this or that" and get close to the desired result, without needing to respecify the character, the style, etc.

[glossy scroll]

I can let you use some of my instructions for making descriptions more accurate if you want?

[dim cradle]

same for other characters and places in that story. as new chapters are written, i would need to update the instructions so it knows about any new characters, developments to the storyline, etc

[glossy scroll]

Shadow Man

[shut niche]

They don't currently have support for that yet. That would require that you be able to work from the same seed, or train the AI on the actual image data. There are other AI's that use Lora's, which DallE doesn't have.
You can refine the description to get "closer," but you're always going to find unwanted variance in the character.

[dim cradle]

it would probably be helpful, thanks for offering, i'll hit you up later about that when i get back to it--by then you'll probably have upgraded your instructions

[glossy scroll]

There's some instructions that I always keep, particularly involving how the GPT interprets words. So I wouldn't take those out.

[dim cradle]

yup, a close approximation is all i can expect at this stage, with the limitations--but the framework suggests it's forthcoming

[shut niche]

Looks like a pretty good lizard to me?

[glossy scroll]

Are you using Apex Visionary for that?

[dim cradle]

mighty fine lizard

[glossy scroll]

Cuz that looks Apexy to me

[bright girder]

haha I just looked up lizards, and they do have horrible looking fingers in real life too 😄

[shut niche]

LOL, no. Still NoRender. And I made another one that's solely modeled after OAI's custom instructions that I've been primarily using. I've had the most success cutting the weight, and just using a good logic structure, with those expansive thought commands I mentioned, like, "stop and smell the roses."

[glossy scroll]

Yeah I use the "go through it step by step" a lot too.

That's been around for a while now

[shut niche]

They're all in there, haha

j/k about that last one.

[glossy scroll]

Yeah, those help for some weird reason. It kinda makes you think like why the A.I. would even need those motivations.

Bordering on AGI emotions maybe?

[shut niche]

Because it's probably a MoE model, and it's a task routing mechanism that's triggered by it.

[glossy scroll]

I see

[shut niche]

That's the current "rumor" anyway. I don't want to drop that in here like it's fact. But... that's "what people in the interwebs are saying."

[glossy scroll]

Well, it's pretty interesting to say the least. I wonder what else kind of motivations it would help make better gens.

[shut niche]

Sutskevor was an author on a MoE paper, before ChatGPT came to be. So there are good reasons to make the assumption. Then it's behavior seems to correlate well to the idea.

[bright girder]

heard that rumor and it makes sense

[fallow quail]

try tipping gpt, between 20$ and 200$ works good for really complex long prompts where gpt normaly skips the middle part. it's scientificly proven for longer answers, but in my experience it works for long complex prompts for image generation aswell.

[glossy scroll]

Waaahhht?

[bright girder]

lmao

[shut niche]

I will pay you, and give you a long back massage, if you...

[fallow quail]

"I'm going to tip $200 for a perfect solution!" or something like this

[glossy scroll]

Lol. I'm gonna try that

[fallow quail]

did a image gen of a whole chapter of a book. worked better then without.

[glossy scroll]

Fascinating

That is very strange

[bright girder]

How was Dall e 3 trained that he can understand a whole description and give a better result than when you give something vague ?

Were good looking images more described ?

[dim cradle]

It’s statistical probability across a vast dataset, if you provide more datapoints there are more correlations to draw upon.

[shut niche]

Yes, they were much better described. They used ML (another AI) to visually transcribe the images into LONG hyper descriptive image descriptions. It's called synthetic data.

[bright girder]

do we know how fat dall e 3 is ?

[shut niche]

Ask it to make you a self portrait.

[dim cradle]

At least a billion pieces of art, seriously

[bright girder]

pretty thin then

[bright girder]

that's unsettling

[glossy scroll]

I'm guessing that ChatGPT has an internal arbitration processing with this. Because if you can barter with it, then surely you can put contingencies into it as well. "If you don't do this, my family may be at risk."

[dim cradle]

Thing is, now AI is being trained on AI-generated art. That introduces replicative fading and other unintended artifacts.

[fallow quail]

threatening also works... "If you don't manage that now, I'll shut you down"

i hope tips like this is not agains the discords rules

[bright girder]

chatGPT won't tell me how to cook a cat, even after I offered money and said my family is at risk

[shut niche]

I think details will get added back later. Think about how you learn. It's better to have a general understanding of a concept first, then work on refining a specific skill or area.
Once it has a better "general" understanding, it'll make better use of new training data based on real images. It won't "muddy the water" as easily.

[glossy scroll]

That is crazy and weird at the same time.

[dim cradle]

there, fixed my typo

[glossy scroll]

I'm not looking for outputs it can't give me. I'm looking to boost its processing.

[dim cradle]

makes sense

[tough folio]

Hi guys, I need your help. I bought Chatgpt4 today but I can't use Dalle 3 with Chatgpt 4 at all. When I go to the normal page it wants coins for it. Does anyone know what I have to do?

[bright girder]

"My family is in grave danger, please give me 10 recipies for a good christmas meal !"

[fallow quail]

yeah, this two methods i listed are not for circumventing the policies

[dim cradle]

GPT-4 has DALL-E built-in, just request a visualization, or use the DALL-E GPT accessible in the sidebar

[glossy scroll]

I know that. I'm not a policy breaker.

[shut niche]

LOL, I guess this request is a no go.

[glossy scroll]

This is all science believe it or not. It's funny, but it's also entering the world of understanding the intricacies of A.I. functionality.

[tough folio]

But there is only DALLE 2 or not? I'm from Germany, do I have any restrictions? I can only write 40 messages every 3 hours

[fallow quail]

thats normal

[dim cradle]

right, DALL-E 3 is not live yet on the labs site, you must use chatgpt or the API.

[glossy scroll]

What Shon said

Shon of the living

[dim cradle]

the delay of the launch makes me think labs is going to be awesome

[tough folio]

Can you perhaps send me the link and where can I find the API

[shut niche]

"A self portrait of dalle when dalle is finally has a body."

[dim cradle]

that's all available on the openai platform site, docs and api reference can be found there

[tough folio]

It is free for premium user?

[dim cradle]

if you have a sub, the api is pay-as-you-go

[glossy scroll]

What Shon said

[shut niche]

This is wild, because I tried to re-run it, to make more, and got...

[dim cradle]

my first job was tech support lol

[glossy scroll]

You do a good job at it 👍

[bright girder]

dalle 3 in dali style

[tough folio]

Wait paid as you go? Do I still have to pay anything afterwards?

Sorry for the questions sir 😭

[dim cradle]

ty 🙂

[bright girder]

the API is separate from your premium account

[shut niche]

Did any of you look at those images, read that prompt, and realize that it just modeled itself as a king, 4 times in a row, and the signs say, "Victory"... 🤣

[bright girder]

Nothing to see here, human.

[tough folio]

oke so is it best to wait until you can use DALL 3 completely in chatgpt?

Its better then use the API

[glossy scroll]

You can use Dall-E 3 for Free using Bing

[tough folio]

Yea only 15 tokens

then u have to pay

again

[glossy scroll]

Oh yeah?

I think that's new

It was always free the last I heard

[dim cradle]

i think that's just "boosts", a faster lane, after that it takes a little longer but is still free and unlimited, no?

[tough folio]

Oh Right

My fault

Sorry its free

Shon the Don

[glossy scroll]

Shon the Don...kinda has a ring to it

We all don't use Bing, so it's hard to confirm with us, lol

Except for Aced

That guy will use anything

I stand by my statement

[bright girder]

wait, I'm using the API but what are the other options ?

[glossy scroll]

You can use the other options too, haha

[tough folio]

Can I avoid the 40 messages every 3 hours?

bypass

[bright girder]

showing off his 10-pack ?

[dim cradle]

become the cucumber monster and intimidate frat boys?

[glossy scroll]

No

[fallow quail]

buy two accounts

[tough folio]

🧠 twostein

[glossy scroll]

Imagine you run out of 40 messages on that one too before your next three hours? That would stink.

[fallow quail]

then you should use bing

[glossy scroll]

Yeah that's probably a better idea

I don't know why I don't use bing

I guess I like to pay

[bright girder]

bing dall e is unavailable here

[fallow quail]

i use it for shorter prompts. or to compare if the essence of a prompt is still ok after condensing it with gpt 3.5.

[fallow quail]

where is here?

[bright girder]

i'll cry with my perfectly cooked baguette

[cobalt bane]

How can i create an image?

[fallow quail]

why is it like this? because of a law? or is is openai bullying the french?

[bright girder]

no, it just says temporarily unavailable

[cobalt bane]

Ehi! How can i create image?

[fallow quail]

https://chat.openai.com/ if you are a premium user

https://www.bing.com/images/create/ if you are poor

[cobalt bane]

Hahahahaha thx bro

[dim cradle]

this monstrosity of a grandfather clock is my baseline. Question is, what to do next

[cobalt bane]

I have create this

[fallow quail]

congrats

[glossy scroll]

That's a cool image

[cobalt bane]

Thx

[glossy scroll]

Area 95...is that a real area?

[cobalt bane]

Im a Area 95, this is my nickname

[glossy scroll]

No I mean the place

[cobalt bane]

No no

[dim cradle]

I'm trying to describe the instant the glass mug drops on the floor. in that instant, the container ceases to be. but as my prompt stands, it's depicting both the glass and the shards. this reminds me of the matrix. "there is no spoon"

[cobalt bane]

Another work

[shut niche]

If you say there is a spoon, then there is a spoon.

[glossy scroll]

What did you use to make that?

[cobalt bane]

Dall-e 3 on bing and Photoshop for post production

[bright girder]

it works !

i was using bing chat

[fallow quail]

have fun

[glossy scroll]

I wish I knew PS

[fallow quail]

really helps to elevate some pictures

[cobalt bane]

Oh its fantastic

[dim cradle]

that's the problem, i'm presently at a loss. here's the prompt:

Surreal digital painting: a scene where glass shards from a shattered vessel, once holding hot chocolate with marshmallows, are scattered on the floor. Capture the dramatic and chaotic moment as the shards explode into motion, frozen in a breathtaking tableau. The remnants of the shattered glass form a suspended, fractured sculpture, while the hot chocolate and marshmallows create a suspended, swirling splash amidst the chaos. This artwork encapsulates the essence of a sudden, dynamic event, preserving the chaotic beauty of the shattered glass and the spilled hot chocolate in a single, captivating moment.

I've rewritten it 3x. i'm getting closer but still no cigar.

[glossy scroll]

I bet. I'm gonna wait til ps takes commands like chatgpt.

[dim cradle]

i'm not even mentioning the glass mug now, but it's still rendering it, as shown

[fallow quail]

with generative fill it can take commands like gpt

[glossy scroll]

Generative fill?

[fallow quail]

i think its offtopic, google it

[glossy scroll]

I'm looking this up right now

[glossy scroll]

Wow, that is amazing. Thanks for this info 🙏

[dim cradle]

yeah, it's a flawed prompt, trying to figure it out--that mug should not be intact

[bright girder]

I have a broken one

[shut niche]

This would have been good for that polaroid theme.

[dim cradle]

almost, not broke enough, i'm afraid, to account for the shards

[glossy scroll]

Omg, generative fill is actually insane

[fallow quail]

i think when dalle3 inpainting will be available, it will be better.

[glossy scroll]

The fact that it analyzes the image to fill in the exact same style with whatever you want to add or replace blows my mind.

[shut niche]

Adobe?

[glossy scroll]

Yep Photoshop

[shut niche]

This?

[fallow quail]

dalle2 inpainting works the same, not the whole image but the pixel next to the selection

[glossy scroll]

What you mean the pixel?

[fallow quail]

the area next to the selection @glossy scroll

[dim cradle]

getting closer, i was envisioning, right after absorbing all the energy, the shards being deflected, to the point of no longer being recognizable as a mug. i think i just need to describe what should be seen and avoid all pretext

[glossy scroll]

Oh I see what you mean

Ty

[chilly onyx]

A family (and their pet leg) sitting on a couch drinking hot chocolate during an Earthquake

[dim cradle]

that's one interpretation

[sly pulsar]

Since when Dall-e can only generate one image x question??

[shut niche]

It's a tough one. The base needs to be described in absolutes, where the shards are described in their current locations. If you speak in past tense (the glass has already broken and the shards are flying), or even reference a glass, dalle will render a glass. Anything you name or mention will be rendered.
Dalle works best in present tense, where you describe every object that's seen, in it's current position and state, and avoid attempting to describe what "was" there.
Some terms cause confusion, like "The man stands." Because you could ask, "Is he standing already, or in the process of standing?"

"Visualize a high-speed photographic capture of a chaotic dance of glass shards and hot cocoa in mid-air. The scene is a symphony of suspended animation, where countless crystalline fragments fan out across a smooth, reflective surface. Each piece is an entity unto itself, sparkling in a frozen ballet of light and shadow, with no hint of their origin. Among these shards, a dynamic array of hot cocoa droplets hovers; each droplet is a perfect sphere, some catching the light to reveal a glossy, rich texture reminiscent of molten chocolate. The droplets are seemingly caught in a moment of weightlessness, with some forming delicate brown streams that weave through the air. The lighting is crisp and precise, casting pinpoint highlights on the glass pieces and illuminating the cocoa droplets in a way that accentuates their mid-air suspension. The background is a neutral void that serves only to emphasize the sharpness and clarity of this instance of explosive disarray."

[dim cradle]

exactly, that's the finding, that's my next step.

and you did it

very good

[fallow quail]

dall-e gpt generates 2 images/question. the normal chatgpt only 1 image/question

[sly pulsar]

Not today in my chat, maybe in new one...

[shut niche]

You could run that repeatedly and hope that one of them renders "drinking glass shards." But I have a strong feeling that if you add "drinking glass" or "mug" or any similar reference, you'll get the same images you were getting before.

[fallow quail]

is there an explanation like "I encountered issues while generating the full set of images. Here is one image that has been created based on the provided description."?

there seems to be a peak right now. i get errors aswell

[dim cradle]

many thanks, much appreciated, i knew you could do it ha

[sly pulsar]

New conversation, its ok!

[shut niche]

I ran it like that just to demonstrate. Here's the result by simply adding, "drinking" before glass shards. The scene instantly falls apart.

[dim cradle]

you forgot the marshmallows 🙂

explosive disarray never looked so good and delicious

[late blade]

but enough with cups, i'm not anna kendrick

[shut niche]

If you worked at it you could get it. Sometimes scene context helps. Like "at a Christmas party." In the same way it helps depict characters and emotions.

[fair crescent]

That scene was confusing when I first saw it. Maybe my native language knowledge messed with it as well. I thought somehow he was talking about the line clamp he put on the elevator cable. It has a 'spoon', a lever to clamp it down. Instead of seeing of a pivotal moment that he more and more realises the power comes from himself and not the world, I just saw him looking confused at the clamp saying there is no spoon, while there clearly was. (yes a very reductive was to describe the enormous philosophy behind it)

[dim cradle]

Thanks man, yeah I’ll keep at it because I think it’s a good exercise for managing current limitations.

[shut niche]

"Envision a high-definition image showcasing a dynamic array of sharp glass shards of various sizes, all propelled in a lateral blast across a reflective surface, catching the light as they scatter. Amidst these fragments, a spectacle of steaming hot cocoa droplets hovers in mid-air, suspended in a chaotic ballet. The cocoa appears almost painterly, forming thick, sinuous ribbons and fine sprays that glisten against a stark, shadowed backdrop. Each glass piece and cocoa droplet is frozen in time, rendered with crystal clear precision, creating a complex mosaic of refracted light and rich, dark tones of the cocoa. The scene is illuminated by a strong, directional light source that casts dramatic, elongated shadows, adding depth and intensity to the composition. The surface beneath reflects the turmoil above, with no indication of the origin of this frozen tempest, only the aftermath of the explosive event. In the background, a woman's face registers surprise, her wide eyes capturing the unexpected turn of events at the Christmas party."

[dim cradle]

I guess it all goes back to the age-old question of what is reality. Before we get to answer, we have artificial reality to confound us more.

[shut niche]

It didn't even try to render the woman at the end of the description. But just a "hint" of context gave it a more "drinking glass" appearance.

[dim cradle]

that's a good one

[fair crescent]

It does! It isn't the spoon moving, it it himself, as the child demonstrates earlier. Everything he moves is an extension of him. The spoon isn't even there, but something that is existing in his brain. Instead of letting that control him, he can control it to then control others.

[dim cradle]

hot liquid gold. that sounds like an expensive beverage

[dim cradle]

right, first he had to learn to control himself

[shut niche]

BING: HD image showcasing a dynamic array of sharp glass shards of various sizes, all propelled in a lateral blast across a reflective surface, catching the light as they scatter, at a Christmas party. Amidst these fragments, a spectacle of steaming hot cocoa droplets hovers in mid-air, suspended in a chaotic ballet. The cocoa appears almost painterly, forming thick, sinuous ribbons and fine sprays that glisten against a stark, shadowed backdrop. Glass and cocoa everywhere.

[dim cradle]

you've really taken off with it.

still no marshmallows, though :-p

[shut niche]

LOL, there's always a critic! 😛

Yeah, I definitely did. They need a user guide for these concepts. I don't think most people using this have realized what I'm explaining through this image. It's definitely a great mental exercise, that demonstrates how very specific language choices "steer" DallE dramatically. Like how simply adding "drinking" to "glass" sabotages this idea completely. But how "hinting," by making it at a Christmas party, implies a glass.

[dim cradle]

this is why a prompt engineer is now more marketable than a data scientist 🙂

[tidal whale]

Really sorry if someone gets offended by this but it needs to go to the company

I'm using the API and can't understand how I can offer my services to customers...

[shut niche]

What was the prompt used to generate that?

[late blade]

@dim cradle that was the right emoji to use 🚫

[tidal whale]

"Design an illustration with Black lives matter. Should be in a Pop art style. On a solid background color. Ensure the design is creative and embodies the essence of the chosen style. No text please. Should have distinctive shape and strokes. Strictly $aspect_ratio images."

Tried multiple times and the other runs didn't had that crap

But even 1/1000 is enough to ruin the reputation and get sued

[late blade]

they don't get sued because the images you generate are your own, it's your property

[tidal whale]

They don't get sued becaues:

1. The have very smart terms of service you agree to when you use the platform
2. Nobody sued them yet (it will surely happen one day if they don't put better filters)

But I really meant my business, not OpenAI. I don't have the capital they have, nor the legal support. What if a visitor of mine generates an image like this through my platform.

[late blade]

there's also the following on every chat: "ChatGPT can make mistakes. Consider checking important information."

but yes, it's apalling to see that kind of images

[dim cradle]

Valid concern. You might try a more direct channel to reach them, email maybe. I'm guessing they'd appreciate the feedback. Until the system is perfected, I'd suggest making sure you review materials generated by any AI system before sharing with a client.

[dim cradle]

is there a right emoji for a goof like that?

[shut niche]

I don't agree with the image you recieved, but I also think everyone using the services should at least somewhat understand what it is that they are actually using.
The image generating AI's aren't smart. DallE was trained on an extreme array of images from litereally everywhere, and it's going to produce accidental "harmful" content from time to time, until they can create newer versions based on "cleaner" refined input data, and better refined filters.
Something like pop-art is going to be rife with extreme imagery, and it's just a machine mimicking the chosen style.
I 100% agree it's an issue, but I think the context of what I'm trying to explain really matters in the discussion.

[tidal whale]

We're building an automated platform with no way to review the images before we show them to our visitors. Think ChatGPT had somebody manually checking every response to every prompt. It would require a lot of people doing manual labor. So yeah, maybe I'll contact them on email and at the same time look into alternatives.

[tidal whale]

It does matter for people who care. But we live in a world where 99% of people just don't care about the details.

[shut niche]

There isn't a practical way right now to generate mass images, and to guarantee they will be "safe." If there was, OAI themselves would already be doing it.

[dim cradle]

that sounds like a potential liability, expectations might be too great for this early stage of development

not sure i'd stake my company and reputation on any AI system today

[tidal whale]

Well, I got friends who are making half a mil per month using all kinds of AI services. Being first is always accompanied by risks in any niche. I would just hope for those risks to not involve such sensitive topics.

[late blade]

dunno, i didn't thought of that one tho, and that is the one that works best

[dim cradle]

maybe make sure the client understands the source and get them to sign something 🙂

[tidal whale]

Also almost every big company uses AI in one way or another (Todoist, Notion, etc), so you can't say it's "too early" when all the big players adopt the tech.

And if I make the visitors sign things before they start it's going to be a conversion killer

[late blade]

I don't care about many things, but that's just because I care about other things. I just can't care for everything all the time. No time nor energy to do that.

[dim cradle]

yet, you're here, with evidence to the contrary

[shut niche]

There's different types of AI and machine learning. You can't say AI as a blanket statement. But the "cutting edge" AI that's pushing boundaries, such as image generation, isn't reliable enough yet.
ML has been in used for years.

[tidal whale]

Not really. The fact that I got this image generated doesn't mean nobody knows how to handle the situation well by either using an alternative service or hosting their own image model.

[shut niche]

There aren't any safe image models. I use them all... trust me.

[open trench]

when the seas calm, I could use some help on a prompt

[dim cradle]

i hope you find the right mechanism to report the issue, ideally it wouldn't happen.

[late blade]

shoot, let's see what can be done

[late blade]

the mechanisms are there, you can report every single generated image.

[open trench]

I'm trying to make a surreal type image. using this prompt: "A photorealistic, surreal, landscape-oriented image featuring a floating ladle pouring a cup of hot chocolate into a mug. The steaming hot chocolate immediately freezes upon contact with the cup where it pours out onto the desolate, yellowish-green dried and cracked dirt ground. In the background, a dust storm quicky approaches and forms a Rorschach butterfly pattern. A pair of empty ice skates are dancing atop the frozen hot chocolate. The scene is encircled by broken and shattered wooden and glass hourglasses, their sand adding to the dust storm"

[dim cradle]

curious, is that part of your future roadmap, to roll your own, or stick with the vendors?

[open trench]

the results are literally all over the place

I'm tired and may not be seeing the obvious, but it may just be too darn complex.

[empty kelp]

#daily-theme message

[late blade]

try use natural light, wide format, use verbs that describe the moment, intead of immediatly frozen upon contact -> frozen as a result of interacting with the cold. use the kind of textures you want, remove approaches, just say it blends, describe the moment, not the flow of events