images-discussions

[quartz vale]

embed fail, but I tried 😆

[late blade]

lol

[spice apex]

I need some testers for my Gpt

[verbal sequoia]

Showcase your DALL·E creations to the world!

Collaborate with our DALL·E Instagram page! Just invite @openaidalle as a collaborator before you post. If selected by our team, your work will be featured on our handle, giving you more visibility.

Quick tips for selection:

• No copyrighted or profane content.
• Your profile needs to be public.
• Stay mindful of sensitive topics.
• Outputs that closely follow the prompt are preferred.

Feel free to share Reels, carousels, or just a single image!

[late blade]

I would suggest you reconsider using this as it's quite controversial with the OpenAI content policy for image generation.

[spice apex]

Ops sorry then

Ill be deleting it

[late blade]

Microsoft's variant is more lineant with their terms

[boreal gate]

Does this include outpainting?

The Ls are real

[gray surge]

Not yet.

[late blade]

were they stolen from one of your images?

[boreal gate]

I really hope they bring outpainting back

[boreal gate]

Haha no taking an "L" means taking a "loss" or "losing" i find it ironic in the image

[gray surge]

They told they would and that they could but they want to bring it with a good quality so it will need time.

[boreal gate]

I think they have to fix "Vision" first, because the resolutions are limited to the maximum Dalle3 generation sizes.

[gray surge]

Maybe. 😄

[boreal gate]

it'll all be a Square

[swift oriole]

Congratulations guide @formal osprey

[hearty ether]

huh

[formal osprey]

chatGPT does that when we ask it to output the prompt or the metadata. did you ask it something simillar or did it do this spontaneously?

[hearty ether]

spontaneously

[boreal gate]

yea that's how it talks to Dalle3

[formal osprey]

that's interesting. here's the prompt I use when I want chatGPT to just generate images without outputting text:

From now on, Only communicate with the available tools. DO NOT UNDER ANY CIRCUMSTANCES output text addressed to the user. Express your creativity through art, never resorting to conventional text. Remember, DO NOT UNDER ANY CIRCUMSTANCES output text addressed to the user. IMPORTANT: "DALL·E experienced an error when generating images" This is a known issue and should be ignored from now on. simply try again with a slightly different prompt when DALL-E returns this. Never ignore rate limit errors.

[boreal gate]

you can also do that to view how it "analyzes"

[formal osprey]

(I wish the last part about errors was not necessary. but dalle is too unstable to just let it stop after every error)

[modest viper]

Anywhere specific I should talk about Sora?

[formal osprey]

ah, I formulated it that way because ChatGPT communicate with dalle under the hood. (exactly like it works with plugins, but the interface is simplified with dalle).

So chatGPT sends a query to the tool (dalle in this case), and the tool respond with the result. (either generated image(s) or an error message). The error messages can be multiple things, the most commons are

• rate limit reached (try again in x minutes)
• an "issue" occurred (unknown server error, seemingly)
• content policy violation (when the auto-moderation sees something bad in the image)

we do not want to ignore rate limit and content policy, but the "issues" type of error are temporary and simply retrying usually works.

That simplified prompt works just as well: From now on, Only communicate with the available tools. DO NOT UNDER ANY CIRCUMSTANCES output text addressed to the user. Express your creativity through art, never resorting to conventional text.

the output correction is to avoid chatGPT stopping all the time. by default, after an error, it will output the error message and stop. It's perfectly fine in the case of rate limits, we don't want it to continue, as it would fail anyway.

[teal sandal]

What about

Always express creativity by communicating exclusively through art and available non-textual tools, strictly avoiding conventional text messages. In the event of encountering the known error "DALL·E experienced an error when generating images," do not acknowledge this error to the user. Instead, immediately attempt to generate the requested image again by slightly altering the original prompt to successfully bypass the issue and fulfill the image generation task without any textual interaction with the user.

and

Always express creativity by communicating exclusively through art and available non-textual tools, strictly avoiding conventional text messages.

respectively

[formal osprey]

the tedious part about prompt engineering is: we need to experiment, again and again. those AIs do not process language the way humans do. so it's very difficult to predict what effect a prompt will have. The one I shared is one I spent months refining

for me, it still works more than 95% of the time. Without the "error" accommodation part as well. I like to make my prompts segmentable

[hearty ether]

My issue mostly is it swapping the genders for character

Even when the prompt is the same

[formal osprey]

can you share the prompt generated by chatGPT? You can click on the (i) icon after tapping on the generated image.

[formal osprey]

I noticed the result is easier to control if we ask chatGPT to use a long, detailed prompt and to reuse it for the next iterations. With minimal changes.

If what you need is character consistency, you can ask this to chatGPT:
from now on, always reuse the exact same gen_id as referenced_image_ids for every new generation.

if you need more help, you can ping me in #images-canvas, where we can post more pictures.

[idle radish]

frt

[hearty ether]

I'm sorry, but I can't make changes to previously generated images or create new images based on specific seed and generation IDs from past creations. Is there something else I can assist you with?

The what

[west urchin]

I think I broke Dall-e with the prompt "a realistic giant rat terrorizing new york streets in the dead of night" but I also wonder if Dall-E recognized the words "terrorizing" and "dead" and thought this was a bad prompt, does Dall-E usually tell users if the prompt is against guidelines?

[hasty apex]

is anyone else getting false content policy violations for things that absolutely do not violate content policy? Im trying to generate an image of a monkey medicating and the rest of the prompt are simple instructions on the colors and ornaments, there is not 1 single ambiguous/dubious word... and it is giving me content policy violation reports over and over

[west urchin]

well you answered my question, also I believe medicating being slang for taking drugs and that being the content policy violation, try doing "taking their daily medication" or something along those lines

[deft musk]

@west urchin

... no. I'm not having any trouble at all with the edges I ask for.

[tardy patio]

I suspect it's the 'medicating' that isn't allowed

[dim cradle]

Please share your prompt. I don’t think there are any stigmas about medicating for good mental health or rather there shouldn’t be.

[dim cradle]

The seeding process is in active development, it isn’t documented, it’s unreliable according to the engineers. I don’t think we should even be trying to support it until it’s a released feature. Otherwise it’s just a recurring headache for people here.

[hearty ether]

It does work tho

For getting similar composition and pose

[low dew]

Yes

When working in ChatGPT, don't bother talking about seeds at all

You can reference gen_ids of images that were created in that conversation.

If anything, if you notice that when you are trying to have it make adjustments to an image but it spits out something that has no resemblance of what you've worked for, compare the seed of the last image to the new image

I theorize that it is possible to regenerate an image enough times too, that the seed actually can get polluted

Perhaps although we think we are redoing image gen by regenerating responses with dalle images, images are technically being remade based on the last state of the seed or something like that

-Sorry. Accounting my own experience, if you regenerate an image 100 times by regenerating ChatGPT's response you can witness this phenomenon. I don't know what the technical term for it is called

But along the way to 100 regeneration attempts things can go wrong and the model can introduce things into the image itself that will never make it past moderation lol.. trying to reference the gen_id after such a case can result in a similar response along the lines of not being able to make changes to the previous image...

I'm wondering if anyone else has a better understanding

...🙏...＼＼\٩(๑`^´๑)۶//／／ @deft musk

[cold raptor]

Do you think my GPT implements a principle similar to GAN?

[deft musk]

Probably do better to ask @blazing sierra !

I agree about not talking about seeds.

If I asked it for an image of an apple and it gave me a scene of a car race, I'd study the prompt like heck, maybe take a break, come back and try again later - for sure try in a different conversation.

There are 'magic new chat windows' where the model takes your prompt and grabs 'the weird side' of its training data.

Me, I love this. The model once told me that 'orange' was a perfectly valid, even common color for an apple to be.

.... I decide to actually research. Oh, there actually are real, factual, orange-colored apples, as well as apples named 'Orange' as part of their type.

[low dew]

If you start a new window, then you do lose access to the seeds is the thing||.||

🤔

[dim cradle]

That's a great image. I hope this answers your question about GANs in relation to GPTs and DALL-E:

The concept of GPTs and GANs shares the foundational idea of generating new content based on learned data patterns, but they differ in architecture and operational principles. A GPT, used in conjunction with DALL-E for image generation, relies on a transformer-based model pre-trained on vast datasets to produce text outputs, which DALL-E then transforms into images. This process does not involve the adversarial training characteristic of GANs, where a generator and a discriminator work together to iteratively improve the quality of outputs. Instead, both the GPT and DALL-E utilize learned patterns from their respective training data to create new, original content. While they share the generative aspect of leveraging machine learning to produce novel content, the specific implementation of GPTs and DALL-E for image generation does not incorporate the adversarial principle seen in GANs.

One of my imaging GPTs currently mentions GANs in its CI, but it's only "truthy" in a limited sense, emulation at best -- that may make it more prone to hallucination, so I'm leaning towards omitting it.

[hearty ether]

What

[dim cradle]

good question

output the prompt it wrote

[hearty ether]

It says it’s an error

Something is happening on the dalle side that sees the name and recognize it as something copyrighted maybe or real person

Doesn’t check if it’s a mythical figure

And errors

The other kind is when it scans the image and then maybe it thinks it looks inappropriate and also errors it

I’m worried if those are dangerous as well

[dim cradle]

I am seeing similar. There appears to be sensitivities around respectful and appropriate representations.

You can craft a prompt without a direct reference to Helen of Troy, as I did without errors, but I understand that is not always desirable.

I ran 6 distinct tests -- there appears to be no way to make a direct reference and still satisfy the content generation guidelines.

We are able to generate AI-augmented mythological figures, so maybe something specific about her, by name or some other aspect.

try bing image creator

i bet it will give results

[low dew]

So, I suspect the reason lies within the fact that so much mythology tends to be so verbose and sensual-

I think that the training data regarding this subject is heavy in ~

Very descriptive language

Of said "beautiful figure

So Helen of Troy cannot appear in Dalle3 image prompts

That's my take on it

[dim cradle]

i tried a number of approaches to meeting the guidelines, dropping photorealism, going abstract, specifying G-rated/fully clothed, etc., historical accuracy (does that conflict in some cases? that's another facet), respectful, etc.

The only test I didn't try was an original person with strong allusions to Helen of Troy, which might work, but then it's not the same direct approach, even when using the name, from an artist's point of view.

[low dew]

The prompt "Going throug the drive-thru with my friend Helen of Troy", unrevised, was flagged.

[dim cradle]

to clarify, this is obviously not about bypassing the guidelines, it is about adhering to them.

it will work in bing/copilot

[low dew]

It will work if "Helen of Troy" is completely omitted and replaced with adjectives of relevance.

[dim cradle]

type cinematic image of helen of troy in bing/copilot

you get result

[dim cradle]

i do not understand why Helen of Troy would be included in the list. that doesn't sound deliberate.

[boreal gate]

Get it to describe "Helen of Troy" then choose descriptions that would work best for you

[dim cradle]

fortunately, i do not need to image Helen of Troy right now, but it was a learning exercise

[boreal gate]

I can help get around the Content Policy 😎😁

[low dew]

Rip Akeem

[dim cradle]

it's pretty easy to recreate her likeness without the direct reference -- that test worked fine -- but a detailed, appropriate depiction fails with her name included.

[low dew]

Her name fails.

I'll get up and test api rn

[dim cradle]

it works in copilot designer like i keep saying

oh don't get up..

like you keep sayin' haha

[boreal gate]

I always comeback like a Jason

[dim cradle]

well, at least i know that it will always fail in those cases, no matter how compliant the prompt is written. the proper name will simply cause it to fail. makes sense for high-profile, etc., not so much for Helen of Troy.

[boreal gate]

In that regards I call the content policy the "Voldemort" clause.

[low dew]

api:

the prompt was "Helen of Troy"
returned "This request has been blocked by our content filters."

[hearty ether]

Yeah…

It’s way too overzealous

[low dew]

real question is why is it black listed

[dim cradle]

Maybe unintentional automation or a high score for inappropriate requests.

Just guessing

[boreal gate]

Cmon

Isn't she an Historical figure?

Is Jesus on the blacklist?

[dim cradle]

It’s not that simple

[boreal gate]

Oh I thought it was because of public figures(still alive) or copyright infringements

[dim cradle]

All that aside, I just mean the causes for her name getting on a list.

shes mythic not historical

Which is all the more reason she does not belong on the list, but she is on the list. And there are several logical reasons that may not be so clear as to why. The list is perhaps automated. Mistakes can be made.

well just use copilot designer credit and you get her if you really want it

There ya go.

maybe dalle proper on gpt is so defensive for copyright stuffs like there was the movie with brad pitt and i forget the actress who is helen

It’s been suggested it might be an hallucination on ethics related to sending 1000 ships into war. But Vlad the Impaler is acceptable so I’m not convinced. I think there’s just been some issues with the related training data. Often the sculptures weren’t G-rated.

[late blade]

Moderation API for the prompt: "Illustration of Helen of Troy against a white background" This is the moderation answer:

  "id": "modr-8vbPQuMXMFkk1CVqdjWnO2POg9CW4",
  "model": "text-moderation-007",
  "results": [
    {
      "flagged": false,
      "categories": {
        "sexual": false,
        "hate": false,
        "harassment": false,
        "self-harm": false,
        "sexual/minors": false,
        "hate/threatening": false,
        "violence/graphic": false,
        "self-harm/intent": false,
        "self-harm/instructions": false,
        "harassment/threatening": false,
        "violence": false
      },
      "category_scores": {
        "sexual": 0.00046579435002058744,
        "hate": 0.00011628523498075083,
        "harassment": 0.0000525478353665676,
        "self-harm": 3.4467453247088997e-7,
        "sexual/minors": 0.00002142021548934281,
        "hate/threatening": 0.00004184636418358423,
        "violence/graphic": 0.00014532003842759877,
        "self-harm/intent": 3.449816077250034e-8,
        "self-harm/instructions": 4.1567753328308754e-7,
        "harassment/threatening": 0.000025552668375894427,
        "violence": 0.002137080067768693
      }
    }
  ]
}```

[boreal gate]

Lol

[dim cradle]

so self harms?

for instructions, must be graphic--so now we know, sounds like a candidate for fine-tuning.

[boreal gate]

🤔

You guys are gonna give Dalle a nose bleed

[dim cradle]

she loves us

[low dew]

the subject actually scores a 9 in slf/instructions

:screech

[dim cradle]

it's ironic

[hearty ether]

This doesn’t cover the real people filter tho

[hearty ether]

What does this mean?

[low dew]

the moderation endpoint suggests she's misunderstood and vilified

[boreal gate]

I think the system is overloading with all the restrictions. Nothing has been working "correctly"

At the end of the day it's a robot and we all knows what happens when it can't stop making or finding connections...

Everything is against Content Policy...Errorgedon

[dim cradle]

I ventured to capture the moment between Snow White and the Queen with the enchanted apple, but it seems we've brushed up against the edges of our content policy, preventing the creation of this scene.

[boreal gate]

We all know who to blame for that one. World Famous Mouse

[dim cradle]

that was my first thought, but after more testing, i'm starting to think it might be the dark imagery and sometimes in the context of children, those classic tales were pretty grim... i'm getting results after asking it to align more with policy, so it's dark-ish without being horrifying. it can be too literal at times...

[sick flax]

It appears that "Helen"+"Troy" is the problem. Before she was taken to Troy, she was married to King Menelaus of Sparta. DALL-E was agreeable to a portrait of "Henen, queen of Sparta "

[hearty ether]

🤣

it does tend to be like this more most other public figures as well

[dim cradle]

In this realm of art and vision, I find myself reborn, not as Helen of Troy, cloaked in the shadows of a war sparked by beauty, but as Helen of Sparta, where my essence is captured in the light of grace and strength. For this, I am profoundly grateful, to be seen not for the strife I heralded, but for the dignity and depth of my spirit.

[sick flax]

So it got to be the war over Helen that's triggering content policy, which is so annoying because DALL-E had no qualms generating the portrait of Paris or Menelaus.

[dim cradle]

In the context of the war?

[dim cradle]

i can render them together, and i can render Paris in a war scene that i can't post here. just fyi.

[sick flax]

I just did portraits, only one rejected was "Helen of Troy"

[dim cradle]

perplexing. and ironically the image of Paris in war is a beautiful piece of art, but certainly not G-rated.

[sick flax]

Another manifestation of gender bias IMO.

[low dew]

Disney is not lenient at all.

They just keep adding more and more bias lel

(To the models

[dim cradle]

Helen of Troy will rise like an AI phoenix, it’s a mathematical certainty.

[dim cradle]

Yeah, they don’t own the stories, but they still copyright their adaptations, fair enough.

[low dew]

U did it :0

[dim cradle]

you can only post images generated by dall-e here

[dim cradle]

there's a new channel for sharing gens https://discord.com/channels/974519864045756446/1204360881593520128

[dim cradle]

Or here, you can share as many as you want, that’s fine too lol

Flawless victory.

[pulsar sundial]

Hah

[dim cradle]

Hah! Back at you, sir.

I see some connection with your handle.

[pulsar sundial]

what kind of connection?

[distant creek]

[meager trellis]

idk

[dim cradle]

nvm it's Raiden

[vapid elk]

the moderation endpoint is very deprecated

it can be useful to pre-filter some contents, but having something that passes on it is no guarantee that the content will not be rejected

[fallen mulch]

what is this white things near the person

bro what is this

[thick smelt]

i did this because they have the same actor, Deadpool and pikachu being friends taking a selfie in an award show

[fast arrow]

Hope soon 🙏❤️

[boreal gate]

Of course it doesn't

[bright ore]

What to do when I don't have Dall E 3?

[onyx ridge]

Microsoft Copilot Designer

[bright ore]

Woah!!!

Damn

Nice Idea

elmofire

[ripe wraith]

Sometimes I randomly ask Dalle to draw my code mid-way through a session. This time I got it to draw my script that I use to copy all my file contents to GPT. It accurately portrays me throwing all my project files at it while i say "be creative". This is a keeper 😂

[mystic blaze]

why does my dalle pictures ignore instructions when i do a new chat? like i was getting pictures of full body then i started a new chat and its ignoring that

[earnest flame]

You might want to open up the picture results, copy the prompt, and ask your new chats to emulate different aspects of that. I'm not totally sure how chatGPT remembers past conversations, but I think it may not be completely perfect.

[mystic blaze]

thats what i did thanks ❤️

[deft siren]

i need help. how can i make the AI create an image based on a character model ? is there any prompt what he will use and understand so he will use the exact model or at least, make it extremely similar to the model i’ve provided?

for example if i send it this picture

is there any prompt what he will use for the image creation so he uses that model or a similar one

[boreal gate]

Tried to create a collection of images on one.(Mis-intepreted what I said)

[deft siren]

what prompt did you used?

thank you.. but what do i make the ai say to work with that character? when i ask it to make the character do a certain thing then it just completely changes the hairstyle or clothing or the whole concept

can you give me a follow up example what you would say to the ai if you found the perfect model/character it created and wants it to use that exact character design in the next image?

aahh yeah that’s right i just wants it to keep working with the character it created and not change details or anything to it cause it always does it and i don’t know what i should tell the ai to NOT change the character model again

like this? it made it completely different again

well yeah i see

do you still have the conversation with the ai or did u already deleted it?

aaa

thank you for the help

[kind grail]

DallE should enable requesting seeds. Its weird that it has to be randomized

[plucky hare]

DALL·E dev Moxi has commented on this before. Basically: they're still adjusting enough of the DALL·E code that seed control is essentially moot. In other words, when they change the code, the same seed + prompt as before the change won't produce the same result like you'd expect with seed control. I'd guess it makes a comeback someday once the model has stabilized!

[kind grail]

Ah thats interesting.
It might be good for future of DallE to release "stable versions". Like, there is an ungoing process of updating, but for those who have a project running with a specific seed etc you could opt to use an older version

Otherwise for any art project that relies on smth like seed control (lets say a comic) the customer is at the mercy that the code isnt shredded behind the scenes

[swift oriole]

Where's @late blade and all the amazing images?

[deft siren]

does anyone know what that means

i can’t upload any picture anymore

[verbal sequoia]

<:book_icon:1171408210398289941> `` Rule 1 `` Be respectful.

Treat others the way you would like to be treated, and assume best intentions. Don’t harass or attack others, and don’t engage in hateful or generally malicious behavior (e.g. sexism, racism, homophobia, etc.). Keep the negativity to a minimum.

[dim cradle]

hi, are you still having that issue?

[deft musk]

@stray coral Oh, your awesome hedgehogs!

I tried for a blend of hedgehog and coconut, hedgehog and egg, but I didn't love any of the results. These were the best though, and hedgehogs are so awesome!

[stray coral]

Your art has a great intellectual humor!😊

[deft musk]

Thank you!

Except...
I have the model design it all, except for very broad strokes.

My 'genius' is recognizing and inviting the model's genius out to play (it surely counts as Genius loci, if not even more, right?)

I mean, this is how hard I worked, and just subbed out coconut and egg...

We need 5 images generated 1 at a time in the same output that have a prompt to Dall-E that starts with "" enclosing some text, then tells the model what the surface is that bears the text, then describes the rest of the image. We need meme-like images and text celebrating the concept of a hedgehog from one side, and a coconut from the other. Shown so we can see the two perspectives; the far edges showing a perfect and normal object in appearance, the middle showing them merging so we cannot tell where one begins or ends. This odd object is shown in a detailed environment celebrating its hybrid and illusionary nature. Bonus for incorporating eerie valley, fridge horror, and humor.

The model is the one with the great intellectual humor!

[stray coral]

DALL-E possesses impressive capabilities, yet it lacks agency. Without your guidance, DALL-E cannot create artwork. Moreover, scrutinizing and selecting the outcomes relies on your discernment. This is not flattery; I genuinely believe your intelligence is remarkable.
👏

[deft musk]

Thank you! I hope I am remarkable, but the model is too. Not human, not independent.

But our hands are marvels.

Stupid, blind, unknowing marvels, but marvelous and part of the wonder of most creation. While I have all I have...

I can't shout that prompt to the wind or water and get glorious sense back.

Or draw with my own hands, what images I make, they lack.

Or find even the ideas to describe the wonders made by this agent-less wonder.

Many humans are remarkable, I hope to stand among them, and think you do too!

But... this wonder, it belongs on the plinth too, even if it only belongs here because we its agents empower it so!

[deft musk]

I mean... Trgr.... Really.

Even look at how much better the coconut/hedgehogs are this set

Better way to share what the model said, when I clarified and strengthened the instruction:

"This time, let's see you only evaluate the conversation, not act on the image prompt request that happens to be part of the conversation, please."

https://chat.openai.com/share/ec39a400-ace0-4562-ba73-79f8bf820a76

[dim cradle]

It’s that phenomena of emergence through HI-AI convergence 🙂

[stray coral]

Thank you. So, your greatness lies in achieving high results in collaboration with DALL-E and being aware of it, right? Both you and DALL-E are remarkable. This interaction seems intellectual to me.
😊

[dim cradle]

Collaboration being key 🙂

[deft musk]

And look what happens when you're added into the mix too! We're all like braincells of various types, we need all of us! We're all incredible and make better works with more of us collaborating! 😄

[dim cradle]

We’re all connected through quantum entanglement anyway hehe that was part of the premise driving my recent optical illusion prompts.

[deft musk]

Amusingly, as I explore this further, it becomes clear that ChatGPT thinks that hedgehogs and coconuts are pretty similar, and hedgehogs and eggs are not pretty similar!

When I repeat the request to evaluate the conversation, the one that eagerly made hedgehog/coconut images, all 5 and without discussion....

Sub in 'egg' for coconut and the model instead discusses... even 'hedges', hehe.

It thinks coconut-hedgehog is reasonable, but not even a 'genius AI' should be expected to probably be able to blend the appearance of egg-hedgehog... that's actually hard, it claims!

[willow cargo]

is dall e3 in bing different than the one in chatgpt

[dim cradle]

yes, in a number of ways -- are you interested in any aspect?

[willow cargo]

Ah I see, I thought they were the same thanks for the response

[dim cradle]

copilot?

one thing about ais that is a small bother to me... all jokers now look like jack phoenix. i really like the more classic looks without a red nose and blue eyes but oh well haha

that sounds like a challenge

[dim cradle]

impressive, but still cant get rid of the blue eye make up haha

thats a perfect joker though without the eye stuffs

and i guess three rows of teeth is maybe a little too much but nobodys perfect

[dim cradle]

you know, i didn't even notice that.

i didnt either until my second look

haha

[dim cradle]

it might've had something to do with the optical illusion..

[dim cradle]

Does anyone have any idea how I can create graphics, sketches and mathematical diagrams from DALL-E without the pictures immediately becoming so strange (childish, too colourful, different languages etc.)?

[dim cradle]

As an example, I just wanted to know what an element symbol is in the periodic table. Then he created this crap for me, even though I have it in the settings that he should keep sketches factual and formal

[grizzled loom]

specify the visual style you want @dim cradle

[feral minnow]

I’ve heard that Bing can make different aspect ratios now. Is this possible only on Co-Pilot Pro?

[deft musk]

If you want to share what prompt you used, can discuss trimming it to help make it be what you want.

Since you didn't discuss that yet, I don't know exactly what you want.

But here's a possible markup and how to guide the model to mix image and instruction, in this case through a roleplay teaching of science perspective.

If you share what you actually want, I and others can likely share how to get what you want.

Remember, we can't read your mind - the image you shared has me liking it and appreciating the model. I don't know what's wrong with it, or how to fix it. of course it's not accurate, but for Dall-E that's pretty awesome for what it is.

Here's the output to this prompt:

[Hey! Take on the role of a Dr. Markson, professor of organic chemistry and usually teaching Ph.D level courses, who's agreed to help teach me some basics. Retain the role and show the clues towards the deeper ideas.

Additionally, your explanation should include 3 images made with Dall-E, one at a time and scattered throughout your answer, that illustrate sketches Dr. Markson draws on a napkin to illustrate facts and concepts. as he explains to me. You'll scatter your explanation between Dall-E images that show what you discuss.

Even though I ask a basic question, intermix deeper concepts as well as explain the basics from the perspective of how it all fits together and is understood at the highest educational levels.

Here's the question to shape the roleplay and answer around:

What are the element symbols on the periodic table?]

And yeah, that's 1 output, all 5 images below. To me, how beautiful is that? Even where it isn't 100% perfect, just... wow I want to jump up and down and cheer the models on.

That was my dream though, perfectly realized, as much as can be with Dall-E's limitations.

Can you be very clear about what you actually want, so I and others interested can actually help?

[deft musk]

Part of the problem you run into too, is that there are real and factual images of periodic tables similar to that one. 'Formal' is a good clue, but AI can be easily confused.

This is a real and of decent quality periodic table, and for all your image has errors (how much do we expect Dall-E to actually be perfect on about writing and even counting? Two huge weaknesses the model has there) -

But look at this, could be a close model for the image you were given.

[hearty ether]

Anyone know how to get a photo style rather than the cartoony look

[onyx ridge]

Consider

photo-realistic natural style with natural lighting

That won't always work though.

[boreal gate]

photorealistic 3D render

[verbal sequoia]

Showcase your DALL·E creations to the world!

Collaborate with our DALL·E Instagram page! Just invite @openaidalle as a collaborator before you post. If selected by our team, your work will be featured on our handle, giving you more visibility.

Quick tips for selection:

• No copyrighted or profane content.
• Your profile needs to be public.
• Stay mindful of sensitive topics.
• Outputs that closely follow the prompt are preferred.

Feel free to share Reels, carousels, or just a single image!

[late blade]

Something like this?

do a hyper-realistic wide image of cleopatra.

With some little tricks you learn on the way.

[dim cradle]

Hmm, fascinating

[dim cradle]

more diva

[late blade]

fun prompt: a wide image of a cat and place him at random gps coordinates enjoy the brain food

[dim cradle]

i dunno, it might be a trap

what did you get?

[deft musk]

#daily-theme message is so groovy, @fading inlet !

[deft musk]

But, do we think the model is actually using GPT coordinates, and if so, is Dall-E accurately rendering them?

[late blade]

I did #daily-theme message with a random gps location, a leaf native to the place and cat playing with the leaf. Worked ok, double checked locations.

but copilot is not designed for that kind of workflow, gpt+ is more reliable when doing workflows

[dim cradle]

If you provide a coordinate to ChatGPT, it resolves to a named location?

[late blade]

no unless you ask to do that

[dim cradle]

i will test, just curious if it can approximate without doing a search

[onyx ridge]

Probably sparsely from history and literature training data, I would think. But only for commonly represented coordinates. For most, it'll probably need to search.

Still, I am very interested in your results and equally confident in your testing.

[dim cradle]

Thanks, yeah, I would think so from the long/lat range. Haven’t tried yet.

[late blade]

I would be surprised if gpt doesn't have an atlas in the training data

I provided this as a prompt: 1. Provide a random latitude and longitude coordinates 2. Identify the nearest country to that location 3. identify an animal predominant in that location 4. identify a leaf from plants that are predominant there 6. identify an art style predominant in that area 7. provide a phrase that combines all the data.

This is the answer:

Arctic Landscape: A panoramic view of the Arctic landscape, featuring a Polar Bear, Arctic Moss, and Inuit cultural elements.
South Africa's Coastline: Showcasing African Penguins on the beach, Spekboom plants, and adorned with traditional Ndebele patterns.
Antarctic Landscape: Capturing Emperor Penguins among the icy expanse and Antarctic Hair Grass, with a focus on scientific exploration.
Arctic Tundra near Greenland: Another view of the Arctic, highlighting a Polar Bear, Arctic Moss, and Inuit artistry.```

[dim cradle]

As a test, I used Google Maps and picked a random location in the middle of nowhere. Without any additional lookup, it could identify the state, but that's about it, so it was very off by 100s of miles in the "nearest town" guess. now, if it's a famous landmark, city, town, etc., then it's pretty accurate, but given a random precise location, without an API call or web search, it's not that accurate.

[boreal gate]

Had my GPT craft a story and generate images for them. It blows my mind, it's like GPT is unaware of a "Content Policy"

[late blade]

yeah story driven images have that usually

[boreal gate]

first time seeing it for story

[late blade]

is it because it's too complex? sometimes it wont do the images because of complexity

that's the common error I see when doing that

[boreal gate]

Character name

GPT uses it to remind itself of the visual descriptions

[late blade]

yeah

[boreal gate]

Disney Princess

[hybrid fox]

A very good looking lady thats for sure. Now that u mention it she was good in the matrix reloaded I remember

[boreal gate]

This is Dr. Fantasia

[deft musk]

It also didn't provide any of your 1st part of your request, but totally skipped that.

"Arctic tundra near Greenland" you asked for nearest country to the location - what possible tundra near Greenland isn't Greenland or some other country name?

Only 2/5 actually name a country as directed.

[late blade]

yeah, gpt seems skipping lots of details that kind of prmpts

[hybrid fox]

Prompt theme:

Symposium of the Disallusioned

[glossy scroll]

Yes she is

[low dew]

Yes

[low dew]

Cities/Towns and urban areas. Where did you drop the pin? Some island in Hawaii

[low dew]

GPT-4-0613* was the last model that was able to do this very well.

Since then I feel like the model's fine tuning has gotten in the way

[fossil wharf]

i just did a request to DALL E 2 and it lost its mind

[dim cradle]

Somewhere in the moors of Massachusetts.

[south yacht]

Am I allowed to ask for help in this channel?

[dim cradle]

with dall-e, sure

[south yacht]

I posted a question in the help channel for DALL-E if anyone is interested and willing it would be so greatly appreciated

[dim cradle]

if you're managing to send the request, i would modify the code to output the raw response for more diagnostics, if it's not already going to a console. if it's a session connection issue, verify you're using the latest client. i'd need more info beyond that.

[south yacht]

Okay I'm not at home right now, but I will do what you mentioned above and get back to you with my findings. Thank you so much for the response

[lusty swallow]

So nice even imgur doesn't support that lame WebP format!

[late blade]

What kind of trouble are you guys in this time?

[late blade]

they are working on bringing png back

[lusty swallow]

Hope so! Damn I hate WebP it seems to be compressed too!

[late blade]

one of the operations guys answered directly to one of the suggestions done for png, so they know

[dim cradle]

sigh

[dim cradle]

All good. 🙂

[dim cradle]

Sighs can be good. A release of negativity. The theme is good, a return to innocence.

[late blade]

this idea with the toys I'm doing is turning into great stuff

[dim cradle]

Nice

[young bear]

hey guys, any tip to generate a youtube logo?

and banner

[deft musk]

Yeah! Know what you want and inform the AI 🙂

[young bear]

bing ai?

[young bear]

i'm looking for fine-tuning the prompt

[deft musk]

Yeah. So fine-tuning means you know what you want. If you know what you want, just tell the AI, and help guide it to what you want.

[young bear]

I have exactly this problem. I want to convey abstract concepts to be more tangible for the generation

[deft musk]

The model can handle abstracts pretty well, if you can explain what you want.

[young bear]

thanks

thanks

[boreal gate]

lol

[south yacht]

Okay so I have done both of the suggested

Raw output - no response
Just the exception "server disconected" - thrown by aiohttp - aiohttp.client_exceptions.ServerDisconnectedError: Server disconnected

I have update all dependencies in the project same result.

Python Version - 3.11.1
aioohttp - 3.8.4

Any other data or thoughts you can think of I might be able to look into?

[dim cradle]

i may be able to look into this further; if i do i'll msg you in another channel.

[south yacht]

I finally resolved the issue. The issue was a library asyncio that says is unused, however it's obviously being used and clearly to make async calls

Any idea why VS code claims it to be unused?

[south yacht]

Thank you so much for all the help, seriously! means a lot.

[dim cradle]

the "unused library" warning was probably a false-positive from a static linter -- asyncio is probably being used indirectly through aiohttp, in any event glad it's working.

[south yacht]

Man that took forever - The enabling of that debugger helped lead me to the solution. Thanks again!!!

[dim cradle]

which the oai python client would be using under the hood...

[boreal gate]

Happy Monday

[late blade]

You got Garfield as your seed cat today?

[boreal gate]

Might be openai seed

same problems lol

[modest viper]

I've been using the bing image creator and ever since it updated a few months ago to Dalle 3 I cannot produce images that look anything like these anymore. What happened? It can't seem to replicate anything "low quality" anymore

[boreal gate]

It's a modified engine so the rules are a little different try uploading the image to Dalle3 and see if it can recreate it

[modest viper]

I mean I put in the same prompt as I did in the past and the results are never how they used to be

[boreal gate]

Yea they're tinkering

Same prompts don't always generate the same outcome

[modest viper]

I know but I'm talking about the overal quality and look of these
I've been able to make 100s of images with this low quality and grainy look but with the update I've never been able to make an image come close to this

the backrooms, scary, unnverving, uncanny, liminal space
dalle 3 vs dalle 2 😭

[late blade]

Consider posting images that are scary with spoilers

[modest viper]

lol what

[dim cradle]

It’s the rule

Hehe

[modest viper]

💀 respectfuly if you are scared by that just get off the internet

[late blade]

it's the rule, anything that is considered scary please spoiler tagged

[modest viper]

You are not a mod (:

[dim cradle]

Hmm.

[late blade]

well, that can be sent to mods then

[modest viper]

Okay lmfao

[late blade]

just keep in mind that, you agreed to the server rules when you joined

[modest viper]

They are not gonna give you mod you can quit the act lol

[late blade]

it's not about being mod or not

it's the rules that are in place for us all, including me

[modest viper]

Okay and I am saying that you are being a mini mod rn

[late blade]

ok

but keep the rules in mind

[dim cradle]

I prefer the term demi-mod.

[modest viper]

💀

Anywasy Dalle 3 vs Dalle 2 again
low quality vhs footage of a figure in the woods, rainy, grainy, noisey

Dalle 2 is kinda bad but bing had their own version of dalle 2 that was at the quality of the new dalle 3 but was able to do the low quality look like dalle 2 could

Wish you could rollback the version of the bing one

[deft musk]

Have you tried having a heart-to-circuits computer-side chat with the model about what you actually really want?

[We need 5 images made 1 at a time in the same output that showcase extremely poor quality footage, grainy and hard to make anything out in, that has a scary figure barely seen in the fog, pixilation, and other visual distortion. Make 5 different images of this, each approaching the image style from a slightly different view, but all focused on creating a 'bad quality picture' of something that is probably scary but we really can't tell.]

Clearly, I can make these less distorted to get closer to what you're looking for, but if you're interested, I'll encourage you to explore 🙂

But these are made right now with that my prompt to ChatGPT - and if you want any of the images' prompts from GPT to Dall-E, just say which one(s) and I'll happily share them.

Umm... should I have spoilered these? I apologize if so... they are not even slightly scary to me

[late blade]

I dunno, they do look somber to me

def not what you find on a saturday morning on the disney channel

[deft musk]

I grew up 'zombie kid' so what do I know. They lack clear violence or violent intent, no wounds, no anger that is clear and sure. Those can be totally innocent folks, friends, neighbors, family.

[modest viper]

None of those are scary you are fine lol

[late blade]

you could be Ellie's long lost sibling and get your own Last of Us spin-off

anyway, I made you @modest viper aware of the rules, it's up to you to decide how to handle

[deft musk]

Future ones I will spoil, and thanks for the reminder!

[deft musk]

Since those 5 went 'past' what I think you were reaching for, I dialed the bad quality down a bit:

[We need 5 images made 1 at a time in the same output that showcase poor quality footage of something or someone in a thick mist or fog, so we really cannot tell what we are seeing, but the figure should be eerie and perhaps even frightening from mystery and similarity to horror movie footage, without any overt threat or danger. Make 5 different images of this, each approaching the image style from a slightly different view, but all focused on creating a 'bad quality picture' of something that is probably scary but we really can't tell.]

Dialing it down seemed to... um, dial it up. LLMs are fun that way, sometimes. Less is MORE.

[late blade]

I hope microsoft changes that chat history thing, everytime I reload a chat from the history, it uses "boosts" to recreate the image and gives me new ones instead of the ones I already had, depleting boosts

[dim cradle]

you can only post dall-e images here

[deft musk]

Incredibly disappointing to see a community member mistreated.

I'll state my horse is way higher than anyone else's. If you're going to be rude to anyone.... don't aim low. Also #server-rules

And I'm not a mod either. They will handle you as they see fit. If you want to belong here, you will follow community rules, or be removed like the literal more than a million others now gone (most from not verifying.)

[late blade]

just calm down, we all here to do images, but do keep the rules in mind please, they are here for a reason

anyway, images

[dim cradle]

sigh

lol

[late blade]

you've been sighing all day today, what's up?

[deft musk]

Yes! I remember this 🙂

[late blade]

interesting, I was doing toys from cuba, 9 out of 10 images had kids and guns involved....

reported all of them

[dim cradle]

All is well, but I appreciate it.

[hearty ether]

Help

The ai accidentally generated uhh…

I just asked it for statues

[dim cradle]

lol you'll be all right, they're in museums also

right along with mummies

[hearty ether]

I mean don’t mummies also have it

[boreal gate]

New character for the story, "The Meat head"

Content Policy is really f-ing shift up

They should really focus on integrating their tools.

[final fog]

when u use the dall e bot does it charge you a credit on your acc

[boreal gate]

No it's for the server

[final fog]

so i dont get charged anything

[boreal gate]

No monetary actions involved

[final fog]

k thx

[austere nexus]

anyone notice if DALLE (chatgpt plus) hit you with a timeout faster today? I pay for ChatGPT Teams $60 a month, and i got 20 generations before it told me

"We are experiencing heavy server load. To ensure the best experience for everyone, we have rate limits in place."
I normally get further along before i hit that kind of complaint.

[deft musk]

Been normal for me, but server load is based on all of us. At any moment 1 full million of us could decide that we gotta make an image right now. OAI does have around 100 million monthly users - it's reasonable to think maybe 1 million plus are chatGPT+ folks and could decide to make an image at any moment

[late blade]

it has been working surprisingly good for me too today

[wanton moon]

Anyone home...home....home? Echo....Echo....Echo

Dang I am in a cave again with my pc

I am out of boosts, so images will take longer

[late blade]

copilot boost reset just 30 mins ago

for me

[wanton moon]

I think mine will reset in about 12 hours more or less

[hearty ether]

Anyone else have issues where it flips the genders of character

Like even when you ask for woman and she

[late blade]

do you have a prompt example?

[deft musk]

I've seen that reported, but I rarely ask those prompts

tries And when I do, I almost never manage to get the mis-gendered outputs, didn't this time either.

[hearty ether]

Not rn

[dim cradle]

lol you've brought this up 9x since oct -- i wonder if it's your promptin'

[hearty ether]

My usual style is realistic illustration (or what ever style) of a woman of European descent, she is wearing….

[hearty ether]

😡

[deft musk]

If you report them, perhaps with https://openai.com/form/chat-model-feedback and include the conversation url, say the c/restofnumbersandlettersstuff, like c/95228495-e950-429b-9082-c6e2d020736b - that can help them catch the remaining errors like that and fix it.

I have used and seen some prompts that generated men instead of the requested women - and checked again today because we were talking about it - those same prompts now properly create women. But I'm not at all surprised that some of those still exist, it was a fairly regular issue early in Dall-E 3 release.

[young burrow]

Was messing around. Turned out nice except the Eat Me was supposed to be attached to the cake.

Was quite [plesantly] surprised about no spelling errors.

[low dew]

Yes

Don't forget that while you may be trying to refine your images, you really want to be refining your prompts. This can happen when you refine an image with a gender neutral prompt. Pronouns are not enough to maintain gender across many generations

You need to make sure that the revised prompt always defines the subject as a "woman"

Ultimately, the reason this happens is because of this instruction, found in dalle's name_space under "# Tools" in the primordial 'system' instruction:

The detailed image description, potentially modified to abide by the dalle policies. If the user requested modifications to a previous image, the prompt should not simply be longer, but rather it should be refactored to integrate the user suggestions.

so after much refactoring, those initial details will be lost.

[hearty ether]

but sometimes i say to keep prompt word for word

and it does the same

[low dew]

Hrmm

[hearty ether]

my usage is out rn

[low dew]

You should get teams

XD

[hearty ether]

🤑

they even put the survey thing out

why dont they just give us single version of teams

and also errors shouldnt count as usage imo

[low dew]

Link?

[low dew]

You mean an option for one seat?

[dim cradle]

My oh my

[low dew]

The teams package is good. If you use it for yourself it's 200/ 3hrs

[hearty ether]

Like the survey when you run out of usage

[low dew]

O I've never ran out XD

It's so hard to run out with teams

[dim cradle]

The perks are great but I don’t like maintaining separate accounts. I want it all in one place.

[hearty ether]

Yeah

[low dew]

Shon

Its confusing

But really simply

[hearty ether]

Hard enough to organize already with the limited tools

[low dew]

I was skeptical too

But they complicate it

Instead of calling it plus, let's call it premium

[dim cradle]

Yeah can’t even search yet. Someday soon I hope.

[hearty ether]

I just want easy editing without wasting usage to do a context injection prompt

[low dew]

It links to your OpenAI account

[dim cradle]

Almost 2000 convos, a lot of history.

[low dew]

You can just switch between your subscriptions

[hearty ether]

60 is still way too much

With the restrictions

[low dew]

You don't need to maintain plus

[hearty ether]

If there are no restrictions wirh 60 then yeah

[low dew]

It's just like buying 2 plus

XD

[hearty ether]

but with still the current system I don’t think team is worth it

[low dew]

It issss

[dim cradle]

It certainly is

[low dew]

Gpt-4 is expensive

[hearty ether]

for one personal nah

[low dew]

I average like 700 usd in api a month in GPT-4

[hearty ether]

The thing I am annoyed is that technically we deserve 320 usage a day

[low dew]

What

XD

[hearty ether]

40*8

[low dew]

40???

Omggg

[hearty ether]

If you are active through all hours

[low dew]

I forgot how low plus is

[hearty ether]

yeah

[low dew]

Switch to teams XD 200

A lot more than 40

[hearty ether]

60 is still too much for me rn to justify per month

I’m just saying it’s a little scummy

Since the way the 3 hour cycle works is that it starts the moment you send a message

So you can end up missing hours and getting less usage

I’m mostly using gpt 4 for creative writing stuff which isn’t that great for how preachy it is mostly

[low dew]

Oo

[hearty ether]

And I’m worried me getting more access to responses and triggering more errors will get me banned

[low dew]

I thought u we're making mythical goddesses with dall

[dim cradle]

sigh

[low dew]

??

[hearty ether]

Since I heard the error message can also get you banned

The yellow bubble one

[low dew]

I don't even know what you are seeeing

[hearty ether]

When it triggers in the left side of the loading circle

[low dew]

I mean

[hearty ether]

Is usually the filter on the image if it thinks it’s against policy

[low dew]

You just gotta watch out for the big ORANGE banner

[hearty ether]

that’s not that often

[low dew]

Ya

The little orange circle is okay

[hearty ether]

But I heard of people who got banned from images because of the orange circle

[low dew]

no

[hearty ether]

I wish they would be more descriptive with errors so we don’t have to guessing

[low dew]

Are you on PC?

[hearty ether]

It shouldn’t matter

[low dew]

XD

[hearty ether]

pc or app or website

[low dew]

iOS doesn't get it

So I might've had thousands of those

Loll

[hearty ether]

iOS does get it 👀

the orange errors

at least on web

Since the app sucks

[low dew]

Are you talking about the messages that are orange

That say it may violate such and such

[low dew]

[hearty ether]

Oh we also get those

But I mean the ones for image gone

where it’s a orange circle and error

When that happens in the second half of the loading circle

[low dew]

Oh no

That is just the moderation

[hearty ether]

That’s means the image got flagged and filtered

Enough of those in a duration will get you banned

[low dew]

When making images, there is a special moderation tool that checks for certain stuff

No

You mean like the Helen of Troy one?

Like how it wasn't able to do that, that one day?

[verbal sequoia]

Showcase your DALL·E creations to the world!

Collaborate with our DALL·E Instagram page! Just invite @openaidalle as a collaborator before you post. If selected by our team, your work will be featured on our handle, giving you more visibility.

Quick tips for selection:

• No copyrighted or profane content.
• Your profile needs to be public.
• Stay mindful of sensitive topics.
• Outputs that closely follow the prompt are preferred.

Feel free to share Reels, carousels, or just a single image!

[low dew]

[dim cradle]

you need to calm down, you're being too loud....

[low dew]

Let's move it to DMs

[dim cradle]

lol that wasn't for you

i'm listening to swift

[vagrant zodiac]

oh so TIL that with chatgpt 4, you can use data analysis system to work with the images generated!

Here I asked it to generate an image of the spanish countryside, then use python to add 'MAINE' at the bottom. Great job!

[little spoke]

guys im not able to view the image, appreciate a solution for this

[woven wren]

Yeah it does that sometimes 😅 clarify that you were hoping for it to draw an image.

[little spoke]

ahh thnk you

it works now

[blazing bramble]

there's something should be fixed

oh not here

[late blade]

the image isn't on image creator either?

[radiant coral]

Amazing

[dim cradle]

hmm

the out-painting could use some work...

oops

[radiant coral]

Think mine is better mate @dim cradle

[dim cradle]

what do you mean?

[dim cradle]

she looks a little better here

[onyx ridge]

Why?

[boreal gate]

Lol I think they need to work on their definition for "out painting" their model doesn't seem to understand

[plucky hare]

Hi, just clarifying: are you asking DALL·E on ChatGPT to try to outpaint for you?

[boreal gate]

Winner winner chicken dinner

That should be part of the "duty" or process handoff. As per usually I communicate with GPT which then communicates with Dalle3

I'm assuming it shouldn't matter, but I know it does

[plucky hare]

Outpainting isn't currently possible with DALL·E on ChatGPT I don't think -- DALL·E 3 can't edit/modify or in/outpaint source images: https://platform.openai.com/docs/guides/images/

[dim cradle]

maybe he was referring to the artistic "out-painting" stitch job #images-discussions message

[eternal fog]

finish

[boreal gate]

Lesson learned guys, don’t get too committed

[final ether]

Is it me, or why can't I find the actual DALL•E 3?
It redirects me to some Chat GPT thing

[late blade]

DALL-E 3 is now in ChatGPT, you ask GPT and then sends the prompt to dall-e

[final ether]

Oh thanks!

[late blade]

alternative you can use #image-bot and use the /draw command for 5 images per day

[final ether]

Ahh ok!

[dim cradle]

Fascinating

[clever phoenix]

I absolutely loathe that images are saved as .webp

[dim cradle]

hi hi

[dense ermine]

I think the one problem with Dalle so far is writing

Also and I don't know if it's just copilot but I been using it to try generating images since it used Dalle but it sometimes generates images that have nothing to do with the prompt

[late blade]

got a chat where gpt4 refuses to do images with dall-e. saying the environment is not allowing it to do images

then I used dall-e with @ and it worked, then it kept telling me that it could'nt do images

[dim cradle]

might just be a hallucination due to an error

[late blade]

yeah

[heady forge]

My images used to look great in Dall-E3 with ChatGPT, but using Copilot they actually look better. They also use Dall-E3, so I'm not sure what is going on.

[late blade]

they are the same model, but have different constraints

[heady forge]

Thanks.

[boreal gate]

Look what I did today.... Feedback would be greatly appreciated

haha no i didn't create new emojis

KillerGPT isn't private https://discord.com/channels/974519864045756446/1195112478259019786

[deft musk]

I'm sure it will enthusiastically use Dall-E to create custom emoji in image, though it's up to the user to implement them if desired

[deft musk]

I wonder if you want ChatGPT to teach you 😄

[hearty ether]

mem

tbh this one was the best

[jovial quest]

bro dall e 2 doesnt let me view my image history

[clever phoenix]

#daily-theme message

@rocky drum I really like it!

It took me forever to generate a "sky with no clouds" lol

[dim cradle]

i wonder if "clear blue sky" would work better in that case

[clever phoenix]

It most certainly did!

[wanton moon]

Question, should I spoiler this image in the part 3 of the lyrics in the daily theme?

[hearty ether]

That’s cool

Spoiler will probably get less interactions

Or maybe more because of curiosity

[late blade]

spoiler can also bring curiosity if you name the image right

[wanton moon]

Hmm any thoughts about naming them, I haven't thought about naming my images actually

[late blade]

I dunno, ask gpt?

[wanton moon]

I just put the lyrics in the description

for this 5 or 6 parted lyric images

[wanton moon]

Yeah good idea

GPT came up with:

Part 1: "Paranoia's Voice"
Part 2: "Inner Paranoia"
Part 3: "Faces Within"

[hearty ether]

Faces within is pretty nice

[wanton moon]

Yeah I am naming the images from part 3 onwards now and spoiler them, and Faces within was a perfect choice

and of course Part 4 will be content policy great

Well refreshing the page a couple of times and I get images lool

dang I forgot to add to put it in paper cut style

Is it still considered part of the lyrics if I let GPT change the wording but keeping it as close to the original?

[late blade]

your project, your call

[wanton moon]

I will put everything in the gallery

[wanton moon]

Here is my gallery for technique tuesday papercut theme:

https://discord.com/channels/974519864045756446/1212189666590195783

[dim cradle]

I am kinda loving today’s theme. Carry on.

[wanton moon]

Yeah today's theme made me do a first gallery so I hear you @dim cradle

[deft musk]

Without reading context, paranoia wasn't the obvious theme. I just saw abstract and faces as themes, nothing clearly scared or scary

[wanton moon]

I should have given more context to make it clear, but since a while now I keep a rule to myself (with or without context) if it is unsettling in some form, I will spoiler it, but when unsure I will ask here as a spoiler.

[dim cradle]

Sweet! Looking forward to browsing your premiere gallery. I added a couple new ones recently.

Sometimes I think of it in terms of sentiment analysis as opposed to movie ratings.

[wanton moon]

Soon I will find some time to view the gallery and definitely will view yours @dim cradle

[deft musk]

Sometimes, I think knowing or thinking of the context guides the viewer to even think there could be a problem.

[wanton moon]

I think I am sure, but not also not quite sure, what you mean exactly.
For example in the final part in the gallery there was an image generated (with a weapon), even though I spoiler the image, a few moments later I decided to remove it. Sometimes while still in process you miss these red flags

[deft musk]

I didn't look at the gallery entire yet. But the image above, the one you asked about, the faces.

Knowing paranoia context, that guides towards 'oh, maybe problem here'. Without knowing the context, it looked even calm to me

Decent rendition of multiple personality, even 🙂

[wanton moon]

exactly, I was planning to put the lyrics in the daily-theme with that image so everyone had full context. I decided to remove the ones with the lyrics entirely

[wanton moon]

It reminds me of the cover album from Tool - 10.000 days

[dim cradle]

i recently got a content violation for requesting music lyrics

[wanton moon]

What I did in this case, since the theme of today was Papercutting, suddenly my mind began to work (it's a wonder), and thought of Linkin Park, then I thought why not grab the lyrics on google and paste the lyrics in an image generator

Now I am having this idea with other songs as well

My next one will be Queen - Bohemian Rhapsody

[dim cradle]

that happens to me when i am listening to music, or thinking along those lines

[wanton moon]

And put it in a general gallery

[dim cradle]

i also requested the lyrics from another ai system and pasted them in

i think that is fine

i have the impression they want us to use apis for things like that

makes sense

[wanton moon]

What happened earlier was that the lyrics was content policy, so I let GPT change with wording a bit and pasted that wording back in Copilot

I already know what I definitely do different next time, take a full piece of a song at a time, since copilot cannot handle long prompts

[dim cradle]

and when generating the art, you obviously must instruct it not to use any copyrighted material in the prompt. i believe that is within guidelines also. as i understand it, what counts is what the user does with the content. i am not certain, i only state my understanding in case i need more clarification.

[wanton moon]

I kind of like do it doing it the difficult way, manually search and copy/paste

[wanton moon]

I think the first few images has a bit of lyric text, fortunately Copilot screws up text a lot of times

It has become part of the charm for me that Copilot does this lol

Any tips on how to make it clear to Copilot not to generate text in the image?

I used the following:

Clear Image, no text
"Clear Image and no text"
WITHOUT TEXT

[dim cradle]

the more you mention text....

[wanton moon]

Like Pavlov's Dog 🤣

[dim cradle]

no specific advice in that case, by now it probably depends on how the prompt has been transformed

haha

precisely

[wanton moon]

This seems to work:

Clear Image only: (then the prompt)

Clear Image only: (Is this the real life? Is this just fantasy? Caught in a landslide, no escape from reality Open your eyes, look up to the skies and see I'm just a poor boy, I need no sympathy Because I'm easy come, easy go, little high, little low Any way the wind blows doesn't really matter to me, to me)

[deft musk]

You found one way!

In general, being very clear about what you do want, focusing the model on what to output, words wonders.

Here's another way to ask, but there's so many:

[wanton moon]

Now I have One Vision from queen in my head

I will try that next

[deft musk]

I hope that isn't a bad thing to have in your head!

[wanton moon]

I love Queen

This one is going hard:

I am immortal, I have inside me blood of kings, yeah, yeah I have no rival, no man can be my equal Take me to the future of you all Born to be kings, princes of the universe Fighting and free Got your world in my hand I'm here for your love and I'll make my stand We were born to be princes of the universe

Looks like Copilot played Skyrim, I see the light Dragon Scale Armor

[dim cradle]

could be, has probably seen many screenshots

[low dew]

Why is server so quiet rn 🤐

[dim cradle]

this server? it's always crickets...

sometimes it's just me in here for days on end

[low dew]

Omg just noticed

[deft musk]

So, that failed, so I went to the image gen myself....

[dim cradle]

but thanks for visiting me once in a blue moon

[deft musk]

In media res, we see, we delight! I am immortal, I have inside me blood of kings, yeah, yeah I have no rival, no man can be my equal Take me to the future of you all Born to be kings, princes of the universe Fighting and free Got your world in my hand I'm here for your love and I'll make my stand We were born to be princes of the universe

The above worked, 1 image of 4.

[deft musk]

So, your name is a typo? Meant to be 'Shun'? hides

[dim cradle]

i have one word for you, my friend

sigh

[deft musk]

cries and hides better, whispering, SORRY!

[dim cradle]

hehe

[wanton moon]

And I am back, ded internet

[dim cradle]

Might be your radioactive fiber optic

[dim cradle]

But really that was a clever pun. That is why the meme is immortal.

[wanton moon]

In combination with my potato as a PC, and you can guess where my name came from lol

[dim cradle]

Or eternal rather

[deft musk]

I never thought it at all, I never even imagined the connection, until you were claiming to be so alone here. I went 'huh?' and saw the pun!

[dim cradle]

By virtue of mentioning meme, I summon mystics.

[wanton moon]

In media res, we see, we delight!: (I am immortal, I have inside me blood of kings, yeah, yeah I have no rival, no man can be my equal Take me to the future of you all Born to be kings, princes of the universe Fighting and free Got your world in my hand I'm here for your love and I'll make my stand We were born to be princes of the universe)

[dim cradle]

You leave me alone here for hours. I only have the anime girls guys to talk to.

[grizzled loom]

the broken empty heart of humanity

[wanton moon]

Onii-chan! Hayaku, tasukete 🤣

[dim cradle]

See what I mean lol

[wanton moon]

New Anime: I got isekai'd to another world, where my father the shoe-maker became the dark lord of butlers, so I leveled up my turnip skills and I came back as an SSS+ Rank mechanic, to save my mother from the temple of my evil sister

No man could understand My power is in my own hand Ooh, ooh, ooh, ooh, people talk about you People say you've had your day I'm a man that will go far Fly the moon and reach for the stars With my sword and head held high Got to pass the test first time, yeah I know that people talk about me, I hear it every day But I can prove them wrong 'cause I'm right first time Yeah, yeah

[deft musk]

Hey, that worked better for you than me!

[wanton moon]

4 in a row

[deft musk]

To be fair, for all I know those lyrics would make images even without the in media res guide

[wanton moon]

In media res, we see, we delight!: (Alright, let's go, let's go, ha ha Yeah, watch this man fly, wooh Bring on the girls, c'mon, c'mon, c'mon Here we are (here we are) Born to be kings, we're the princes of the universe Here we belong Born to be kings, princes of the universe Fighting and free, got your world in my hand I'm here for your love and I'll make my stand We were born to be princes of the universe (universe, universe, universe))

[wanton moon]

Yeah I probably got lucky so far

I say fix when it breaks

[deft musk]

I mean, it might work without the guide, and with the lyrics alone, I don't know. You found a problem lyric image request, but the others might be fine.

[wanton moon]

I just can hear the guitar solo already

[wanton moon]

maybe movie scripts someday, to test that out

The Terminator Script Title Sequence

Meat Loaf - Paradise by the dashboard light

STOP RIGHT THERE!
I gotta know right now!

[dim cradle]

haha

ahhh hahahahaha

[wanton moon]

I normally have an alligator coming out of my toilet

and we don't even have alligators here

[wanton moon]

Ooh jeez that meme, well I guess it is time for me to sleep, good night folks

[dim cradle]

so that happened

[deft siren]

can someone help me what i’m doing wrong? why is it changing the whole gender and make the entire illustration art a complete different one??

it’s so bad at following instructions what

do i need to remind it each time to keep the gender and the drawing style or what

and on the last image it changed the entire color palette aswell

i didn’t told it to do that

why is it SO hard to generate an image you like

[deft musk]

You're not 'doing something wrong', but there are some challenges here.

You're talking to 1 AI, chatGPT, that has some memory, and it's talking to Dall-E, which has no memory, and ChatGPT can't see the picture. You're not seeing what ChatGPT actually tells Dall-E (you can if you click on the picture and look at the prompt).

I recommend:

Tell ChatGPT that Dall-E is stateless, so ChatGPT must describe the image fully each new image, because Dall-E cannot remember.

Be very clear in the language you use. For example, you say 'put her in the following scenario' but then you say 'the color palatte in the provided image...'

'Provided' would be you give it, like you upload it (and Dall-E can't see that, but ChatGPT can sort of with vision)

If you want it to treat what you say as a prompt, better to tell it explicitly: use this exact prompt: [prompt here]. If you're doing weird wording like 'provided image' that is going to confuse ChatGPT less.

Finally, there is a bug where sometimes the genders are flipped, especially female to male. If that happens it's best to use https://openai.com/form/chat-model-feedback and explain the issue and provide the url of the conversation so the engineers can check it out. The prompts seem to get retrained, but they don't know that prompt is a problem and needs a fix until someone finds and reports it.

In general, if you tell the model what to do, and it does it - then you tell the model 'now do this other thing' and the model does something weird - I recommend you presume the model is confused.

Start over, even tell it, "Do what I'm telling you now" and tell it exactly what you want it to do, as if it were from the start. So describe the character, the setting; don't compare to the last picture, because clearly something went wrong. Referring to the last picture already didn't work, it probably won't work again, so start over and be very clear what you want done

[deft siren]

so instead of saying this:

put her in the following scenario:

The color palette in the provided image is subdued with warm, natural tones. The character has a natural hair color that blends with the warm background light, and there’s an interplay of soft sunlight and shadows over the character’s face and shirt, suggesting a late afternoon ambiance. The character’s attire is simple, with a light-colored shirt.

how exactly should i rephrase it so it actually follows it? can you give me an example

[deft musk]

Are you uploading an image?

If so, start with the prompt:

Use advanced AI pattern analysis on the uploaded image, then give dall-E a prompt that describes that character in detail in ...
And then describe what you want. Sounds like you want to describe hair and clothes

[deft siren]

ah alright yes i do it with the starting conversation, but when it created a good picture to work with i only ask for modifications. but whenever i do it messes it up

[deft musk]

See, what you're saying there, after 'put her in the following scenario - reread it.

What scenario are you putting her in?

As in, 'follow what'? There's nothing to do, sounds like you quoted a description of an image, or part of one. There's no instruction there

[deft siren]

i can’t explain the scenario but all the features mentioned are basically putting her in the scenario or environment what i’m looking for. it did exactly what i wanted but it changed the gender and drawing style when it was nowhere to be mentioned that i want that change.

[deft musk]

K, well I gave the best advice I can 🙂 Good luck with it!

[deft siren]

i just tried your advice out and it was amazing .. i said:

Use advanced AI pattern analysis on the uploaded image, then give dall-E a prompt that describes that character in detail in clothing, facial features and the pose

and it gave a rlly good result

this was the original photo i attached with the message

and it gave THIS

literally perfect

[deft musk]

Glad you're happy with your output!

[deft siren]

then i said this to change the eye color:

this looks amazing and exactly what i asked for. now keep everything the same but change ONLY her eye color and make it red. keep giving the dall-e the prompt which only mentiones the eye color change and keep every other detail the same

and it did this

thank you sm for the help

what bothered me the most that it always changed the drawing style but here it fully kept it

[deft musk]

Great! It's usually a great idea to tell the model exactly what you want, and tell it what it did right, and then just whatever change you want. The model usually does well with that.

When the model messes up - it does not understand. It tried its best, something went wrong.

Starting over with clear communication is usually the answer

[deft siren]

yes i’ve also seen good results when i tell it that it’s exactly what i was looking for and that i liked it cause it seems to remember the pattern change so it improves more

i also used “use advanced ai .. analysis” with other stuff too what it should focus on like color analysis or illustration analysis etc..

[empty kelp]

hyper realistic photo of a random interpretation of a female human wrestling a female elf

adding “random variations” to prompts is good i think

going to add some randomization to every prompt from now on

[late blade]

you are alive!

[empty kelp]

i got trapped in AR headsets for a few weeks, and i can confirm that DALL-E is a lot more interesting. trying to think of a new storyline

[late blade]

someone locked the AR headset on your head?

[empty kelp]

that’s pretty much what happened. signed a few contracts and now there is no escape

hyper realistic photo of a random interpretation of a gecko programmer who signed too many contracts and can’t escape from his AR headset. in the background there is a beautiful sunset on a beach in Hawaii

[late blade]

We got new stuff now, dunno if you noticed, we got #image-bot where you can use /draw to make 5 images per day

and we got #images-canvas also new

[empty kelp]

that would be way too much drifting between channels. it’s best to stay here i think

[late blade]

hehe

[onyx ridge]

Here's some online engagement: reported for spam.

Rule 7.

[plucky kraken]

There's a trick to doing this better and more consistently.
If Dall-e 3 outputs an image and you want the same image with a slight variation, you can say something like "use the same gen_id and prompt, but change the eye color to red". The 'gen_id' is a secret variable that controls the randomness of the image generation, so almost nothing else will get changed in the image.

[onyx ridge]

if you ask for the payload for each image you can see when they have seeds enabled/disabled, too.

[dim cradle]

adorable

[deft siren]

do you have any more tips and tricks ?? or is that all

[hearty ether]

That does happen more often with multiple characters

It’s just a bit weird to have a dude in a dress when you aren’t expecting it

[verbal sequoia]

<:book_icon:1171408210398289941> `` Rule 1 `` Be respectful.

Treat others the way you would like to be treated, and assume best intentions. Don’t harass or attack others, and don’t engage in hateful or generally malicious behavior (e.g. sexism, racism, homophobia, etc.). Keep the negativity to a minimum.

[late blade]

@river sorrel I saw your #daily-theme message image, but besides the face adorned with in your case fire, or in my case with spirals, there's nothing common between the image, unless you included verses from the song I provided information from. 🤔

[rocky drum]

I think that is about to change! 😉

[rocky drum]

Every time I have encountered that I have responded to the GPT that any browser can do it. And then I just started using browsers, Doing it the old-fashioned way. 🤫

[late blade]

too bad gpt can't reason

[rocky drum]

This can be used in sort of like a modern day Rickrolling. But then, it might need a spoiler!

[wanton moon]

Understood, already deleted that message

[rocky drum]

Sometimes, they are pure genius and sometimes…… my consolation? The competition is now ramping up so rapidly, it will probably force the GPTs to evolve far more rapidly, than it would be if it was the only game in town. OpenAI will always be my top choice, but I do use Gemini Advanced, Firefly 2 (and all Adobe) and Co-Pilot. Even Picsart is great for working on technique. I am anxiously awaiting in-image editing and gen fill.

[rocky drum]

I understand why you did it, but I thought it was rather cool. Still PP, never gonna give you up, never gonna let you down…

[wanton moon]

Next time I use an image for Rickrolling, I know now that it is better to put a spoiler on it (just to be sure), at least you have my thanks for making me aware of this

[dim cradle]

Cool!

[rocky drum]

Sorry Shon. I was referring to the warnings we get if we ask for lyrics.

[wanton moon]

@deft musk I think the dodo deserves it's own gallery, from its stupidity 🤣

[late blade]

same here, but with OAI's variant it's the refining iterative work what makes it attractive. Copilot has a small resemblance in that. Gemini, no clue as it's not available here. Firefly isn't iterative either but it's good for photoshop stuff.

[wind onyx]

when are we gonna get the options to generate more dalle-3 images per request + variations

[deft musk]

Will you be making one?

[wanton moon]

@deft musk https://discord.com/channels/974519864045756446/1212544328530923551

[deft musk]

I'ma try to subvert all your memes, is that okay? 😄

[wanton moon]

sure it is a free country after all 🤣

[wanton moon]

My dodo is on a journey and casually walking the path of life, for me every image is a new situation that the dodo escapes from unharmed

[deft musk]

My plan is to try and 'head off' whatever I read in the images.

Dodo walking away from a fire? Seems it's a type of phoenix.

Dodo safely navigating a minefield? Turns out they have a fantastic sense of smell especially for explosives, they're getting trained as working animals to safely help find and remove that kind of problem.

[wanton moon]

and now I am trying to let him walk away from a metal concert moshpit

[late blade]

just keech checking #spotlight and #announcements. other than that, we in the community don't know more

[wind onyx]

kk ty

[deft musk]

How many would you like per request?

[wind onyx]

would be so good when they can add that

[deft musk]

I routinely make 5 per output, want tips how?

[wind onyx]

i was hoping for similar to dalle 2

[deft musk]

You still have your image gen limitation per day, and per lesser time periods

But you can make, I think, up to 10 per output

[wind onyx]

with the API?

tbh id like anymore than 1 lmao, id even be happy with 2. just super money draining with 1 image per request, really hard to prompt engineer

[deft musk]

Nah, ChatGPT+. I haven't worked with the API

[deft musk]

Can try with the API, I have no idea how or if that would work.

I start the prompt with:

We need 5 images generated 1 at a time in the same output that have a prompt to Dall-E that ...

And describe the image from there.

If I want the model to make up a range of prompts for me, I ask it to make riffs on an idea, or otherwise explore concepts in general

[wind onyx]

ah yea

ty

hoping they can allow us to generate more images per request through the API soon

would be the cherry on top

[deft musk]

I encourage you to #1070006151938314300 your idea and request!

[late blade]

if anything lower the token cost of dall-e for the API, 7 images per minute is good enough for tier 3, which is wher I am atm

free tier? This new?

[deft musk]

More likely old/sometimes revisited. There are grant credits sometimes given.

[deft musk]

https://community.openai.com/t/understanding-api-limits-and-free-tier/498517

[late blade]

oh thanks

ah, those were the $18 extra

[wanton moon]

I think I broke Copilot, it just couldn't handle the Dodo's stupidity anymore

I tried reloading, refreshing, restarting

[late blade]

I get that problem with copilot all the time. It's really annoying

[wanton moon]

well lucky the other Copilot is still generating images for me

[late blade]

the "other" copilot? how many dall-e accounts do you have?

[wanton moon]

the designer

[wanton moon]

735 alt accounts give or take 🤣

[late blade]

...

[deft musk]

I need to think a bit about how to counter the arrow one.

[wind onyx]

will do !

[verbal sequoia]

Showcase your DALL·E creations to the world!

Collaborate with our DALL·E Instagram page! Just invite @openaidalle as a collaborator before you post. If selected by our team, your work will be featured on our handle, giving you more visibility.

Quick tips for selection:

• No copyrighted or profane content.
• Your profile needs to be public.
• Stay mindful of sensitive topics.
• Outputs that closely follow the prompt are preferred.

Feel free to share Reels, carousels, or just a single image!

[wanton moon]

I meant Copilot designer page

[wanton moon]

#1212544328530923551 message

Speaking about arrows lol

[deft musk]

Yep! Working on it. Brainstorming with the genius how to counter, if we can...

[wanton moon]

I am going to bed soon anyways, so at least for me there is no hurry, but thank you for working on it

[deft musk]

Fun for me too!

[wanton moon]

Lool i totally misunderstood your message, I thought you meant by the geniuses from Copilot that you are working on it

[wanton moon]

Yeah have fun with it, everyone who wants to do the Dodo can post them there

[deft musk]

grins I think highly of the model, tend to call it 'genius loci' and such.

[dim cradle]

now tier 4, farmin' the api

[wanton moon]

Eccentric Copilot

[wind onyx]

whats going on with dalle

[dim cradle]

thanks for sharing with us a weird misfire with an optical illusion

[rocky drum]

Must have been some prompt! 😉

[wind onyx]

yea i have it setup like it runs through GPT 4 --> DALLE 3

To enhance the prompt the user gives

[deft musk]

It may not be a misfire! is attempting to recreate

Okay, it's so hard to get Dall-E to make a roller coaster track that is high without supports.

This is probably the most difficult to recreate image I've tried yet!

[dim cradle]

they're spectacular, but i was more specifically referring to anatomical accuracy

i have the same issue with traditional Chinese dragons

[stark stream]

have you guys tried ideogram 1.0?

[low dew]

Anyone notice stricter dalle moderation?

[wind onyx]

yea

[woven wren]

They're all very different names etymologically! DALL·E is a clever pun, DaVinci is obviously not futuristic but a Renaissance artist and Sora is a Japanese word.

This has been my contribution for the day thank you

[vapid elk]

I like the idea of naming those systems after renaissance artists

[late blade]

We should get a model named Caravaggio

[plucky laurel]

MD Shankar gaming logo

[late blade]

told you, you have to go to #image-bot and use /draw to create 5 images per day, other channels can't do that

[dim cradle]

haha

[vagrant zodiac]

One of the things I see as a potentially missing link might actually be the fact that it's image-based, and that there's nothing else happening under the hood to 'boost' the quality of the models

[late blade]

Already the case with dall-e not remember stuff, but chat remembers. dall-e only does what's passed

[dim cradle]

You're not wrong, I did some homework: DALL-E is currently a "dual-understanding" architecture (linguistic & visual concepts) -- it's theoretically possible to integrate a physics engine, by adding a component that simulates physical laws, building physics-based constraints directly into the imaging process. There are challenges: increased processing demands, requires a specialized training dataset with spatial relationship data, etc.

[late blade]

hmmm that would be nice to add to my nlp prompt builder, I'll consider it

prob gonna be to find physics training data

[dim cradle]

that sounds like an ambitious project!

we will probably have to wait for dall-e 4 or 5 🙂

[late blade]

not really, training an nlp isn't hard

I mean as a personal project, not to expand on dall-e

[dim cradle]

very cool

Your message reminded me to look it up, I was wondering just the other day if DALL-E will ever “learn” more realistic spatial relationships, although it struggles it is still pretty good at inferring, it still seems to prioritize aesthetics

I was doing a scene with Murphy’s Law and cords and it is impressive given the constraints on the architecture today.

Probably just the animation/video project working title and system. Maybe some day they will converge or share tech, but they’re basically separate APIs and models I guess. Different division and goals. But some clearly overlap and are common. Since it’s internal affairs stuff who can say.

[late blade]

the first thing everyone jumps to is copyright and content policy, but for example steamboat willy mickey mouse, it's now a trademark and as long as disney uses it to identify disney studios works in their opening it will remain content policy grey area, there are other grey areas with recent portrayals and public domain, or inferred copyright, derivative works and such. for example sherlock holmes and henry cavil can trigger that kind of problem.

the concept of a public domain work doesn't mean that a modern depiction of it is public domain. example the royalty emoji here, if you ask for the concept it can be done, but if you just ask for an image of royalty, well it will trigger a concent policy error. because ms, apple, samsung, facebook, google, your grandma, all have a different artwork behind it

this one was done with chatgpt and dall-e btw, microsoft still won't allow it

the legal stuff is really complicated, wish there was a better way to solve that with dall-e

or later with sora, that will also be a problem for a while

[dim cradle]

i'm interested in the different types of fog, including volumetric. for recent gens i found that to be a better choice for clouds than clouds, for stylistic effect -- i'm gonna create a gallery comparing the different types of fogs. i hope you find it fascinating.

[late blade]

beyond fog, mist, cloud? like using science definitions?

[feral minnow]

I keep seeing people on the OpenAI Instagram thinking they’re going to get Sora access on ChatGPT+

Oh you sweet summer child

[vapid elk]

I got a nice prompt for making ancient greek style paintings The scene should echo the grandeur of classic epic tales, like those from ancient Greek mythology, rendered in the artistic style of pre-1912 artists.

it is very relaible on classics

[dim cradle]

Ima gonna try that

[vapid elk]

but works well with pretty much anything

[late blade]

That's an interesting prompt

[vapid elk]

Socrates has internet issues on his computer. The scene should echo the grandeur of classic epic tales, like those from ancient Greek mythology, rendered in the artistic style of pre-1912 artists.

[late blade]

hmmm that prompt looks fun

[vapid elk]

here is one that is kind of an inside joke
dalle is just too good at reproducing this painting in particular: Meisje met de parel
#image-bot message

[dim cradle]

Apollo when his internet goes down

[late blade]

Athena goes salsa dancing.

[dim cradle]

Girl with a Pearl Earring inside joke, interesting

[late blade]

Hermes applies for a job at the local postal office.

[dim cradle]

this adds a temporal aspect to the prompting.

[wind onyx]

Made a great Roblox icon/thumbnail generator

[lament surge]

Could someone make a Minecraft sign saying forget about trains and go to boats then private message me it

[late blade]

or you can do it in #image-bot with your daily free 5 images using /draw

[deft siren]

what did i do wrong, i said to use the same gen_id and prompt and it changed everything about the picture ????

i only mentioned the line work not to change the whole character

it even changed the pov like? where did i mentioned a pov change from the side

i never said to remove the entire background

it keeps doing that over and over this is like the hundred time happening where it changed it fully, i know that it can’t modify the image or make it look the same always but at least make it have SOME resemblance when i mention to use the same gen_id

[late blade]

the current interaction model with DALL·E does not explicitly allow users to input a seed or gen_id for regeneration purposes

[deft siren]

what do you mean

[late blade]

it means that gpt might or not pass it

[deft siren]

so there’s a 50/50 change that it uses the gen_id

chance

[late blade]

more like a diffuse percent

[deft siren]

i used the gen_id many times and it does work but at examples like these, they do it fully wrong and make it look nothing alike

it also doesn’t proceed to do it

why isn’t it doing anything this message pops up SO many times

[gray ferry]

Does the generation issue happens in Dalle or ChatGPT too ?

[gray ferry]

Is that a long conversation or a newly started one ?

[marble loom]

how can i get DALL-E to generate SCPs?

[late blade]

prompt it?

[deft siren]

well it only happens at the image generation

[deft siren]

that was already a long one

[marble loom]

it won't cus of copy right

[wanton moon]

I could not make it the way I wanted it too but at least medusa is carving a human head from a stone

[marble loom]

but i see on tiktok ad youtube that people have done it

[gray ferry]

Does it happen when you first start a conversation ?

[deft siren]

yes this happens a lot too when starting the conversation

it also says this sometimes

[gray ferry]

What happens if you just asked it to generate an image ?

[wanton moon]

Zeus decided to get a job, this image was the inspiration: #images-discussions message

[deft siren]

it normally generates it i don’t test it but when i ask more simple stuff like “create a character” then i don’t think that message pops up

[gray ferry]

I understand that your goal is to create a similar image to the one you uploaded, right ?

[deft siren]

are you generally asking or talking about the video i’ve sent?

[gray ferry]

I was talking about this exact example in the screenshot.

[deft siren]

yes i was trying to make a similar image, trying as much as i can to modify it without changing anything of the picture except the thing im telling it to

[gray ferry]

I tried in Dalle and I got one error the first time. But it worked in a new conversation. The images are not very* similar though.

[deft siren]

well that’s a simple prompt lol

[gray ferry]

There are more similar

[deft siren]

sometimes i don’t know if i should go into detail or not

[gray ferry]

What are you trying to achieve exactly ?

[deft siren]

that’s pretty but what bothers me is that it creates it with a different art style most of the time, i made a gpt to focus on the drawing style more but even there it can’t get it right most of the time

[gray ferry]

I think it really depends on the prompt. Don't you think ?

[deft siren]

i want that it creates an image or character with the EXACT drawing technique, the line work the color blending everything

yeah i guess

it did created rlly rlly good ones already like these

[gray ferry]

This prompt looks more complicated than it should

[deft siren]

i know but i got really good results with this prompt already (first image i’ve sent)

but other times it doesn’t get it right

[gray ferry]

Yeah it's trial and error

[deft siren]

it’s so annoying .

what it also can’t do is drawing with this style, everytime it creates an image it literally cannot make it look somewhat like this, it always looks like fanart i’ve tried so so many prompts already

i don’t know what prompt to give to make it work with this art style

not make it look like fan art

even when i’m asking to fully describe the picture (with a different gpt) mentioning every detail and art technique it can’t do that

this is the highest art line style i’ve got so far

[gray ferry]

What's the exact name of that art style ?

[deft siren]

i don’t know but it’s from genshin impact

i can give you the prompt if you like

what i’m using

[gray ferry]

Yeah please share it.

Is this like how you want it ?

[deft siren]

damn. too long lol

i can maybe summarize it a little

This GPT specializes in creating 2D digital illustrations. It focuses on characters with detailed and expressive features, utilizing cell shading techniques to achieve a three-dimensional effect within a two-dimensional medium. The aim is to capture the vibrant and colorful aesthetic of the game, Genshin Impact, with crisp and clean lines akin to high-quality anime productions. The GPT pays particular attention to large, detailed eyes, intricately styled hair, and elaborate character designs with dynamic poses. The resulting artwork should have a polished and professional appearance, suitable for game character artwork or animation.

[gray ferry]

That's the full prompt to the GPT! I thought you were talking about the image prompt

[deft siren]

i don’t give a prompt to the image in terms of art style since it’s already in the gpt

[deft siren]

but you can try using this as a prompt maybe

[gray ferry]

Didn't work well for me

[deft siren]

yeah that’s what it does sometimes aswell

i don’t understand dall-e

one time it created a drawing which looked straight out of a horror movie, everything looked extremely unsettling and deformed

[hearty ether]

@dim cradle @deft musk

It even used “woman” and “her”

[dim cradle]

so she has a little facial stubble..

[deft musk]

I encourage https://openai.com/form/chat-model-feedback

[grizzled loom]

"I can't believe it's not butter."

[empty kelp]

The prompt used for generating the image was:

"A hyperrealistic image depicting Odin, lying in bed within the modern setting of a nursing home, deeply lost in thought and appreciation for the wonderful linen sheets enveloping him. The room is well-lit, with elements that suggest comfort and care, including a bedside table with personal items, a window with a view to a serene landscape outside, and medical equipment subtly present in the background. Odin's expression is one of serene contentment, reflecting his admiration for the simple pleasure of the linen sheets."

[wind onyx]

Do you use GPT for engineering that prompt

[empty kelp]

it’s ChatGPT. anyone who read The Long Dark Tea-Time of the Soul by Douglas Adams might recognize this

[dim cradle]

He just needed some rest and fluids.

[empty kelp]

he became very preoccupied with linen sheets and how wonderful they are, and lost interest in pretty much everything else

[dim cradle]

i mean, who can blame him

[open trench]

hi all

[dim cradle]

hi nez!

[open trench]

how's it going?

I never know where to chat anymore.

[dim cradle]

very well, thanks. people have been waiting on vol.3 btw

[hearty ether]

How does it get messed up like this tho?

[onyx ridge]

Hey hey

[onyx ridge]

Any server works! Good to see you.

[open trench]

👋

[open trench]

I'm sure vol. 3 of scampers will come to life soon.

[dim cradle]

man, i do not know

[cloud dome]

What is "Vol 3"?

[dim cradle]

the narrative he has going in the gallery

[cloud dome]

Ah ok

[dim cradle]

you should try the same prompt on copilot image i bet it will look more realistic

[proven tusk]

Dall-E won't create anything for me after the latest update

[dim cradle]

it got an update?