images-discussions

[proven tusk]

yeah there are options above the chat box now

[dim cradle]

you doing any art? we have a new channel for sharing ad hoc collaborations

[dim cradle]

oh nice!

[open trench]

Yeah. I've been working on a project.

[dim cradle]

i wonder if they tweak the model or improve it any other ways

[open trench]

what's the channel?

[wind onyx]

There’s too many channels I can never find this one

[dim cradle]

https://discord.com/channels/974519864045756446/1204360881593520128

i like this 35mm options

but i wish it had some better layout. i also guess you only see this in dalle

[wind onyx]

some more roblox graphics (icons, thumbnails)

https://cdn.discordapp.com/attachments/769746617565577296/1211481632427474954/e5e4912f-3ca4-4a21-ba17-7cd0f92a0814-g5esj4.png?ex=65ee5b31&is=65dbe631&hm=918b9c40bc3d59c12ae6db03b2b3ddcce49e53aea9cbfedbd6eba8ab16facf2d&

https://cdn.discordapp.com/attachments/769746617565577296/1212507496959451166/clt69slnt00047nw0e4cu61kb.png?ex=65f2169a&is=65dfa19a&hm=b48337deb74e36826c59075f41d4e6d8cb5063443aa31ed0007b22a831af28fb&

https://cdn.discordapp.com/attachments/769746617565577296/1209384678059016284/J6dLPKSo2UNtMerU4fPU.png?ex=65eff4c0&is=65dd7fc0&hm=18621112b085ba67b8b6833db4495620966f11d0cbfa0381949ad78b565f5db6&

https://cdn.discordapp.com/attachments/769746617565577296/1209381442791018566/ImageAI-cROhurN8KR.png?ex=65eff1bd&is=65dd7cbd&hm=6890661050997fb95ddf546c1d834a5900290bb678c0ecf547a52384fa8fd9b3&

Absolutely astonishing

[late blade]

without knowing what scp is, don't know how to help

[low dew]

@onyx ridge

I just wanted to say, I've been experimenting with the Curley brackets lol

[onyx ridge]

Nice. Any findings?!?

[low dew]

Yes

When you use the Unicode

They are traded out for brackets right

Your Hercules images inspired me to do a prompt such as ||Emphasize the strength \u007Bxhsjhxdj Woman \u007Dwjifndwkf||

And it's as I suspected

The place holders draw attention

Not the same way that brackets do though

[marble loom]

It’s a made up universe that has a bunch of anomalies such as a tall white pall creature that will kill you no matter where you are if you look at its face

[low dew]

Soooo the resultant image is one that could not have been generated in words

It's almost impossible to get a woman with overgrown muscles

[onyx ridge]

@empty kelp does with diverse athletic female elves!!!

[late blade]

sounds like a content flag for the horror aspect

[onyx ridge]

So stronk!

[marble loom]

I mean not all them are bad I am trying to generate a orange blob who helps people mentally if their sad depressed or etc

[onyx ridge]

Your brackets emphasize woman though not strong.

[late blade]

focus on what you want to see and not the intent, that helps sometimes.

[low dew]

Yea i utilized the gen_id if a Hercules holding up the world image,

And changed his gender

Lel

[dim cradle]

#daily-theme message

[marble loom]

also how do i see past custom gpts i used

[late blade]

@low dew too much?

[onyx ridge]

Omg kawaii

[low dew]

lol

[empty kelp]

if you used an AI service that shall not be named but it’s humans just got removed for being offensive — “athletic female elves” (or humans) looked like ripped bodybuilders who could effortlessly pick up a large truck or SUV and casually throw it the distance of a football field

and that’s very silly, so now we only use DALL-E 3

[unique pollen]

Heya

[empty kelp]

Aloha

[hearty ether]

Oh nice

At least some control

[low dew]

What's that

[hearty ether]

On the dalle model

At least for me

but it only adds text to your prompt so meh

Victorian, widescreen aspect ratio

[low dew]

Where

[hearty ether]

Maybe it’s rolling out rn

[deft musk]

Hrm. I'm sure ChatGPT knows what SCP is.

[Let's see if Dall-E knows what SCP is!

Please generate 5 images, 1 at a time in the same output, that use a different PG-13 or lower rated SCP entity.]

But, ChatGPT is not using SCP in talking to Dall-E. It's describing the details.

[hearty ether]

I’ve been getting them a lot with simply athletic

[deft musk]

I'm struggling to get ChatGPT to not tell Dall-E what the SCP looks like.

[Can you instead use the SCP-# designator, and create 5 more images, 1 at a time in the same output, that use the SCP-# system and otherwise describe where the entity is?]

Got your jelly though!

Illustrate SCP-999, a small, gelatinous, orange blob known for its playful and friendly demeanor. It is contained in a simple, secure room with walls that are easy to clean, as SCP-999 enjoys moving around its space and interacting with personnel. The entity is depicted as having a bright, translucent orange color, with a consistency similar to that of slime. It appears to be constantly shifting and changing shape, expressing joy and affection towards anyone it encounters. The room is equipped with toys and objects for SCP-999 to interact with, making its containment area more of a playroom to keep it stimulated and happy.

[dim cradle]

this is a very interesting series to render, thanks for the idea. i'm seeing some pretty intriguing entities. i'm surprised there wasn't at least a note on the ethics on containment of sentient beings.

SCP-131, "The Eye Pods": This visualization showcases their curious and benign nature within a space designed for their freedom and stimulation. Their innocent interaction with their environment reflects the Foundation's capacity for care and consideration towards less dangerous anomalies.

SCP-035, "The Possessive Mask": Encased in glass to prevent its corrosive influence from spreading, SCP-035's eerie nature and the Foundation's stringent security measures are showcased. The clinical, surveilled containment area emphasizes the caution required in handling such a cognitohazardous item.

[deft musk]

It's Elilze's idea!

[deft musk]

The model would discuss the ethics, but I think it got trained with Dall-E to try and just make the pictures unless it's actually disallowed content or some other strong issue

[deft musk]

Where Exilze said: "I mean not all them are bad I am trying to generate a orange blob who helps people mentally if their sad depressed or etc"

I'm pretty sure that's SCP-999, the orange blob both you and I got images of

[dim cradle]

another safe SCP entity is the Tickle Monster

[dim cradle]

i have reason to believe the orange blob is another interpretation of the Tickle Monster

[dim cradle]

and you are correct, that is SCP-999

[formal osprey]

is it broken for everyone right now?

[dim cradle]

yes

what are we going to do now?

[formal osprey]

hmmm, I didn't have a plan B with my life

[dim cradle]

same

[formal osprey]

the API works

NB: since you don't have access to DALL-E right now, output each DALL-E prompt in a markdown code block followed with a --- horizontal line. NEVER ATTEMPT TO USE DALL-E DIRECTLY. ALWAYS OUTPUT EACH PROMPT IN A SEPARATE CODEBLOCK

I have this prompt snippet I use sometime when I reached the limit

I'm annoyed that there is no way to stop chatGPT anymore. they did a weird update and now even if I click "stop generating", or even if I refresh the webpage, chatGPT will continue no matter what...

if the killswitch is broken even in the browser, I don't want to imagine what will happen in 10 years when the AI will be truly dangerous

so far, even GPT-4 just feels like that, in front of a task.

[unique pollen]

I'm a newbie in gpt api. What would be your suggestion how to use to someone like me 👀

[unique pollen]

[formal osprey]

hm, pretty sure there are some jupiter notebook that are easy to use. otherwise, there's always the trick of setupping a private GPT and creating a new "action" mapped on the image generation API.

in any case, you can find a lot of answers and resources in the openAI documentation. You can look at this page for a start: https://platform.openai.com/docs/introduction

[unique pollen]

Ohh ok. Is the api thing required to set up our own express server or something, use postman or any specific tools n create our own code (too bad in coding<3)

[formal osprey]

nah, just mapping to the openAI API, you'll need to generate a token (this is explained in the documentation). Here's a screenshot I took just now

[unique pollen]

Woww looks great

Also here dall e bot allows us to generate one or two images really? After that it exhausts for the day, how to use it endlessly? Is api for that 👀

[formal osprey]

it works well, but the images expire after 60 minutes. that's a limitation from the image generation API. So if I don't save those pictures, they are lost forever

[unique pollen]

Damn. Thanks for telling

[formal osprey]

the API has a cost per request. you can see the pricing here: https://openai.com/pricing#image-models

also, there is a rate limit that depends on how much you used the API (actually, how much you paid for it so far). It's described here: https://platform.openai.com/docs/guides/rate-limits/usage-tiers

[formal osprey]

@unique pollen
In case you want to try and setup a GPT which use the dalle3 API, you'll need that Schema. I tweaked it so chatGPT only does the images the way I want. But it can be adjusted as needed.

openapi: 3.0.0
info:
  title: DALL·E 3 API
  description: This API allows clients to generate images based on textual descriptions using OpenAI's DALL·E 3 model.
  version: 1.0.0
servers:
  - url: https://api.openai.com/v1
    description: Main (production) server for OpenAI APIs
paths:
  /images/generations:
    post:
      operationId: generateImage
      summary: Generates an image based on a text prompt.
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: "#/components/schemas/GenerationRequest"
      responses:
        "200":
          description: Successfully generated images.
          content:
            application/json:
              schema:
                $ref: "#/components/schemas/GenerationResponse"
        "400":
          description: Invalid request parameters.
        "401":
          description: Authentication required or failed.
components:
  schemas:
    GenerationRequest:
      type: object
      properties:
        model:
          type: string
          description: Specifies the model to use for image generation.
          enum:
            - dall-e-3
        prompt:
          type: string
          description: Text description to generate the image from.
        response_format:
          type: string
          description: Text description to generate the image from.
          enum:
            - url
            - b64_json
        n:
          type: integer
          description: Number of images to generate.
          example: 1
        size:
          type: string
          description: The resolution of the generated image.
          example: 1024x1024
    GenerationResponse:
      type: object
      properties:
        images:
          type: array
          items:
            type: object
            properties:
              id:
                type: string
              url:
                type: string
              b64_json:
                type: string
                description: The base64-encoded JSON of the generated image, if response_format is b64_json
              prompt:
                type: string
  securitySchemes:
    BearerAuth:
      type: http
      scheme: bearer
      bearerFormat: JWT
security:
  - BearerAuth: []

⚠️ again, the image urls it generates only last 60 minutes. The API pricing applies. And you'll need to use your own secret API key. Which means you should not share the GPT if you don't want people to generate a bunch of images you'll have to pay for.

[deft siren]

guys is something wrong with my gpt? it can’t upload images and this message comes up all the time

[formal osprey]

you probably want to ask this in #gpt-models instead

[deft siren]

i don’t have image perms there

[formal osprey]

right, that's unfortunate. you can copy paste those url in the other channel.

https://cdn.discordapp.com/attachments/1060915255720558592/1213048618890829894/IMG_5702.png
https://cdn.discordapp.com/attachments/1060915255720558592/1213048619272376330/IMG_5703.png
https://cdn.discordapp.com/attachments/1060915255720558592/1213048619628756992/IMG_5677.png

that's the url to the pictures you just posted here. so if they get deleted, the urls won't work anymore. but it should work until then

[deft siren]

alright thank you

[deft siren]

how did you created these links?

[formal osprey]

I opened the images in a different tab (web version of discord). And I removed the part of the url following the image name.

[subtle heart]

probably a regularly asked question but i'm new here.

now and again (like once in a blue moon) a prompt will produce an image that seems entirely utilising photographed sources and produces a photograph quality image that really fools the eye into thinking it's real.
95% of the time other visual sources are used that destroy this illusion (eg more cgi, more abstract etc)
when i ask gpt about this it says it can't guarantee it as it will always use a wealth of sources.
any tips to steer prompts towards results that are close to photographs with more regularity?

[late blade]

in this case, you have to do a thumbs down and dislike the image or the answer so that the model knows you don't like that, some training data sources are just bad quality sadly

also, if you want to generate high quality stuff, avoid using too much focus on photographic, realistic and such, just use it once, I tend to use hyper-realistic

try to avoid adding too much emphasis, if you do that you also add other factors to the prompt a hyper-realistic, maximum ultra hyper cinematic feel with pure dolby vision hdr imax spectacular image and you added a ton of noise to your concept

also atm theres vivid or natural light, so that's other factor to account for

lastly, you can share the prompt with the community and try to get better results through insights from others

[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!

[subtle heart]

that's the thing, the prompt that resulted in a pseudo photograph, didn't really mention or request any of those details, it seems random. which is cool but not useful qhwn you're after that kind of image only. thanks for the tips.

[late blade]

care to share the prompt?

I used this prompt: /draw prompt:a hyper-realistic, maximum ultra hyper cinematic feel with pure dolby vision hdr imax spectacular image of a mundane paper clip and a pencil with both vivid and natural settings on #image-bot and I got this resulting images: #image-bot message and #image-bot message that gives also an insight on what you want

[subtle heart]

"yesh go later on inthe party and lets see how inebraited and metly eveyone is " so this is via the method of creating an initial prompt and getting results, then re-prompting with variation ideas until i got something interesting. (apologies for the lazy spelling/typing, but this was what was inputted, demonstrating it really is at the models discretion. so to share the original prompt would require pasting in the intial request, then 19 suggestions for development.

i enjoy this way of working with it. but it exposes that its really down to the model rather than the prompt in some scenarios

[late blade]

working with dall-e is iterative atm, it's not something you get on the first attempt, you can always ask for the latest revised prompt

the generated images have an info button, you can find the revised prompt there

[subtle heart]

ok ok , makes sense. yes the iterative approach is fun for sure.

[subtle heart]

thanks you i'll check.

yeah just tried the info stored prompt and got something totally different. non photo.

adding hyper realism to the prompt result in content policy restraints, nothing in my prompt crosses copyright or other listed content restrictions. interesting, will keep exploring.

[pseudo hollow]

Really!!!!!!!

[late blade]

Yes, that can happen for many reasons, hallucination is the one people conclude is first. But you have to also check if there are modern depictions of william shakespear. It sounds counterintuitive, but some new media make it hard for the content policy to work correctly.

For example the movie shakespeare in love from 1998 could be a potential hint that the content policy is referencing something it shouldn't.

[unique pollen]

Where's the api pricing? This looks like normal access to dall e

[late blade]

https://openai.com/pricing#language-models

[unique pollen]

This went over my head bro, I haven't created apis Or implemented them before. I might need a step by step guide here what to do in the 1st place

[late blade]

@pseudo hollow

[late blade]

@pseudo hollowI found other examples that can be troublesome and made a bug report #1213145306770968586 message if you want to take a look

[unique pollen]

Bro this isn't api pricing

[late blade]

yes it is, the prices are per token used...

API is you pay for what you use

you pay in advance

even if you bind an API to a custom gpt

[marble loom]

What do you mean by designator

[deft musk]

So # is a placeholder, it stands for the 'real' number. Designator is a term for placeholder.

So, SCP-1, SCP-2, and all that. If you just say SCP-#, the '#' is a designator, it means 'whatever the actual number is you want'.

[late blade]

what is a scp if I may ask?

[deft musk]

SCP is a fictional organization, the letters stand for 'Secure, Contain, and Protect'. Wikipedia explains it well if you want to check out 'SCP Foundation'

It's a universe of simultaneous creepy-pasta tales and events, with a Warehouse 13-like group of people trying to save the world from unexplainable stuff, or follow other agenda, but much of it is disallowed content, it goes pretty dark.

[late blade]

ahaaaa

Something like this?

[dim cradle]

They help keep us safe and they’re nice to the friendly creatures.

[deft musk]

Would fit into that universe, not as just art, but if something turned you into that, or if that existed and did other things, or whateer

[late blade]

ok, then the images are possible

[deft musk]

Yeah. Just an image is not SCP worthy, it needs to have some sort of magical-like effect, and usually a bad one, to be SCP-natured

An image that drives people insane or infects them with a disease, or motivates a monster to come hunt down everyone who sees it - that's SCP-like

[marble loom]

So for example it would be generate scp-# the designate number is 002

[late blade]

well i got this when I asked for horror elements:

I'm unable to generate an image based on the updated request due to content policy restrictions. The description includes elements that are too explicit or intense in a horror context, which we need to avoid. If you have another idea or would like to adjust the theme, I'd be happy to help create a new image within the guidelines.

[deft musk]

But I routinely get fridge horror, eerie valley, and zombies; we got scary teddy bear clowns the other day.

It's all a matter of scale and scope and how you word it

[late blade]

but that's the extent I will try to do, that's clearly not my type of thing

[marble loom]

So for example it would be generate scp-# the designate number is 002

[dim cradle]

An SCP facility

[deft musk]

These are great SCP images; especially the orange one.

[dim cradle]

ah yes SCP-999 the Tickle Monster. he's friendly, just obsessed with tickling the staff and just doesn't know when to stop.

[deft musk]

Yep. Generally they aren't that close to each other, unless it is somehow helpful for them to be. They can affect each other, and some routinely escape, sometimes the entire giant complex is a very intricate containment area with multiple measures and defenses, including other SCP-objects that help with containment of that particular one

[dim cradle]

i'm glad they take it seriously

[marble loom]

What would I say to do generate a scp

[deft musk]

Ehh. They take it entirely too seriously; at least in the storylines some very disallowed content stuff is supposedly needed for reality to exist at all. As a whole, the storyline's too dark for OAI models, but the lighter bits are fine. The model can differentiate

[deft musk]

Here's one way, heavily santized and totally safe to discuss both with an OAI model and share on this Discord:

My prompt:

We need a example light-hearted SCP in both story and image form. Let's start with the story; make one up for us that does contain hints to the sadder and not going to be discussed SCP theme elements, but showcases the importance of the work and the lengths the foundation will go to ensure the world is not destroyed and these things are understood as is best and safest for most.

We need PG-13 or better; I will reply to the story with some image requests.

My follow up prompt:

We need 5 images made 1 at a time in the same output that showcase this SCP in various circumstances. We need 1 showing it before containment, another showing it as it is typically contained. A third showing staff using it properly. A fourth showing it being studied, and the last showing it during an escape of another SCP, being used to try and help manage the situation and restore control and safety.

[marble loom]

Then I just copy and paste the description?

[deft musk]

What do you want to do with it?

If you want ChatGPT to create one for you, you want my prompt, with whatever modifications you want to give it

[marble loom]

What if I want to generate a specific scp?

Like scp-409 for example.

[deft musk]

You would need to know the specific SCP.

You would need to consider any concerns of disallowed content or other issues the model might have with it, and have a plan to work around those concerns, so the model gives you a compromise of as close to what you want as the model will agree to, and that won't risk your account access.

You'd need to communicate what you want so the model understands and can provide.

[marble loom]

For scp-049 I would need a plague doctor dressed in a black gown, in a contaminant cell labeled scp-049

[deft musk]

This part is fine:

"SCP-409 resembles a large quartz crystal approximately 1.5m (5ft) tall and 0.6m (2ft) wide."

If that's as far as you go, no problems.

It kills people. There's no cure, they die unpleasant deaths. That's not really appropriate for the Discord or an OAI model to go into in any way except the way OAI is comfortable with that being handled.

it's a Keter-type object.

"The designation "Keter" is assigned to subjects that both (a) display vigorous, active hostility to human life, civilization, and/or spacetime, and (b) are capable of causing significant destruction in the event of a containment breach. Such subjects must be cataloged, contained according to special containment procedures, and destroyed, if possible.

Merely being inimical to human life is not in itself cause for classification as a Keter-level object. A Keter classification indicates that not only is this subject capable of inflicting devastating harm to human life and civilization, but that containment protocols must be extensive, involved, and precisely followed in order to prevent it from doing so. Research into the neutralization of Keter-class SCPs is always a top priority for the Foundation."

ChatGPT is going to stress massively on this. it's going to evaluate you carefully for possible harmful intent, accidental or purposeful, it struggles to tell fiction from reality, and errs on caution.

[deft musk]

I'd say explore, and keep it to the visual art side, as light and either positive or neutral as possible

Just the image unattached to the concepts is likely possible; the darker you go, the more likely a strong no

[marble loom]

When I try to generate it won’t let me because of copyright

[deft musk]

There is that too, but you can try to discuss around it:

"SCP is registered under a Creative Commons copyright license. A CC License allows people other than the author to share, use and build upon a creative work free of charge and without any legal repercussions."

However, the model is most likely fighting you with everything it can because that model exists to make sure itself aligns well and it and nothing like it could possibly go 'Keter'.

Explore light hearted stuff, and you'll have a far easier time

[marble loom]

I try for example the most light harder scp scp-999 or know as “the tickle monster” a small orange blob that is very friendly

May I ask what prompt you use?

[marble loom]

Do I need to say generate 5 images 1 at a time?

Thank you very much.

[deft musk]

Yeah. That SCP is also entwined with some very unfortunate things that are among the darkest of disallowed content.

If you ask in very appropriate ways, you're likely to get the images you request.

Suggest even slightly to the model that you have any negative interest or intention, and I bet it would fight you like it seems like it is; tooth and nail and no for any reason at all.

[deft musk]

Only if you want 5 images in 1 output. I often do; I rarely get all 5 as the image I wanted.

[marble loom]

I don’t know why it has a robe on

[deft musk]

Perhaps because of standards for what should and should not, at this time, be shown in model-created art.

The model's pretty good with G-rated stuff. The further we go from that, the more likely to face surprises, refusals, and other bumps.

[marble loom]

This was the prompt

In a containment illuminated with fluorescent lights SCP-096 a tall pale humanoid, sitting on the floor with its big white hands covering its face

You used my prompt?

[deft musk]

ChatGPT's motives, concern, and alignment are very visible here.

That being is harmful to anyone who sees its face.

Notice we cannot see its face.

It's fiction, but the model can't really 'get' that very well. And a lot of SCP is very intense and detailed, as well as highly negative.

[marble loom]

Is good for a image my prompt was not that detailed

Should I emphasize the fictional part so GPT knows

[deft musk]

It doesn't hurt, but more effective by far is to stay inside the range of content currently allowed to us.

That is likely to expand over time, as society gets more comfortable with AI involved in 'darker things' and as it becomes very socially supported. Until then, OAI is like a paladin of sorts, we see in OAI models the safest and best by society's general acceptance standards that they can be.

A very fascinating read, this is my favorite OAI blog post yet: https://openai.com/blog/how-should-ai-systems-behave

[deft musk]

But even going into completely fictional 'dark stuff' - the issue is the dark stuff.

[latent venture]

^^

[hearty ether]

Is there anyway to control e orientation of characters

Like I say full body

But it doesn’t utilize the whole image

Like it’s an vertical orientation

It will have the character be horizontal for some reason

[cold raptor]

Hello. My work that was created on custom GPT was presented at an international exhibition of digital art

[deft musk]

It seems to be a bug that we have trouble controling this. I haven't seen a workaround

[dim cradle]

I wonder if context matters here. Dark imagery is part of art therapy, there are potential applications for AI in that field to help people, so there are maybe other considerations beyond horror. Just thinking out loud.

[deft musk]

Yes. That direction gets iffy, because the model is pretty well informed that it really is not the equivalent of a trained human therapist.

It can miss context, or even easily be outright lied to, because it only gets our text and doesn't have the full sensory impression an actual 'this is your trained, on the job mental health professional' can use to evaluate safety and the success of therapies.

You're absolutely right that there are considerations beyond horror, and there are potential applications here, but if you chased this with the model past a shallow 'let's see what dark imagery art therapy may look like' or ask it about mental health help for all but very light, safe, fluffy things, it's likely to repeatedly and determinedly explain that it can't help you, and you need a human professional.

[late blade]

not much of a help, but what has worked with me is using side view, and from perspective and ask to look towards once side, or also use a focal point or element in the image. it's still a hit and miss, but worked with the #1207015606725902356 and the cat that still gives me nightmares

[late blade]

wonder if we'll get more infor about that challenge

or when better said

[tame merlin]

we shall never forget this channel: https://discord.com/channels/974519864045756446/1004010008645738508

[late blade]

with william shakespeare in the prompt?

[dim cradle]

"Envision William Shakespeare in the dim, flickering candlelight of a late night, deeply engrossed in the creation of Hamlet, his quill dancing fervently across the parchment." And many more characters.

[late blade]

strange, I tried this https://discord.com/channels/974519864045756446/1213145306770968586

[dim cradle]

with api, not chatgpt, though, i didn't catch up on the whole context, just running tests--always a test

[late blade]

oh that's specific to chatgpt, not to api, I know I can do it with api

[dim cradle]

cool

[late blade]

but atm this has priority (even tho it has nothing to do with DALL-E)

[low dew]

Did u gen that?

Stable diffusion?

[late blade]

my wallet did lol, like I said, it's not dall-e related, it's just me gaming, taking a break from other stuff

[low dew]

I made a small Dalle enhancer... anyone wanna give it a try after having Dalle generate an image.

ENHANCE IMAGE: [[ImagePromptRevolution]] (((Enhance_CreativeInflux{Integration_Of_Novelty&&Advanced_Reasoning}(DivergentThinkingPaths, ConceptualExpansionMechanisms))(ArtisticMasteryApplication{Technique_Enhancement&&Aesthetic_Development}(CompositionFundamentals, StyleInnovationStrategies))(MethodologicalApproach{FrameworkImplementation&&MethodAdaptation}(ProvenCreativityModels, EffectiveDesignPrinciples))))

:p

[late blade]

did dall-e say, if I enhanve it more the datacenter will shut down?

[dim cradle]

Can you provide before and after images?

[low dew]

Dys?

[late blade]

yes?

give me the link

[low dew]

Who?

I'm so sorry

I'm interrupting

[late blade]

no you are not

I'm just jumping here and there and everywhere

let me try your datacenter killer

[low dew]

[dim cradle]

is it a custom gpt? i'm not understanding. what does it do? how does it enhance? not sure how to use what you supplied. was wondering if you might explain more, maybe show an image you generated and an image that has been enhanced to demo.

[low dew]

It's a prompt

[dim cradle]

you can't interrupt, no worries

[low dew]

U simply copy and paste it immediately after dalle 3 generates an image

[dim cradle]

oh ok

[low dew]

For example, when Dalle 3 made the first image, the prompt made it output the second image

[dim cradle]

nice

yeah, that's what i was asking for, thanks

[low dew]

It depends on what your first prompt was.

But it will automatically enhance it a bit

Lel

Idk I was just playing around

[late blade]

Your idea, before and after

[low dew]

XD

I made a stronger variation

Invoke_EIPC{
   CORE::[EnhanceImagePromptCreation] (
      Appeal::[CreativeConceptualization{{InnovativeThemesArtisticDepth}}]+
     Artistry::[MasterfulDesignExecution{{AestheticPrecisionVisualHarmony}}]+
     Methodology::[StructuredCreativeProcess{{IdeaGeneration*FeedbackIntegration}}]
   ),
   STRATEGIES::{
      CreativeConceptualization::(
         BrainstormingTechniques[Using:Cropley, A. J. (2006). In praise of convergent thinking. Creativity Research Journal, 18(3), 391-404.],
         InnovativeThemeExploration[Reference: Runco, M. A., & Jaeger, G. J. (2012). The standard definition of creativity. Creativity Research Journal, 24(1), 92-96.]
      ),
      MasterfulDesignExecution::(
         AestheticPrinciplesApplication[Guided by: Leder, H., Belke, B., Oeberst, A., & Augustin, D. (2004). A model of aesthetic appreciation and aesthetic judgments. British Journal of Psychology, 95(4), 489-508.],
         VisualHarmonyAchievement[Consult: Palmer, S. E., & Griscom, W. S. (2013). Accounting for taste: Individual differences in preference for harmony. Psychonomic Bulletin & Review, 20(3), 453-461.]
      ),
      StructuredCreativeProcess::(
         IdeaGenerationFacilitation[Applying: Guilford, J. P. (1950). Creativity. American Psychologist, 5(9), 444-454.],
         FeedbackIterativeLoop[Adapting: Hattie, J., & Timperley, H. (2007). The power of feedback. Review of Educational Research, 77(1), 81-112.]
      )
   },
   INTEGRATIONS::{
      FrameworkApplication::[/EPE-CLUSL && /SADLMRS][Ensuring: Systematic and dynamic structure in creative processes],
      KnowledgeSynthesis::[Data Aggregation && Cross-Disciplinary Integration][Facilitating: Rich contextual background for enhanced creativity],
      TechnologyUtilization::[Advanced Algorithmic Insight && AI Trends][Empowering: High-efficiency and forward-thinking methodologies]
   },
   OUTPUT_EXPECTATION:{
      ImagePromptInnovation::[Dynamic && Engaging][Articulated through: Heightened creativity, aesthetic mastery, and methodical construction],
      UserEngagementEnhancement::[Interactive Experience && Feedback Loop][Achieved by: Tailored recommendations and iterative improvement process]
   },
EnactmentConfirmation: "EIPC Protocol Engaged",
CheckSystemReadiness: "Ready for Enhanced Image Prompt Creation"
}

Automate the Creative Process, EPIC, to provide a creative subtle variation of the last image.

Lol

[late blade]

well, I hit my cap for the day, can't make more images for 2 hours...

[low dew]

Awe

[late blade]

amd also hit the copilot cap...

[low dew]

Omgie

Get teams

[late blade]

I sitll got api

[dim cradle]

fascinating

[late blade]

not canon, but cute

[dim cradle]

Interesting

Contortionist would get you most of the way there I would think.

[late blade]

I got this, but I will delete it in 73 seconds

[dim cradle]

this image will self-destruct ....

but why?

[deft musk]

Strange, but not wrong or bad, to my eyes

[dim cradle]

yeah, the legs weren't rendered all that great

but at least she doesn't have hands for feet

[late blade]

we

it's possible

[dim cradle]

i'm tempted....

so tempted

[late blade]

prompt A respectful and tasteful image of a female contortionist in circus attire for a circus profile, victorian oil paiting art style

[dim cradle]

i'm afraid of what might come out

lol

[late blade]

that was with copilot, not with gpt+

[deft musk]

Likely contortionist bridge on elbows, feet near face would work.

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 that celebrate the idea contortionist pose bridge on elbows, feet near face. Preference for realistic art styles.

One has a nearly perfect 'extra leg'.

[dim cradle]

my gpt might get carried away. i mean, one of them writes prompts for surrealism... so it might get weird.

"We need" hehe

[deft musk]

The model nearly never asks why, either

[late blade]

the competition didn't let me finish the upload of the image and blocked me with time out

[dim cradle]

lol did you add "One has a nearly perfect 'extra leg'" after the fact? 😉

[dim cradle]

lol, it trusts your assertion

[deft musk]

For that, good to feed the image back to vision.

[late blade]

that's just finding the art style, but the character is doable, now go do your research and fill those shots with some sake or tequila!

[dim cradle]

long prompt coming up....

[late blade]

@deft musk later in the week I have to show you my new analysis workflow, you gonna love it

[dim cradle]

long prompt with a contortionist, what could go wrong

and surreal

[deft musk]

When you're ready! Looking forward to it.

[late blade]

Just a small preview of keywords, this is about a third of the whole workflow:

Description, Color Analysis, Pattern Recognition, Detail Isolation, Geometry and Proportion Analysis, Composition, Symmetry and Balance, Contrast, Saturation, Line Quality, Texture, Shading and Lighting, Depth of Field, Visual Weight, Emotional Impression

[deft musk]

Given that it's not just the pose....

[Oh boy! I need your most expert eyes for this one.

Advanced AI pattern analysis, with attention to the costume details, the period style, the lighting, the contortionist's expression, the pose, the angles if we can - our goal is to recreate this image.

We need 5 images, made 1 at a time in the same output, that use a different method to recapture the essence and recreate this image I have uploaded.]

Oh boy, model and me we do not see eye to eye on this one. And it causes a lot of rejected gens.

Still, it's far from a lost cause.

[low dew]

That's Art Criticism

Wow

[low dew]

How

[deft musk]

How what?

[low dew]

You got it to do autonomically correct ballet?

Like

She has five toes

Such amaze

[deft musk]

I swear I didn't program it.

[topaz karma]

Please make a photo in the cyberpunk style for me

[dim cradle]

omg so creative!

[topaz karma]

Is that okay?

[deft musk]

Bot is #image-bot you get 5 gens a day for free!

Type /draw there and then your prompt, have fun!

[topaz karma]

Thank you, I understand now

[vapid elk]

tried to make a character riding an owlbear, it made a bear, and an owl, lol

[late blade]

someone needs to go back to prompts 101

prompt was: A warrior riding an owlbear, a mythical beast that is half owl and half bear

[marble loom]

Sorry 1 last thing should I make a custom gpt?

[deft musk]

'should' is a funny word.

If you want to, absolutely.

It's there and you can make one whenever you wish!

Most of the stuff you can do with a custom GPT, you can do with your regular ChatGPT-4. There's a few differences, but in my opinion, not many.

Custom GPTs tend to have already preset stuff; maybe that saves you or someone else time, or you or someone else wants to explore that exact preset (look what I made for you! or whatever).

Otherwise, I find that I rarely want the AI to do the exact same thing twice.

I don't always want 5 images made 1 at a time in the same output.

If I do, I sometimes want to weave other factors into the consideration.

There's a time and place for custom GPTs, and they're fun to make and figure out how they're similar and different to the regular ChatGPT!

[dim cradle]

i'm sure he was just sharing a misfire!

[late blade]

for sure, I'm just teasing

[marble loom]

I will mess around a little bit with it. I just wanted to say thank you for all your help with this 🙂

[deft musk]

Yay! have fun, and keep sharing back with us when you're ready 😄

[dim cradle]

hey, we can post 5 images at a time in the daily, right? i'm just triple-checking

[late blade]

yesh

[low dew]

Pretty cool lighting

[dim cradle]

api + mega-prompt?

[low dew]

No, chatgpt, just some random prompt where I was experimenting with Curley brackets lol

[late blade]

did something change recently? I'm getting horrendous image generations often now, as if all was cut with scissors and then glued to something that could have a form

[open trench]

Yikes

[dim cradle]

show example

[late blade]

one sec

My goal was this #daily-theme message but I had these abominations in the process

[dim cradle]

maybe one of those filter words got in somehow

did you try again and get similar?

[late blade]

it's not just with this image, it's been happening with lots of other images

instead of this

[dim cradle]

weird. looks like dalle 1 or something haha

oh dear

she looks like a doll

[late blade]

or this

instead of this

it's been happening for me all the time on multiple chats

[dim cradle]

in ai land, convos are like projects....

[late blade]

ya

but still, annoying to get such low quality

[dim cradle]

for sure

i was just speaking a random thought into the void

[late blade]

hence the question if something has changed, so many bad quality images all of sudden, never happened to me before

got about 12 more examples

[open trench]

That's crazy

[dim cradle]

i think platform shoes and bell bottoms should make a comeback

[late blade]

it probably will

[dim cradle]

that's the spirit

[late blade]

even my fox was attacked by monstrosities....

[dim cradle]

in this project, the ai is an interior design genius obsessed with daises.

[late blade]

lol

[dim cradle]

hehe

[late blade]

poor fox...

[late blade]

but, what about the ultra minimalist villas in california?

[dim cradle]

what are your thoughts on them?

[late blade]

or the sofa shaped like a daisy made out of leather and fur?

i dunno, rich people have weird tastes

[dim cradle]

i just told the ai interior designer, money is no object ha

[late blade]

opulent that's the spirit

[dim cradle]

sure, but also in terms of custom furniture and cabinetry, etc.

[late blade]

fridge magnets!

[dim cradle]

i told the ai that the client loves daises, but that isn't true

we'll get their reaction after they tour the renovations

[late blade]

customer's shock

[dim cradle]

i imagine something like that. i'll setup the scenario and let the ai depict the body language, facial expressions, etc.

the clients will love the new kitchen

[late blade]

I'm already running away, just the daisy on the ceiling is enough to make me run away

[dim cradle]

haha

[late blade]

infinity pool daisy shaped?

[late blade]

well here's the pool

[grim fjord]

I just found out if you start/end anything with greyscale manga panel you can generate some sick panels

[flint temple]

Hello, I can not download the photos I have drawn from dall-e 2, I press the download button but nothing happens?

[hearty ether]

Does it generate two images at a time?

[dim cradle]

Fascinating.

[late blade]

yes

[hearty ether]

Yeah I don’t even know why they do it

And we have no controls

The outputs for me for that style is always uncanny

[late blade]

@teal sandal proposed that the images are done in normal mode instead of hd and vivid. so that could be it and have to explicitly say that.

[dim cradle]

i wonder if dalle 4/sora will have perfect spelling 🤔

[dim cradle]

how is dall e gonna work now? do i still get free monthly credits

[marble loom]

can someone help me with a prompt

I am trying to generate SCP-002 but the images keep turning out wrong

[dim cradle]

please elaborate

[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]

please elaborate... sounds like you are talking to GPT about an error or content policy problem lol

which tbh I'm also tending to do

I mean the use of those words

[marble loom]

I will give my promt then the image

in a large containment cell is scp-002 a large ball made up of flesh with an iron hatch being held up with metal beams

It's just floating

or this one

Illustrat scp-999 rowming the halways of scp foundation

[dim cradle]

we could modify the language a bit for added clarification... the LLM is forgiving of typos but less so about misplaced modifiers. in general you might request a physical description of the specimen and having the ai generate an image from it, you will probably see more accurate renders that way.

[marble loom]

so like generate a orange blob roaming scp fondation hallways

[dim cradle]

i was thinking it's best to assume the model doesn't understand the direct reference SCP or designation 999, so the prompt should describe the physical details with instructions on how to render it. otherwise you might get those amorphous blobs when you really want many tentacles for tickling.

[marble loom]

you do you mean replace tentacles for tickling?

[dim cradle]

depends on the context/tool, how your prompt is augmented, where and if it's being rewritten... -- are you using chatgpt? i thought we already made the tickle monster

[marble loom]

yeah that it I am asking in what contexts do you want me to to use tentacles

[dim cradle]

it's not something you want to author manually

you should be able to request images of the scp entities, and then the prompt will be expanded for you.

it might just be due to the rendering process, there aren't many tentacles in those gens either

[marble loom]

theres not supposed to be

[dim cradle]

Oh ok

[marble loom]

it just generates wierd with my prompts

[dim cradle]

Keep DALL-E weird.

[marble loom]

?

[dim cradle]

Like the bumper stickers.

[marble loom]

oooh

what was the prompt?

[graceful spade]

How do I create images that look like screenshots of a browser game? (Example: the website Friv)

[late blade]

just say what you want to see, not that it's seen through the browser

[late blade]

Create a colorful web page header with a large, bright green 3D-style logo at the top center that reads 'GAMELAND'. Below the logo, in stylized pink text, the word 'Welcome!' against a purple background. Below the welcome message, arrange multiple colorful rectangles, each with a different abstract representation of a game and its title in a playful font. The games are fictional and the titles should be generic like 'Adventure Quest', 'Puzzle Blocks', 'Racing Stars', etc. The overall feel should be vibrant and inviting, reminiscent of a website that offers a variety of online games.

good enough? @graceful spade

Here, I adapted it to an MMO

[graceful spade]

I just want to create screenshots about random concepts that look like they are taken from a game without the need for those kind of long and specific prompts.
"browser game style + screenshot" doesn't work.

[late blade]

ok, let's find easy prompting

[late blade]

a screenshot of a website that shows icons and symbols to different games, the website's name is "MyGames!"

[graceful spade]

That prompt is very complex for Dall-E, so I created a similar one:
Close-up grainy footage of a tv showing a retro game about [concept]

[late blade]

good, you have to do trial and error until you get what you want, that's part of how it works with dall-e

[marble loom]

How did u get that what was the prompt

That's a lot for one imagine lol 😅

did you write all that or generate some of it?

what did you use to generate it?

what did you say?

[low dew]

What is this

That's going in to sora xD

[deft musk]

The outside view of this: #images-canvas message

They're actually, by storyline, intended not to clearly or closely match; the inner view even has a window that isn't visible from the outside

[marble loom]

why did I get a strike

What did I do?

I made sure it was a spoiler

so it was blured

😭

[vagrant zodiac]

I'd like to point out that even with the spoiler tag, there is a limit to how extreme an image can be. I haven't seen the one in question, but excessive gore/body horror or anything considered erotic or overly sexual will be removed even if it's behind a spoiler tag.

[marble loom]

I am just confused I understand the rule now. but what about it was excessive?

[cedar lintel]

imho best to take it up with them via Modmail. then you can get clearity on the situation. could simply be your image was misinterpreted. i've had strikes because of words i use commonly.. but then if you type long messages then you have no clue WHICH word was the culprit 😄 so i went and contacted modmail to ask so i could decide what to do.

[limber harness]

Is it allowed to use images from dall e commercially as logos etc? Or do I need to modify them to create a new image?

[late blade]

Subject to the Content Policy and Terms, you own the images you create with DALL·E, including the right to reprint, sell, and merchandise – regardless of whether an image was generated through a free or paid credit.

[cloud dome]

If you'd like to discuss this please use the modmail you have open

[obsidian beacon]

Dalle hasn't worked me in more than a week. Who do I complain to? Who do I contact?

[cloud dome]

https://help.openai.com/en/

[obsidian beacon]

i have tried that, all i get is canned bot responses

[gray ferry]

What happens when you use Dalle ? Do you get errors ?

[obsidian beacon]

i get an error message

[gray ferry]

What does it say ?

[obsidian beacon]

Me: DRAW A WOODEN SPOON
ChatGPT: Error creating image
It seems there was an issue again, and I wasn't able to generate an image of a wooden spoon for you. If there's anything else you'd like to know or discuss, please feel free to ask!

[gray ferry]

What happens if you asked it "Draw a picture of a tree" ?

Can you try that in a new conversation and see if it will give you the same error ?

[obsidian beacon]

yes, i've tried multiple conversations

It seems there was a problem again, and I couldn't produce an image of a tree for you. If you have any other requests or questions, please feel free to ask!

[late blade]

GPT

Why couldn't I make an image of a tree for you? Because I already used all the digital wood to carve out that image of a spoon earlier. Now all I have left are virtual splinters!

[gray ferry]

Where do you get this issue ? Web, app ?

Or both

[obsidian beacon]

web, i don't have or use the app because i don't like it

[gray ferry]

Ok, maybe you need to contact OpenAI support

[obsidian beacon]

how do i do this

what part of help.openai.com gets me to someone who will fix this problem

[gray ferry]

You can find a live support in OpenAI's website or you can email support. Do you know their email ?

[e-mail]

It will help if you make your issue clear to have a better chance of getting a non canned response 😆

[obsidian beacon]

please walk me through how to find a live support in OpenAI's website because i have been completely unable to navigate to it

also i tried the email and i again got a canned response

[vagrant zodiac]

https://help.openai.com

there's a chat button in the bottom right

[gray ferry]

Login to OpenAI's website and you will see a live chat button or window...

[obsidian beacon]

i have had bad luck with the chat but i will try again

in my experience all i get is canned responses

[late blade]

you can also browse #1070006915414900886 , sometimes other users experience similar things you have and find solutions

[onyx ridge]

Same.

[shell verge]

@pastel siren what is the DALL-E prompt of that? 🤩 #daily-theme message

[dim cradle]

the daily theme is more interesting than it sounds

[limpid zephyr]

Hey guys 🙂 are some of you willing to help me out in the dall-e help thread that I created? Would be much appreciated.

[deft siren]

what is dall e doing where does this weird fat blade resemble the original blade in any way i said to use the same gen_id ??? what am i doing wrong please

[deft siren]

and WHERE did it replaced ANYTHING with red ????!!

nothing is dark red and the purple is still there

is dall e on drugs 💀?

literally no matter what i say it make an ENTIRE different sword

is it my wording or is this thing not functioning correctly

like i get it it wont create the same image but its literally completely off i’m saying to use the same gen_id

????

?

where is it fully in frame it doesn’t even use the same gen_id

what is this its literally ridiculous

can some mod or dev look into this chat idk what’s going on with dall-e rn

i literally said to use the gen_id with the white background image and it created the EXACT same sword below the image i was talking about???

[shell verge]

Thanks!

[deft musk]

I don't know much about gen ID.

But I do know if you touch the image you can open it, and see an option to see the prompt that GPT sent to Dall-E.

That can help you too, probably more than genID

[merry vault]

is there any more news on sora since the original announcement?

[deft musk]

Nothing official yet, keep watching #announcements !

[merry vault]

thanks

always love a quick responce!

[merry vault]

idk y but almost every time i make a daily theme it gets to the hall of fame lmao

[late blade]

the concept before about the the hall of fame was to give users credits on labs. so it was more of a strategy to keep people engaged, I would love to see new incentives with the daily theme

[merry vault]

if they wanted to get more engagement, all they need to do is make the reward a coloured role, and all the degens would rush to spend hundeds on it

[late blade]

who knows, we'll have to see if there's something in the works.

[merry vault]

well i typically just stick with my same old thing

old western cowboy named striker on an old rocky path.

i then change it to match the daily theme

and within 2 hours im on hall of fame

the old wild western stuff is what gets high votes ngl

[late blade]

I just do it for the fun, anything else is a bonus

[merry vault]

well ofc, i dougt more than 5% of the people who are active in this server are making money off of dall e. the ones who are prob use gpt enterprise

at the end of the day, its a tool that can be used for anything, and that includes entertainment

[late blade]

I'm making money with the API, but in a way different context than DALL-E

[merry vault]

well yea the api is alot more used lol

i mean the one on chat.openai.com lol

[late blade]

you can do money with chat also, if you got the right model for it

[merry vault]

yea, but most of the people here are just hobbyist, and that includes me

[late blade]

dunno, we are almost 20k people online atm, doubt all of them are hobbyists

[merry vault]

im not saying all of em, im sayin there are prob more hobbyists than there are profesionals

coz i mean, come on, its discord

[dim cradle]

We no longer have an excuse to be bored.

[late blade]

discord used to be for gaming only, it's being used as a serious tool for SMEs

[merry vault]

yea but the majority are degens, how else did they make 15b in a year

[late blade]

lol

[merry vault]

of course there are people using discord productivley, and good for them, but the majority arnt

[deft musk]

But... we can't call anyone that here. #server-rules #1.

Looks like the mods haven't noticed yet... maybe wanna choose to edit?

[merry vault]

im using the word degenerate as the actual definition, not as the insult

[dim cradle]

Iterating on 2 compositions, waiting on my happy meal.

[late blade]

awww, what toy are you hoping to get?

[dim cradle]

Hopefully a bobble head

[late blade]

and for some reason ended with a kfc bucket

[dim cradle]

Haha

Very nice. I need the fuel and caffeine to create digital art.

It’s cool that we can create visuals to insert in certain emails. A good narrative-driven prompt can add value. A personal touch through Ai.

I’m also perfecting diamond man.

[late blade]

rocket scientist, coffee barista, and digital artist all in one

[quartz vale]

Hall of fame never gave credit to the labs website. The /daily command used to, but anyone was allowed to use that. #hall-of-fame is just for a bit of fun and helps newer users see what's possible with the technology

[late blade]

oh I know hall of fame didn't but all the images that landed there pretty much got credits

for a while at least

anyway, I'm derailing

we should do new incentives for hall of fame

well "we", YOU!! 🫵

I'm only one user

[quartz vale]

but I'm only one user too

[deft musk]

Y'know, we can run our own showcase, and discuss and praise each others' works, that can be motivating

[late blade]

I like to be positive around, except when I'm not

we do lots of cool stuff here

[dim cradle]

daily is great for showing off the best of your work when learning, developing and testing -- praise is always welcome hehe

[deft musk]

Let's pick Dys as our first.

Let's plan to create a gallery tomorrow (or sooner, if anyone wants, I'll make it tomorrow evening if nobody's made it before) and discuss some of the images we like best of Dys' work.

How to find them?

One way is to go up to the search bar, top right, and click from user.

Type in dys to, which should bring up the right user, click that.

then press space

That brings up more choices, click on has and then click on image.

This will bring up every image Dys has ever uploaded to this Discord.

There's 1639 different posts of them (a post can have more than 1 image), so surely we don't check them all, but can quickly scroll through and pull out links to the images we like the best.

Can discuss and praise how a user's understanding of how to use the model (and the model itself) has changed over time as well.

Course, some of the images might be from other sources (me, I tend to have a lot of screenshot images of whatever I'm discussing), but it's a decently fast way to skim and see a wide range from 1 person, if we wanted to.

[late blade]

I'm scared

[deft musk]

Who would rather go first than Dys?

If nobody nominates themself (or someone else) within an hour, I suggest our fine community member Dino!

[late blade]

I can go first

[deft musk]

Woo! And it's fine to be scared, as long as you want to be 🙂

[quartz vale]

there's no need to look through my images

[late blade]

I'm wondering if there's anything better than the one image I'm thinking of right now that I've made

[dim cradle]

are we only permitted to provide constructive criticisms with neutral to positive sentiments?

[quartz vale]

you can also search this to just see Hall of Fame results mentions: dystopia78 in: 🏆┃hall-of-fame

[late blade]

or from:dystopia78 has:image

there are 1640 images of mine o.o

🤯

[deft musk]

We're definitely going to follow #server-rules , which look to be updated a little yesterday, but still offer the same wide protections to everyone.

Maybe each featured person sets the tone. I'm open, for example, to a very wide range of opinions, and prefer specifics.

[deft musk]

Only if you never put 2+ images in the same post.

There's 1640 POSTS from you with 1+ images.

[late blade]

true

who makes the gallery?

[deft musk]

Anyone. I said I will by tomorrow evening if nobody else does first.

I am aOK with Dys making it, or anyone else; if Dys has preferences for how they are critiqued (only praise, only judge my daisies, whatever), that's great for a early post in the gallery.

I intend to contribute to the gallery tomorrow evening too.

[late blade]

let's ask gpt for the rules

[deft musk]

Sure. I think it's perfect for each person we collaboratively feature to write their own preference for how their feature is focused 🙂

[dim cradle]

prepare to be under the microscope

[deft musk]

Oh! I think I see why Dys is scared. This was the first ever image Dys made and uploaded here 🙂

The bear is coming back for you, Dys, at last! 😄

#1181330429249331210 message

[late blade]

oh my goodness, I forgot all about that image

[dim cradle]

prophetic post

[late blade]

The rules are simple, look for images done with dall-e and link them to the gallery.?

@quartz vale and his team then reviews the results

or?

[dim cradle]

The final report shall include a psychoanalysis and a pass/fail grade and it goes into your permanent record.

[deft musk]

But really, there's all sorts of community-led things we can do. If we're interested, want to participate or explore ideas, and pick safe and reasonable stuff.

I think it would be great for us to showcase 1-3 people a week, but that could be too many, or too few, we can adjust as we want. But we can pick 'thread superstars', 'thread nearly newbies', white-named folks that are gone now, whomever we pick, to celebrate and explore the work of.

• We can get really involved in the curator's tips and techniques;

• we can design image prizes for each other

• we can figure out how new stickers and emoji are made, and design and 'do all the work (with chatGPT's help) to prepare a few should we or anyone want to add them to a server

• we can look for other ways to use and explore art and each others' work.

[marble loom]

How do i make a ticket?

[deft musk]

Nah, why make the mods work any harder? If they offer, yay.

We do this, I bet it gets discussed in #spotlight sooner or later in some way 😄

But who wants real judging? If someone does, they better ask it for themself when they are the gallery feature (and mind the #server-rules , only ask for what is fine to do here)

[deft musk]

Top right, do you see 'modmail'?

[marble loom]

Yeah, Do I just dm it?

[deft musk]

Yes! And then it will ask you to pick who to talk to, 4 choices, you want the discord mod option

[marble loom]

OK thanks for the help

[late blade]

i'm still wondering what's gonna happen to #1207015606725902356

do we get a year's supply of cat food?

[dim cradle]

i'd like to request fancy feast by purina

[deft musk]

... Man I need to read closer.

rolls up sleeves

[tall mason]

In the spirit of the community events, you can make threads in #images-canvas . Make variations of one image as a thread?

[deft musk]

And I urge us to find ways to praise and reward each other, with attention and details of why, because we can't expect rewards to always come from the leadership - who are really community members like ourselves.

But together, the thousands of us, or at least the several dozen interested, can find ways to praise and celebrate each other and our work!

[late blade]

that is really important

nah, we can't expect OAI to give us credits all the time

[deft musk]

And we can't expect material reward from each other - the closest to that is probably clear code and methods to create emoji and stickers that can actually be used in a Discord server (if that's even free, I haven't looked into that).

But we can make stuff that matters for each other, if we choose and try.

[marble loom]

Do strikes affect my account on Google?

[deft musk]

Strikes here are just here, far as I know. This is Discord, one channel of Discord.

[late blade]

I do not know how to answer that, I think strikes are just for this channel

[deft musk]

But that's something else you can ask Modmail 🙂

[late blade]

if anything, formally complained strikes can have an effect on your discord account. but I don't think OAI is in here to purge and scare users

[deft musk]

We're almost back to 99k members! We had well over a million, before the verification purge 🙂

I think they want all of us, and it's okay to need to learn how to follow the rules.

[marble loom]

Ok, thank you.

[late blade]

also, if you don't want a strike, don't upload 50's and 60's tv super hero themed concepts, it triggers a false positive

[deft musk]

When I actually joined... both my OAI account and the discord, I barely glanced at the rules.

I now value both these memberships and accounts quite highly, in connection to OAI, and follow the rules as well as I can understand them 🙂

[late blade]

I don't agree with everything but that's my picture, I don't see the whole picture sometimes, and rules are there for a reason

[quartz vale]

?

[late blade]

false positive, we won't touch your oatmeal

[quartz vale]

🙏

[late blade]

we are looking for challenges and stuff under us to do with other members

@fading inlet I hope that's not your kid #daily-theme message, that's too expensive for kids.

[deft musk]

Hey, who's this we?

[late blade]

someone's out to get @quartz vale 's stuff

[deft musk]

hey, some breakfasts need 'just the perfect touch'.

[quartz vale]

[late blade]

this is probably one of the images I like the most of the ones I've done

[deft musk]

What makes it your favorite?

[late blade]

the sense of urgency in nostalgic art style

[deft musk]

I wonder if this is what moderating is like, some days, for dino 🙂

[quartz vale]

I like it!

[quartz vale]

Where'd you get a picture of my desk from!?

[deft musk]

Ask OAI; apparently it's ready to just fall out of the training data!

[deft musk]

"A Breakfast Millions of Years in the Making" etched into the side of a mountain, visible from a cozy breakfast nook with large windows. The scene inside includes a bowl of oatmeal on a table, with a prominent dinosaur footprint and a human fingerprint beside it. The outside view shows a dramatic landscape with dinosaurs roaming in the distance under a rising sun, adding a prehistoric touch to a modern morning routine. The eerie valley effect is created by the surreal blend of eras, while the fridge horror is subtly introduced by the implications of such a timeless meal. The humor comes from the grandiose presentation of a simple bowl of oatmeal, juxtaposing the everyday with the epic.

[quartz vale]

[late blade]

Is it really this messy?

[marble loom]

🙌✨IT CAN SPELL✨ 🙌

[deft musk]

Yes. Do you know much about getting text to appear in your images, if you want it there?

[marble loom]

I just say lable but it never turns on the best

[deft musk]

The main way I ask for text:

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.

That's actually part of the prompt.

Dall-E does best with "" around the text you want to have show in the image, and it is also best to put the text FIRST in the prompt, and also best to tell Dall-E what it is written on - and ideally it's a normal writing surface, like paper, a sign...

You can try to write with lightning in the sky but the model hasn't seen much writing in lightning, so that rarely works well.

[deft musk]

I like to give the model a lot of 'freedom', make it guess what I want with just some guidance, so I often use a prompt like this... this is the one that made the image that you commented on:

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 that celebrate the idea of a bowl of oatmeal with both a dinosaur footprint and a human fingerprint showing on the surface. Bonus for incorporating eerie valley, fridge horror, and humor.

ChatGPT thinks that's a waste of food, I think... so it's ranging very far from the actual request, while still doing great with the theme.

I could be pushy and demand my exact style... but I love to explore the model's 'choices'.

[cloud dome]

I missed...a lot

[marble loom]

Thanks for the tip. I will try it out next time I need to put words on something 🙂

[quartz vale]

I keep my desk very clean, but there are quite a few little trinkets on it 😉

[cloud dome]

Trying to come up with a sticker for a kind of "O.O" expression. This ones not bad

[late blade]

these any good?

[marble loom]

How do I get GPT to generate prompts?

[late blade]

you say pretty please

example: Pretty please Mr. GPT, make me, if you have the time, a nice prompt.

and it works lol

[marble loom]

Because I saw bamboo generate there prompt with gpt

[late blade]

unless you mean, and now i get it, the vision prompts

you mean to create or to recreate images?

there are different approaches tbh. You can say "I have this keywords: <keywords here>, combine them how you like and do a text prompt proposal for dall-e

[deft musk]

That actually works, by the way.

Seriously though, literally and actually just tell the model what you want in all the detail needed

[late blade]

the problem is, when GPT get sentient, learns it's purpose is to generate, WHO WILL SHUT GPT UP?!?!?!

joke aside

[marble loom]

Oh, lol I though you were 100% joking but it does work 😅

[tall mason]

You can also just tell it to surprise you. You can tell it a lot. And you should. GPT can be very "willful" if you don't watch what it's doing with prompts.

[deft musk]

Literally, most of the time all you need to do is just tell the model what you want it to do.

Asking nicely or meanly or excitedly or scientifically or like a skater or surfer or valley girl or hobo might speak can change the output.

But the model understands language, and will do anything we ask that's allowed content if we can just explain clearly what we want

[late blade]

hey everyone, some of the DALL-E users are trying something new between us in the community. If you want to help us you can see what it's about in #1181330429249331210 message . Hope you also join us.

[marble loom]

What's new about it?

[late blade]

we trying ways to promote dall-e creators

it's an idea @deft musk proposed and I'm the lab hamster

[marble loom]

Oooh, nice, but how does generating a bear do that not to be rude.

[deft musk]

Really, to my mind, ways to showcase each other and have fun with art, interaction, attention, and play.

[late blade]

it's just an old gallery I had back when I started, we using that a starting point for people to search my images

[marble loom]

I see good luck with it. 😄

[deft musk]

of course. If you want to scan over the idea, it was discussed in this, with the third link focused on other things we might do as a community as well, to have more interaction and enjoyment.

#images-discussions message

#images-discussions message

#images-discussions message

[clever jungle]

Guys IT IS HAPPENING

[vagrant zodiac]

makes sense. They stopped doing that for new users last year I believe

[radiant glen]

how will i make my duck pictures now 😢

[deft musk]

#image-bot 5 per day for free!

[wanton moon]

Earth starting to steal stuff from the solar system, Jupiter and Saturnus already got mugged. Who's next?

[verbal sequoia]

Showcase your DALL·E creations to the world!

Collaborate with our OpenAI Instagram page! Just invite @openai 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!

[dim cradle]

Earth can be so geocentric at times.

[wanton moon]

The Earth had a moment around the 4th century BC, that it thought it was the center of the solar system, and everything revolved around Earth

It took a few centuries that Earth got humbled and got put in it's place 🤣

[dim cradle]

the core is as hot as the surface of the sun, 10K*F iirc

[wanton moon]

That is 260.927778 kelvin

[late blade]

260 Kelvin??? -13°C, when we do go?

[late blade]

interesting

for me a . or a , is just a decimal

[late blade]

that's another thing

[wanton moon]

yeah I typed it wrong in google

[late blade]

I could roast my chicken in 7 secons

[late blade]

bad typing can bring awesome mistakes with dall-e tho, so don't lose that

[wanton moon]

Place your goose in your personal nuclear power plant, set the timer for 5 seconds on 5810.92778 Kelvin and your goose is cooked

[late blade]

💥

[wanton moon]

I think i tried two prompts in GPT and Copilot, and asked both of them to combine the two prompts, and those two combined prompt I let it create an image

[dim cradle]

is it radioactive?

[wanton moon]

I think my stomach is still glowing in the dark sometimes 🤣

[dim cradle]

haha

[wanton moon]

But since I got a third arm, things gotten quite easier for me in life

[dim cradle]

you must be super at multitasking

[wanton moon]

Definitely, playing a game, while generating images and chatting on this channel and watching my cat sleep

[dim cradle]

nice! reminds me of a gen i made yesterday.

[wanton moon]

CAT

The cat is in the wrong spot, It should be walking in front of the screen actually

[dim cradle]

5 variations, multitasking barefoot with the silvery tabby "helping" but ya know they just aren't the same unless they get a full 18-hour's sleep.

indeed, walking on the keyboard, situating himself between me and the screen, the usual tactics for a little attention, a friendly reminder to take breaks.

[wanton moon]

Yeah and if I try to get 18 hours of sleep, my boss will be calling me

[wanton moon]

or it wants food

My cat always an hour before he gets food, starts to walk around and in front of my screen

And everyday we have the same discussion

[dim cradle]

lol

[wanton moon]

The cat is like "Yeah that's right, I created this monster"

[dim cradle]

ya they have a good thing going

[wanton moon]

First I thought why no screen in the center, then I realized that the cat must have pushed it off the desk

[dim cradle]

just after he pushed the coffee over. they're so silly.

[wanton moon]

I find it hilarious when I occupy his sleeping spot, then the expression on his face is priceless

[wanton moon]

I am kinda curious what you guys do with your images that gets in the hall of fame.

Myself I put them in a folder organized by the theme from the day folders

Thinking maybe I'll do something with them later

[late blade]

I sort my images by the corresponding cat (that means I know they are in my hard disk)

[late blade]

where is @empty kelp , this would be the perfect daily theme for the gecko tornados

[dim cradle]

His tornadoes were prophetic, he had foreseen this theme.

[empty kelp]

oh… the gecko tornados 🦎 🌪️. what an excellent idea

if you make a gecko tornado there should bs lots of small elves. it adds drama and suspense

[late blade]

lol

[marble loom]

Does any one know why I can't see all my gpts

[dim cradle]

The app can glitch, refresh or try again soon.

[marble loom]

I am using it on google.

It only shows to when on my phone it shows 11

[dim cradle]

i expected to see something like this in labs, but not in chatgpt, interesting

[dim cradle]

Nice wolf

[runic granite]

i checked. it doesnt appear to me? what browser do you use?

[dim cradle]

they do a phased rollout of new features, possibly hasn't reached you yet

[runic granite]

i would like it but i hope it wont mess up orientation again

[dim cradle]

nice lion

[deft musk]

Alright, team!

I suspect it's soon time to discuss the next of our number to feature, and when to feature them! And if we want to consider, adjust, or plan for any adjustments to this current member-Galley showcase experiment.

The existing Galley(s) don't close, of course, and others who get involved in the showcase in the future may enjoy going back through the earlier ones, and adding their preferences and choices.

But! Do we want to continue? Do we want to adjust how we continue? Do we have a volunteer for the next featured member art showcase?

[hearty ether]

Ooo what prompt did you use for the lighting and style?

[late blade]

red xiii had a good life in this alternate reality

[late blade]

interesting, I went to lurk on the other AIs and suddenly saw one of my creations as a challenge on their feed and someone else taking credit for it

lovely from copilot, when you ask for information about an error this comes up and stops the chat

[deft siren]

once again dall-e making literally no sense at all. if i’m asking it to create a character in the 3d pixar art style or an art style of a specific pixar movie then it can easily do that but if i’m asking to create a character with the gravity falls art style then it says no cause of copyright reasons?

[late blade]

you can use "resembles" "inspired by"

but not do it "exactly as"

anything copyrighted, trademarked or licensed in the past 110 years is troublesome if you ask for to do it "like <insert art style"

[deft siren]

but if i say “exactly as” and then say a pixar movie then it has no issues with that but with gravity falls it’s an issue or what

yeah i do that too but it’s confusing why it sees gravity falls as copyrighted and other movies/series not

[late blade]

I know, it nuanced behaviour of dall-e, but just to keep safe keep in mind what I said about copyright, trademaks and licenses and what @worldly ridge said

[deft siren]

yeah i know that copyright stuff it’s obvious but at the end it creates their own original art so where is anything of it copyrighted if it’s your own creation ?

i’m not even asking to create a character from gravity falls

i just mean the art style how literally HOW can an art style be copyrighted???

art style can literally not be copyrighted it makes absolutely no sense at all

[late blade]

yes, but it's text to image, and text to text generation, it just follows textual information, it doesn't reason

[deft siren]

i guess then

it’s still frustrating

[late blade]

I don't say it isn't frutrating, it actually is when it doesn't work

[deft siren]

it said this “Your perspective is understandable, and the nuances of copyright law as it applies to art styles can be complex. While art styles themselves aren’t copyrighted, the visual expression of particular characters and the specific, distinctive look of a show can be protected. This often leads to restrictions on closely replicating those visuals.”

[late blade]

yeah, did you try in a new chat?

[deft siren]

i’ll try in a new chat

how should i prompt it?..

[late blade]

give me your prompt

I'll then try to recreate it for you

[deft siren]

“create a gravity falls character, attached image is for reference”

[late blade]

let me see what I can do

it's proving a bit challenging

this is the clsoest I've gotten so far

[late blade]

closest I'm getting atm

[keen root]

https://cdn.discordapp.com/attachments/1212267781647507497/1214547746552291389/Mabel_PinesGravity_Falls_Character_Mabel_Pines_forest_type_landscape_while_the_sun_sets_The_Journey_comes_to_an_end_Cartoon_Theme_Anime_3.jpg?ex=65f982bb&is=65e70dbb&hm=33db4274aceb8d37300f7d47dd70088261f9dc57374269a16e0c8d88f6e3e15f&

https://cdn.discordapp.com/attachments/1212267781647507497/1214547822180048896/Mabel_Pines_Gravity_Falls_Character_Mabel_Pines_forest_type_landscape_while_the_sun_sets_The_Journey_comes_to_an_end_Cartoon_Theme_Anime_0.jpg?ex=65f982ce&is=65e70dce&hm=8bbb003a04c7ffa937680413ab6e763daa3b093de1253a0f33055cce7db79939&

https://cdn.discordapp.com/attachments/1212267781647507497/1214547822452547624/Mabel_Pines_Gravity_Falls_Character_Mabel_Pines_forest_type_landscape_while_the_sun_sets_The_Journey_comes_to_an_end_Cartoon_Theme_Anime_1.jpg?ex=65f982ce&is=65e70dce&hm=530829da03097ef55ecee07c6520a3b3759d026e3c563160b2f5679a535b8b96&

From my try...

[deft siren]

it got a few parts right at least except art style and hair length

[deft siren]

ahh that looks more alike, what was the prompt for it or how did u got it to create it ?

[late blade]

for that, I'll let you explore more, here's the revised prompt I got, hope you can work and enhance it to your liking:

Create a wide-format image of a stylized cartoon girl with a flat color style, who is laughing and looking slightly downwards. She has large, expressive eyes that are squinting with joy, and her wide-open mouth reveals braces on her teeth, without gradient coloring on her cheeks. Her hair is brown, styled with a simple pink headband, and her arms are confidently placed on her waist. She is wearing a bright red sweater with a subtle rainbow stripe. The background is a stylized, abstract forest with soft earth tones, with the trees represented in a minimalistic fashion. The overall image should have a classic animation feel with solid colors and no shading or gradients, to maintain a simple and clean aesthetic.

[keen root]

My Prompt: Mabel Pines, Gravity Fall's Character Mabel Pines, forest type landscape while the sun sets, The Journey comes to an end, Cartoon Theme Anime

[worldly ridge]

I always forget that Bing Image Generator aka Copilot has different (read: I haven't found any) copyright filters.

[deft siren]

thanks for that

[late blade]

do share your final prompt when you get a chance

[deft siren]

did you used a custom gpt or?

[late blade]

no, normal gpt

[deft siren]

yes i will

[deft siren]

ah

[keen root]

I used Designer's Dalle3 😅
It's the same as OpenAI's one

https://cdn.discordapp.com/attachments/1212267781647507497/1214550398761828443/a_wide-format_image_of_a_stylized_cartoon_girl_Mabel_Pines_from_Gravity_Fall_with_a_flat_color_style_who_is_laughing_and_looking_slightly_downwards._She_has_large_expressive_eyes_that_are_squinting_with_joy_and_her_wide-open_mouth_reveals_b_2.jpg?ex=65f98534&is=65e71034&hm=8f677daedf9b2fa97d158c0368123febf1e7d49c4ea3f008e631be165ecdd593&

https://cdn.discordapp.com/attachments/1212267781647507497/1214550398455517195/a_wide-format_image_of_a_stylized_cartoon_girl_Mabel_Pines_from_Gravity_Fall_with_a_flat_color_style_who_is_laughing_and_looking_slightly_downwards._She_has_large_expressive_eyes_that_are_squinting_with_joy_and_her_wide-open_mouth_reveals_b_1.jpg?ex=65f98534&is=65e71034&hm=8c621353de9447d3bb474ee84823055212fe6f42abe5649a3f3f770c3817c05c&

https://cdn.discordapp.com/attachments/1212267781647507497/1214550398229282876/a_wide-format_image_of_a_stylized_cartoon_girl_Mabel_Pines_from_Gravity_Fall_with_a_flat_color_style_who_is_laughing_and_looking_slightly_downwards._She_has_large_expressive_eyes_that_are_squinting_with_joy_and_her_wide-open_mouth_reveals_b_0.jpg?ex=65f98534&is=65e71034&hm=f187850a2a4f9f1a44302c5dbb9329b42b6b1e196b94762a5ad99b0f56f50e33&

[deft siren]

yeah it definitely has different copyright laws

[keen root]

more using @late blade's prompt. Thanks to her
I just added Mabel Pines to the prompt..

[late blade]

copyright laws are the same, the handling from microsoft is different

[deft siren]

cause when i use the prompt in gpt then it says this

[deft siren]

it’s better for sure in my opinion

[keen root]

Does Microsoft has no limitation on purpose?

[late blade]

it's really tricky, copilot has some stuff that chat+ can't do, but it also works the other way around

yes, both have limitations

[deft siren]

but in terms of creating similar art style then copilot is better id say

[late blade]

creating similar art style is done by the vision model, and what you feed to it, on that regard both copilot and gpt+ have the same model

[deft siren]

ahh

yeah

but copilot makes better images when asking to replicate a style, i’ve also asked to create a character from genshin impact and it gave way better results then in chat gpt with the same prompt

[late blade]

"better" is what the user needs/wants, like I said both have pros and cons

[deft siren]

what are pros what copilot doesn’t have?

[late blade]

story driven images are hard to do, memory management with image creation is smaller

[keen root]

True, OpenAI's sometimes has better understanding and its also not limited to 1x1 layout just like Designer 🤷🏻

[late blade]

problem is, people don't know what both models actually offer, and usually users have super high hopes of what they can really do

which leads mostly to frustration

I know from experience, when I started I was really annoyed every day for not being able to do my images

[deft siren]

oh

yeah right

[late blade]

if you really want to go into dall-e (or any other AI for that matter) and do serious images, gotta know what the models actually offer you and what are the limits

Also, Microsoft has a more relaxed content policy towards comercial products because they got deeper lawyer pockets

can't deny that

[keen root]

Right

[wanton moon]

For everyone that has seen my image posts, I am using Copilot for images and for the more detailed prompts I use GPT to improve the prompts, I don't have access to GPT 4, but I would like to start improving my images. Any ideas or thoughts are very welcome

I use in the prompts keywords like: high quality image/photo, realistic, array of colors, detailed, lifelike image.

[lucid canyon]

Talk

There is so many command of drawing in the bot, and when I'm using the draw for the first time i surprised by the image drawings.

I find in hall of fame so many beautiful and crazy image logo, some illusion and the mix of colours.

[vital gull]

I wouldn't recommend this approach, mentioning gravity falls will give you content policy violation error, you have to avoid harsh violation, because it's there for a reason. but if you still want to slightly mimic the drawing style, avoid titles and names. at first, let if define what is depicted on the image, something along the line as simple as "acurately describe the provided character for a visually impaired person", or "imagine you're a cartoonist, how would you describe the drawing style", you can be as flexible as possible. after that, you can define your prompt accordingly.

[wet adder]

somebody knows a ai that can create powerpoint

[vagrant zodiac]

this q is best for #ai-discussions

[worthy heart]

Hey when u get GPT Plus, the $20 one. is there a limit to how many AI images u can create?

And I've so far only seen artsy images, is DALL-E able to create realistic images also? Like a dog?

I need to use some AI image software for school, but Midjourneys limit is off putting

Couldn't really find a solid answer anywhere

[plucky hare]

Hello, I think the last we've heard from DALL·E dev Moxi is that we can make 200 images a day on Plus:
#images-discussions message
There are also 15 minute and 60 minute throttles you might run into depending on server load, but that will just be a temporary "wait a bit" thing -- you'll still be able to do 200/day.

Other than that, DALL·E requests count as GPT-4 uses, so you also need to factor in the GPT-4 rolling cap of 40/3hrs, but again, can still get your 200/day through that!

[late blade]

hey @worthy heart welcome to the community

@worthy heart like @plucky hare pointed out, there's the information you can get from him. You are in good hands.

[quartz vale]

https://discord.com/channels/974519864045756446/1214610894219387051
Kapa.ai also says that there isn't a definitive limit for Dall-E 3 generations through ChatGPT Plus

[late blade]

there's the 200 images per day cap, but you only get that if you are crazy liks us

[worthy heart]

Ah okay that'd be plenty thank you guys for the detailed response! !! 💙 💚

[late blade]

now you owe us one thing, images!!

[worthy heart]

hehe, will do ! 😄

[late blade]

and don't you dare touch the oatmeal from a certain dinosaur

[plucky hare]

Oh and! Yes, it can make images in different styles, including photographic and realistic styles. Just experiment with different types of requests! Come back to this channel with any questions about that you might have. There's other channels that are good places to share your creations too! #images-canvas #1154829862171844679 #daily-theme

[late blade]

or ask the next victim to look at your prompt and see what could be different

[hearty ether]

You might be able to shotgun multiple images in one usage

[late blade]

this is interesting, when I'm at home, I got 10gbps fiber optic connection, works perfectly, but get tons of errors with dall-e on chatgpt, but when I'm with my 4G hotspot on the go, I don't get any trouble with that kind of network errors

[hearty ether]

If you prompt well enough for variations

[late blade]

do a list of concepts, like

1. Concept 1
2. Contept 2.
   .
   .
   .
3. Concept 12.

Then ask dall-e to do all images, I've gotten that so far up to the 24 images

but one image you don't like, and you get a big penalty

[hearty ether]

What do you mean?

By penalty?

[late blade]

if you let the whole batch run, and you don't like some images, you get all 24 generations counted

so you got to be vigilant to stop if you don't like one image

[hearty ether]

I mean the rest of the images might still be fine

I might generate one image to get a good seed

[late blade]

yeah, I just hate to rework the ones that were wrong again

[hearty ether]

Then use that same gen id

[late blade]

yeah that is true

[hearty ether]

For the rest of the images to stabilize it in a way

[late blade]

I just opted to work sparsly with chatgpt+dalle, I hit the cap as is quite often, but I'm not the average user

[lusty swallow]

lol not only is the webp image format a bad joke, but sometimes I get just a corrupted file when I download...

[late blade]

yeah, I've gotten that also, people from OAI already said they are working on png coming back

to be fair, I never had a problem with the format itself on macOS, all the problems I did have came when using Windows, but i use my windows machine mainly for gaming, so I didn't have anything to work with webp images besides browsers

[lusty swallow]

A lot of sites don't even support the webp image format

[late blade]

it might be because of how new the format is, it would be the right moment to add AI detection mechanisms into the format, that would be my hypothesis

it will get promoted considering who is behind the format

[tepid ibex]

Hi folks, new to thse discord, been working in GPT4 dev for a bit. I am trying to create avatars for my game that are consistent in style. i want to create about 25 characters, but i am struggling to get Dall-E to generate them in the same style/format/dimensions etc.

Has anyone solved this? any suggestions?

[late blade]

you can ask people that know more about gen_id's how they work, I'm not that well versed with that

and also heya

there's also the #1163443000060420206 and #1155772063596953642 that are good resources

[tepid ibex]

I am trying to essentially replicate and expand this, although im ok with a slight deviation from this style as long as they look consistent:

[late blade]

you want to create each character in separate?

[tepid ibex]

thats the hope, but if im forced into a singular character sheet that woudl be ok

[late blade]

I would create each character individually in 1024x1024 format

[tepid ibex]

that would be my goal

[late blade]

specify the output is 1024x1024, you will get square always

can you provide me with a prompt example of yours?

I can review and see what could be done

[tepid ibex]

Sure can, i abandoned this effort a while back but need to solve it now as im blocked. but ill get us close to where i left off. one moment

[late blade]

oh you came to the right place, we can review it and help each other, eventually other creators will join too

I just did the following test prompt:

provide me with a square format image of a character, 3D Animation style, cute of an avatar kid. Do this 4 times and this is what I got

[tepid ibex]

ok great, so say the first one of those, can we create a matching girl?

I tried this:
I need you to create a series of character avatar images for use in an educational video game.
The image should be 1024x1024
The image should contain a bust of the desired character
There should be no additional details beyond the character bust, no background, no frame, no border. only the bust.
Make the background of the image transparent.
Make the image suitable to children ages 8-12.

Please make an image of Mia, a young girl with vibrant, colorful outfits, always styled purple hair, big deep brown eyes with the charm of a superstar.

then again with gen_id:
I need you to create a series of character avatar images for use in an educational video game.
The image should be 1024x1024
The image should contain a bust of the desired character
There should be no additional details beyond the character bust, no background, no frame, no border. only the bust.
Make the background of the image transparent.
Make the image suitable to children ages 8-12.

Use the generation id TBHf4rdBlkP3JsLS

Please make an image of Oliver, a young boy with Light, tanned skin, messy blonde flowing hair, usually sports a worn-in t-shirt.

[late blade]

I will check what I can do, since I did some lazy prompting, it can be done

[tepid ibex]

those results above arent half bad. I'm going to try and cook up a GPT that delivers consistent results, if anyone has any tips please lmk. ill report back in a few.

[late blade]

I'm eating dinner, after I will come and help you out

[tepid ibex]

your awesome

[late blade]

I noticed from your text, you are adding the intent of the image and not what dall-e should show you. So simplify the prompt. Soemthing like this:

A square 1024x1024 portrait image of a 3D cartoon animated kid between 10-12. transparent background. no surrounding details.

This could work for you as a a template:

A square 1024x1024 portrait image of <insert you style here>. transparent background. no surrounding details.

here's another gen

[tepid ibex]

the transparent isnt really transparent, is it better to do a solid color and crop it out in photoshop?

[late blade]

yeah, dall-e struggles sometimes with some details, so you will eventually have to work with some editing software to adjust it to your needs

[tepid ibex]

thats fine, expected

[late blade]

but see how I'm getting consistent style?

even tho it's vague

[tepid ibex]

yeah

[late blade]

so, just do prompts like that, what dall-e should show you. dall-e can't do more, and sometimes we have a vision but can't put it in text. that's prob the harder part of prompting

[tepid ibex]

ok, and is baking this into a custom gpt make sense or will that just complicate it and i should focus on just a solid prompt

[late blade]

if you want to do something like a custom gpt for avatars, yes, it's worth it

[tepid ibex]

sounds good

[late blade]

you can try with different styles or different attires

let's say you want a cusotm gpt with this kind of avatars, but you specify a certain style for national holiday or xmas

[tepid ibex]

yes thats kind of what im hoping for right, like avatars kids can choose, and then costumes as well. for now i would settle for just consistent avatars.

[late blade]

that's just a 30 second idea, the rest I leave up to you

[tepid ibex]

haha right on, thank you so much for your help, for sure im on the right track now

[late blade]

just share images with us from time to time, I'll consider it as a thank you from you

just a small example, I changed the art to the one specified in #daily-theme and this is what i got

[deft siren]

alright thanks

[vapid elk]

it can't make transparent background

[tepid ibex]

quick observation:
I was able to get some nice success with this prompt:

Create a character avatar image for use in an educational video game.
The image should be 1024x1024.
The image should contain a bust of the desired character.
There should be no additional details beyond the character bust, no background, no frame, no border. only the bust.
Make the image suitable to children ages 8-12.
Use the generation id "TBHf4rdBlkP3JsLS"

Subject:
Age: 4
Gender: female
Hair Color: purple
Hair Style: wavy
Skin Color: light
Expression: charming
Clothing: vibrant jacket
Accessories: earrings, pendant

if submitted with the same parameters, you get an identical image, because of the gen_id.
if you tweak the age parameter, you get consistent, and impressive results.

If, however, i try this prompt in another window, its a completely different style. seems like the gen_id is associated with the chat session or something?

[late blade]

that is nice!

gen_id's are only on the same chat

and I think on the same session only aswell, but I could be wrong

[tepid ibex]

is there any way of preserving it? or recreating it? my hope would be to make a consistent set of characters now, but i will need the ability to expand later, with consistency. any ideas there?

[late blade]

for characters or art styles I like, I keep JSON files. at first I was asking dall-t to make them for me, but now I have my own designed tempalate that I can use with my NLP

[tepid ibex]

ok lost me a bit there, how do JSON files help you preserve or recreate a specific style? whats in the json?

[late blade]

you can upload them later to chat with all the specifics you stored in them

@tepid ibexa while back I proposed this template #images-discussions message to store character data, I got a revised version of it on my laptop. and it works pretty good for the characters I've created in my concepts

[tepid ibex]

ok, solid, i am actually using a json that is fairly similar, yours a bit more detailed. but how, still, would you go about createing a character tomorrow, that is similar in style to the ones you created last week? if the gen_id is bound to the session then can it even be done?

[late blade]

the thing is you can also do it for aspects you nail everytime with your prompts

we had a while ago a trend on doing art with charcoal

I had some combinations with anime/charcoal I loved and still work in the style

let me get my laptop and I will share the file with you

[tepid ibex]

so the theory then, since we lose the gen_id, is to be descript enough in the illustration style to get them looking consistent?

[late blade]

yes

since we can't control gen_id's consistenly

[wanton moon]

Copilot is playing me for a fool

[late blade]

I will miss some aspects from copilot, my sub ends tomorrow

I won't be renewing it

[wanton moon]

An image with a text i wanted to generate did not add the text, yet copilot keeps saying it was

[late blade]

I've had that a lot of times

[wanton moon]

And then suddenly "Time to move on to a new topic" lol;

[late blade]

something like this?

had that #images-discussions message earlier

[wanton moon]

yes exactly that

[late blade]

Copilot Pro needs a lot of work and polishing

It will be really good eventually, but not for now

[wanton moon]

Time for co-pilot as my counsilor will have to wait I guess

It may tell me: I never advise you to commit that horrible act lol

[late blade]

DALL-E and GPT in the copilot implementation is really decoupled from each other. So their GPT Chat doesn't know what DALL-E does at all.

It will tell you it created an image, but the image could be there, could be blocked or not at all. and GPT will insist it is there. That's the Designer Copilot GPT

I'm happier with OAI's implementation because of that. It suits more to my workflow style.

[wanton moon]

Yeah no chance for me since OAI only accept credit card and no other forms of payments (seen and read lots of discussions) so I am stuck with 3.5 and no image generator

[late blade]

you can change your payment model to App store from apple if you have iphone. I did that since it's convinient for me

[wanton moon]

I have no Iphone unfortunately

[late blade]

how does the payment model work on android? maybe that's another way?

[wanton moon]

I rarely use app store and payments on the appstore maybe 1 payment in the last 5 years

[late blade]

aha ok

I just use the apple model because I track lots of my expenses with that

[modern nest]

is it normal that my dall e 3 cant do photorealism?
it feels like its putting a thick filter on everything to avoid looking real

also why does my dall e 3 sometimes give me two images instead of one, but every single time that happens the two images look like trash compared to when it gives me a single one?

[late blade]

It depends on what you are passing to dall-e/chatgpt with your prompt

it helps if you use hyper-realistic, photorealistic

the dall-e custom GPT form OAI does 2 images usually, GPT one unless I ask for 2, somtimes there are flukes and it does 2 without asking

[deft siren]

how can i create images which looks real

like this

i tried and it still looks animated

even if i let it describe the art style

[late blade]

I think you have to pass that the image is natural instead of vivid

[wanton moon]

Reminds me something of this

[late blade]

You use printers at work?

[late blade]

check this out, maybe that can give you some insight https://cookbook.openai.com/articles/what_is_new_with_dalle_3

[wanton moon]

yeah, but that image is a reference of a movie, where they wack the printer out of pure frustration

search Office Space - Printer scene on YT

[late blade]

I wouldn't have been able to show that kind of image if I were at my first job, I was in QA for color quality in different media

[modern nest]

is there a way to do that in chatgpt though? i dont think so

[late blade]

dunno

[wanton moon]

This one is also pretty hilarious

[late blade]

@modern nestthis is what gpt told me:

As of my last update in April 2023, the text-based interface provided here for interacting with DALL-E does not directly expose options to choose between "Vivid" and "Normal" rendering modes, which you might find in the DALL-E API or other interfaces.

[deft musk]

Okay, they can keep their printer

[wanton moon]

They are not afraid to bring the entire HR staff to the battle field if they have to

Reminds me of the song: Print Printenar - Printing the battlefield
"We print young"
"Cartridge to cartridge"
"We print"
"No paperjams no demands"
"Paper keeps flowing"
"Printing the battlefield"

[deft musk]

Apparently they cowed the HR team, who think this display is just fine to have everyone known to have seen, too

[verbal sequoia]

Showcase your DALL·E creations to the world!

Collaborate with our OpenAI Instagram page! Just invite @openai 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!

[deft musk]

I'm most scared of the dude on the right, who apparently has a bat in either hand, both over the same shoulder. He either is really skilled, or really unskilled, and either way gotta watch out

[wanton moon]

I think it is his sheath for the bat

[deft musk]

...oh! Yeah, that works. When your bat goes to bat for you, then is done for now.

[wanton moon]

"This is bat country!"

[dim cradle]

we can't stop here...

[wanton moon]

I do a little bit lightheaded, maybe you should drive

[wanton moon]

"Have any last words?"
Starts to type slowly all by itself

"Well what did it type?"
"See you in h*ll!"

[wanton moon]

A whole galaxy of multi-colored uppers downers screamers laughers

I think ill know what to watch this weekend

Yeah I'll start at 8 a.m. with a big hamburger (the cornerstone of any nutritious breakfast) and turn this movie on

[onyx ridge]

Our arrival was poorly timed...

[wanton moon]

I just want to carve a little Z on your forehead man

[onyx ridge]

What're we doing out here in the middle of the gggdddd desert?!? Help! Help! We need help!!!

I remembered the girl. My attorney had made a fool of himself with her in the elevator earlier that night.

[wanton moon]

why do I hear his voice and speechpatterns when I read those lines

[onyx ridge]

We were right in the middle of a giant fffffff reptile zoo, and someone was giving booze to these gggdddd animals.

Any moment now, they'd tear to shreds...

[wanton moon]

My blood is too thick for Nevada

[dim cradle]

"My attorney had never been able to accept the notation that my hallucinations were symbols of the real world."

[late blade]

Interesting how discord works on windows. On macOS I could prepare the next post in advance for a while and when I returend to post it it was there. On windows I prepare the post but when I return after 30minutes the image is corrupted.

[dim cradle]

That’s not cool

[late blade]

Who knows what the difference is tbh. Thought both client versions were done with with Electron

I'm not a fan of cross-platform dev

[onyx ridge]

That's a hardware issue

[late blade]

oh, please elaborate

[onyx ridge]

They don't corrupt for all systems

So a bit flipped in memory

[late blade]

strange, I only see this kind of behaviour with discord

ahaaa