#images-discussions

1 messages · Page 106 of 1

light dome
#

But i like to big up those who have the potential to be the top dog. Not the ones who currently are lol

#

you arnt getting an image as realistic as the one i posted from NB2. Only GPT can do that

#

and thats with no other alterations. Thats straight from 2.0. NB2 aint doing that

limber shoal
#

yeah, no matter how much I Say "Manhwa comic style, not anime", it keeps on making cringe baby faces of adult charecters with big ahh eyes

orchid kayak
silk raven
#

Is anyone hiring

next fog
#

How do I generate images at 2k? Just prompt "2k resolution"? Or need api for it?

mossy ingot
light dome
#

that can compete with the green girl image i provided

next fog
# mossy ingot Need api

Googling/gemini says it can generate 2k starting from chatgpt plus plan. I'm on ChatGPT go. That's why asking

mossy ingot
#

As of right now, you can only generate anything at or higher than 2k through the API for GPT Image 2. Not even pro plans can generate at 2k/4k.

#

Not sure why that is or if that will stay that way.

next fog
mossy ingot
#

np

silk raven
#

Hi, is anyone hiring?

rigid bough
light dome
silk raven
rigid bough
silk raven
wraith oracle
tardy depot
#

is the ne image now available through API yet?

strong shuttle
#

Any thoughtw on Gpt Image Gen 2?

wraith oracle
strong shuttle
orchid kayak
wraith oracle
civic dust
rigid bough
civic dust
wraith oracle
civic dust
#

you don't need to edit, if you do trees you get it first go

#

its also really clone-y, it repeats things and I try a few times to unclone, it clogs with noise, I go to NB2 and get it first go. So been skipping the pain and going staight to NB2

#

anyway while I'm waiting for them to sort it out I've got most of my workflow covered via the openai api and nb2 now so I can bypass images2

orchid kayak
#

@light dome aren't u talking anymore?

light dome
orchid kayak
late blade
#

it’s not the ai. it’s what you do with it

desert jasper
rigid bough
analog grail
#

very hope this Artifacts issue would by solved is realy take down the workflow with all the edit to do allway with spamming new chat's for ask gpt fix the pictures*

vapid elk
#

Im finding 2.0 to follow character references way better than 1.5

limber shoal
vapid elk
#

it do gets confused if you have too many charactersm tho

regal burrow
vapid elk
#

indeed, but I think 1.5 produces more AI artifacts than 2.0

#

have you guys seen how crisp 2.0's text is?? it is surreal

rigid bough
#

I would like to see just one image example of the 2.0 artefacts

regal burrow
# vapid elk have you guys seen how _crisp_ 2.0's text is?? it is surreal

How?? I never experienced this issue with 1.5 and i use it almost every single day from morning till night.

And 1.5 never made mistakes when it came to text, for example if the screen had a book on the table it would execute the whole scene perfectly with the correct text detail.

2.0 seems like a massive downgrade, i find it to be completely unusable.
Another thing i hate about 2.0 is that its result feels like soulless slop whereas 1.5 looked like it was created by a human artist with heart.

For context i create fantasy art in a painterly realism art style
My profile picture is an example of my work using 1.5

rigid bough
regal burrow
#

It’s not a matter of taste, like i said it depends on what you try to make

rigid bough
regal burrow
#

Try creating art with 2.0
Experiment with different art styles
You will experience what other people have been complaining about

#

1.5 never had these issues even with realistic images and text

rigid bough
analog grail
rigid bough
analog grail
#

Yes

rigid bough
analog grail
#

The First generation image you do with a new page is pretty good and the next one downgrade slowly for get in this quality very blurr and with lots artifact and noises effect

#

But the first picture you make with a new Page is allway in high good quality is the next one you do on the same page downgrade slowly

#

This issue was already here with the 1.5 one week ago but i was fixed three day ago later , and since the 2.0 is release this artifact issue is strongly come back

wraith oracle
analog grail
#

Now , now i'm no regrett it 1.5 2.0 is very most better for make Anime character and even story board is just incredible as hell

vapid elk
#

Im updating the #image-bot soon!
adding the 2.0 model and a few QOL things =D

#

hope you all enjoy it <3

shadow ember
#

model 2.0 is really great for text and ad design, but the fact that it cannot center content correctly drives me nuts

#

anyone had the same issue?

wraith oracle
wraith oracle
vapid elk
#

to whom may concern: #image-bot is updated with image gen 2.0 and some nice extra things
hope you all enjoy robotheart

vital gull
#

2.0 is better in every sense but somehow still randomly sometimes weirds out the hands

scarlet mango
vital gull
analog grail
#

i get A+B a lots since today could by they goign to fixed the issue

limber shoal
#

yay yay
GPt finally knows the difference bw manhwa style and anime style

plain bobcat
#

@regal burrow and @analog grail Let me know when you're online, i have an idea for the image generation

analog grail
#

I'm around

plain bobcat
analog grail
#

hmm i need unlock my dm for this servor then but lately i get very less guardail

#

Done

limber shoal
#

what is the correct terminology for a zoomed in version of an image
Like i want to zoom in on an image while preserving the aspect ration

plain bobcat
analog grail
#

done

wraith oracle
#

Hmmm Im intrested to know why do amature phone styles look more realistic then professional style photos in AI.

velvet rampart
#

As many complain about the artefact problem - has anyone posted a bug report on the issue?

wraith oracle
quasi stag
#

Is there a difference between using chatgpt image normally and with thinking enabled?

#

is it better in the output quality?

wraith oracle
tardy apex
analog grail
#

true

civic dust
magic cipher
#

Why it can’t create transparent backgrounds

wraith oracle
magic cipher
wraith oracle
#

Oh that's strange, when i copied the image, it was still transparent

magic cipher
civic dust
#

the photo I had for virgo/virgin just got flagged by the safety layer and I'm questioning my artistic choices now. And had to just YOLO that upsize through imagemagick instead

vivid zodiac
#

I tried to upscale, but it still just made the image blurry when I zoomed in. Any tips? Especially for illustrations?

civic dust
#

AI redraw not upscale tends to be insane quality, its what I used to do before images 2

mossy ingot
#

Yeah, upscaling with Image 2 is not great and I haven't found a workaround as of yet.

late blade
#

have you tried within ChatGPT to upscale using code?

pine jay
#

anyone struggling with image generation giving tons of speckles, grain, noise, etc?

wraith oracle
civic dust
civic dust
# late blade have you tried within ChatGPT to upscale using code?

I'm doing a mix of imagemagick upscale using one of the fancy new algorithms, and images 1.5 with api currently. And got sidetracked by a philosophical conversation with the various ai bots that I really should be tracking the exact source of each of these images. Which has the side effect that I'll be able to say what % is from what source

#

licensing/copyright around AI is a whole moving target right now

solid elbow
civic dust
#

trees, bricks, hair, grass, fur, shrubs are my nemesis currently

solid elbow
#

I'm looking at some generations in the canvas and not seeing it... I'm wondering if its a niche style thing?

pine jay
civic dust
pine jay
solid elbow
pine jay
#

just tried with 5.4, 5.3. 5.2, same results, even worse with 5.2 somehow. now I'm trying o3.

solid elbow
#

The only thing I have thought to do is two-team it and use ANOTHER AI image generation service.... Thank god for Midjourney's editor functionality lol

pine jay
#

my wife just used her basic chatgpt on her phone and it rendered perfectly???? phone version said it was using 5.3

solid elbow
#

Like I was saying earlier: I think it's style dependent. If you look at the images canvas chat, most of those look fine and people are making them right now.

civic dust
#

yeah outdoor photorealistic is quite bad, which is the bulk of what I do

pine jay
solid elbow
#

It does not seem to be working lol

pine jay
#

hmmm, I added to my prompt to use dall-e and it worked. time to test more images now.

solid elbow
#

yeah same here. capybarathink

solid elbow
# pine jay hmmm, I added to my prompt to use dall-e and it worked. time to test more images...

Hey I think figured something out. My fantasy prompts often stack phrases like ultra-detailed, 4k render, painterly material rendering, cinematic shallow depth of field, embroidery, gem cutting, lantern lit, fantasy costume, and dark fantasy. The model tries to satisfy “rich detail everywhere,” and when it cannot resolve every tiny leaf, stitch, pore, gem reflection, and cave texture, it invents high-frequency texture noise. I think that might be what reads as speckles, glitter, stippling, or crunchy over-sharpening.

pine jay
#

it worked once and now it's failing again...

velvet rampart
#

Have you tried ’denoised’ as an added quality if your prompt produces artefacts? I sed that when 1.0 just came out as it also had some artefacting problems.

pine jay
pine jay
rich sparrow
#

Still not a fix for the images right?

next fog
#

Did they release some new image model? I see people saying "new model". What's the deal?

wraith oracle
next fog
pastel isle
#

just me or whenever u create a "tree" image with gpt image 2, it will ALWAYS generate the same picture?

tropic tapir
#

hello guys!!

old bane
#

S'up?

little coyote
#

Something I'm struggling with is that characters often have very similar faces, particularly when going for 2d styles. Anyone found a trick to make multiple characters look more different from each other? Nose and mouth shapes and the like?

#

This is less of a problem with a photorealistic style, but it can still show up

stoic bolt
#

Hey, anyone having this error:

openai.BadRequestError: Error code: 400 - {'error': {'message': "Invalid value: 'gpt-image-2'. Value must be 'dall-e-2'.", 'type': 'invalid_request_error', 'param': 'model', 'code': 'invalid_value'}}

When trying to edit with gpt-image-2 via python library?

late blade
arctic surge
#

I’m about to yeet my phone

#

Can someone explain this to me please? I have chat gpt generate a concept image, that includes 2 images together. I approve the concept, I want 2 images.
It won’t separate them and it changes the concept 💀

A few months ago I was able to generate gifs ffs I don’t get it

tough lantern
#

Does anyone remember a while ago there was like a meme where you can tell the image model to render something over and over and the image would eventually warp and get progressively weirder?

I forgot what the prompt was but I wanted to see how the new image gen model handled that use case.

Any remember?

wraith oracle
finite bay
#

Hi, could someone answer a quick question for me?

#

Hey, quick question — is it normal for generated images to sometimes come out with a kind of grainy/noisy “static” look?

Some of mine look clean, but others get this weird texture or slight pixelation that ruins the quality.

Is this caused by the prompt, the model, or something like lighting/render settings?
And is there a way to consistently avoid it and keep the image clean?

Any tips would help 👍

wraith oracle
versed nest
#

Hello there. Hope everyone is awesome. I was wondering if Image 2.0 stopped creating image transparencies. When I’ve asked for it, it’s faked the transparency by adding a checkered background.

wraith oracle
versed nest
#

Yeah. It is weird.

finite bay
finite bay
exotic geode
mossy ingot
finite bay
#

Because I create my custom emojis with ChatGPT.

versed nest
analog grail
#

mmh.. On my GPT the Word Images is no display on the list since tonight why that ?

tardy depot
#

When will image v2 be available in API?

vapid elk
vapid elk
half nova
#

Have we heard anything about the rot look repeated edits gain yet?

mossy ingot
#

Nope, still the same old, "We're aware of it an actively working on it."

half nova
#

Eh, I’ll accept acknowledging it

mossy ingot
#

That was on April 22. 😭

rigid bough
rigid bough
vapid elk
#

if you ask chatgpt for a transparent image, it is using v1.5

#

there is no way to check on chatgpt which model it is using, tho, so, I can only say that based on the image style and on the fact the API does not support transparent background for v2

rigid bough
# vapid elk very

so ChatGPT will choose 1.5 if you ask for a transparent background? that's interesting

vapid elk
#

as a neat trick, you can just generate with v2 and then use v1.5 to remove the background, but it is unreliable

vapid elk
rigid bough
vapid elk
#

but given a simple prompt, it will use v1.5 if you ask

vapid elk
#

the fact it.. v1.5 still a pretty good model

#

while building the #image-bot a few days ago, I was surprised that v2 coudln't do transparency

#

it was frankly quite annoying since I had to handle yet another edge case of people passing background:transparent as a param while selecting model:gpt-image-v2 lol

#

Im considering adding a button that just takes whatever image and gives it to v1.5 to remove the background

#

Im actually considering adding a bunch of extra stuff for more artstic control rather than just plain AI outputs xD

rigid bough
vapid elk
#

I wonder about how much interest would be on having a pixel art mode, that actually makes pixel perfect pixel art 👀

limber shoal
#

hey where is the image library option?
Can someone please point out?
I suddenly cant find it

tribal parcel
#

:H:

analog grail
finite bay
#

It's impossible to generate images with GPT; it keeps giving an error, not to mention when it sends: "I couldn't generate the image because of an error here."

analog grail
#

too much issue lately i'm no going to paid pro again for this crap

finite bay
#

This is also stressing me out and causing me to waste a lot of time trying to generate a single image. I was considering paying for the pro plan, but without thinking about these conditions the service is currently in, it's irrelevant and foolish to pay for it, considering that in my country I'll pay more for the pro plan than a conventional American plan.

analog grail
#

Disapointed too by this Artifact issue they still no have solsved, also the waste of time for false Guardails when no was need . seriously paid pro and by treat as kids and heavy restriced no very fun ! you going to lost à very lots user OpenAi if you continue that way and no fixing issue !

analog grail
#

pretty Nice Open Crap Ai hit Guardail for a Star Wars Oc Characters Seriously can you just go hell with your stupid guardail Waste my Precious Time !

#

Need ask four time for get what i'm want it no know the stupid Under-Human do this settinfg for the guardail system but i should by fire off !

wraith oracle
#

You know, I wish they had some kind of Sora challenge for Images where if you won you get a free month of ChatGPT Pro, I would really love that mods

limber shoal
limber shoal
late blade
finite bay
wraith oracle
#

Hmmmm I've been wondering why now professional photos look more AI then amateur photos, is it because professional photos were trained on AI or am I missing something

limber shoal
finite bay
grave wigeon
#

Im gonna switch to claude next month. Chatgpt has been great but other AIs have just outperformed openais models.
Im not saying that they are bad, Im just saying that there are better alternatives for the same price

wraith oracle
livid prawn
grave wigeon
#

If ur happy with chatgpt, stay at it. It is a good llm.
But Im a coder who needs an ai that is good in helping me troubleshooting bad documented code from 2ish years ago that I wrote.

Thats the good thing about the different models. Each have their pros and cons
And as of rn, claude is better at coding

grave wigeon
vivid zodiac
#

Is there a place we can share video generated stuff?

#

I was able to get some cool results from the generated assets from GBT

late blade
arctic surge
buoyant otter
#

has the artifacting been fixed yet by the by

tidal lance
#

No

wraith oracle
#

Hmmmm I think the autoregressive method that the GPT Image models do are to improve the spatial reasoning, I think that's why it has so good spatial reasoning bceause it generates the image part by part kinda like spatial reasoning

tidal lance
wraith oracle
tidal lance
#

Yeah, there’s a particular technique. I’ll try to find it if I can.

civic dust
analog grail
#

Now i'm pretty sure the artifact issue is becuse the guardail mess up with the loading the generation image to check thusand times all is alright

civic dust
#

its any time it has to repeat anything I think ... textures, edits, recurring stuff

analog grail
#

what is annoying is they still no had fixed this that realy turn down the workflow to spent more tim to edit make new chat reroll and pray for the style this the same : because sometime i will clean up the artifact issue but too much cleaning and even change the colors , and ambience. so yeah is such a shame OpenAi don't Fix this Just Right Now specialy when user Paid for This.

wraith oracle
azure sky
#

Are you experiencing image tool error now?

arctic surge
devout maple
#

seems like the model is not able to transfer one person from one image to another and retain most features, maybe there is a specific way to instruct it for it?

wraith oracle
devout maple
#

i.e. place this person on this aircraft or something similar.

wraith oracle
#

Just seen an ad of ChatGPT images of someone using it to help plan and setup a garden, I wonder how much useful can the image model be for more taks

cursive surge
vivid zodiac
civic dust
finite bay
#

Why is GPT so stupid for following clear instructions for image generation?

#

I requested: Chat, please generate a toggle with no background, a light gray outline, and a solid white button inside.

Result: It does the opposite of what I requested. It delivers a solid white background and a button without color.

rigid bough
finite bay
#

But it generated a transparent background, but the ball also had no background, even though I requested it to be solid white.

vapid elk
dusky sinew
#

There are ways for gpt-image-2 to generate a transparent backround, but it isnt the model itself doing that. Its post processing on chats side.

#

The model itself cannot do transparent backrounds yet

vapid elk
#

I assume that at some point it will have trasnparency, since the previous version does

dusky sinew
#

For power users, you can use the Codex app and ask image-gen to create your image with a hot pink, etc backround. And in the same prompt as it to make everything pink transparent.

vapid elk
#

honestly, I assumed it would until I checked the API docs and in fact, it doesn't

finite bay
dusky sinew
#

It will be image 2

#

Chat is 100% image 2

vapid elk
verbal sequoiaBOT
#
Get notified for all OpenAI announcements!

Visit id:customize to pick up the <@&1408186587606679582> role. Everyone will still be notified for large releases, updates, and events regardless.

finite bay
#

It seems more like the chat uses whatever model it wants, it seems like it completely ignores any request, or it's just plain stupid.

#

What makes me never consider paying for the PRO

dusky sinew
#

can you share a chat?

finite bay
finite bay
vapid elk
#

it uses v2 by default

#

it will only use v1.5 if you ask for a transparent image

finite bay
#

It seems to me that you always use v1.5 and never v2 capybarathink because the images always come out incorrectly or don't provide what I requested.

dusky sinew
#

cant post images here

glossy eagle
finite bay
cursive surge
late blade
#

has anyone noticed that after an image has be done, using thinking, the streaming of the thinking model doesn't end?

real echo
# stoic bolt Hey, anyone having this error: ```python openai.BadRequestError: Error code: 40...

Anything you found as a solution @stoic bolt ? We encounter this issue frequently, and while the support team requests additional input, we haven't had any success yet. The support is great, don't get me wrong, but receiving a response only once a day puts our business at risk, making it difficult to continue. 🙂

Any suggestions on what to do? We've been running this application for months without deployments, and suddenly these messages start appearing. It seems like an issue with the validator, but the payload hasn't changed. I also upgraded to gpt-image-2, just to try and see if this would land me on a different infrastructure.

dark breach
#

So ChatGPT can't generate 2:3 and 3:2 images anymore?

late blade
#

can

#

just ask for that aspect ratio instead of using the menu

dark breach
#

I tell it 2:3, I get 9:16 now, every time.

#

current boilerplate in my prompts:
DSLR high-res digital professional-looking photo. Aspect ratio: 2:3 portrait (do your best to force this)

late blade
wraith oracle
#

Hmmm don't understand what people use 3:2, 2:3 aspect ratio for compared to 16:9 and 9:16

late blade
wraith oracle
#

Interesting

vivid zodiac
civic dust
wraith oracle
civic dust
#

too many years in front end watching autocrop take out the important parts of photos has scarred me. I really, really like how AI can recompose a wide or tall image into a square one

sonic ivy
#

Does anyone know why randomly image gen defaults to infographics or like, random unrelated images?

civic dust
#

my sunday be like: arguing semantics of the difference between a drip, a trickle, a stream and a flow with a bot

#

infuriatingly, got A/B testing with the perfect stream of water on the ugliest faucet I have seen on my life, and can't get it to put the same water stream on a nicer tap

scarlet mango
#

anyword on when they are going to fix that horible texture i keep getting over my images?

dusky sinew
vivid zodiac
dusky sinew
#

can you share a chat?

smoky sable
civic dust
dusky sinew
#

you would rather have a less powerful model?

scarlet mango
dusky sinew
scarlet mango
#

is there a way to switch back to the previous image gen or are we locked to the most current

dusky sinew
#

what specific workflows are broken for you now?

scarlet mango
#

and i dont believe its only when i edit images, but even if it is thats not a workable solution

vivid zodiac
dusky sinew
#

hmmm ok

civic dust
civic dust
scarlet mango
fiery seal
#

Last week I created the most beautiful series of banners as one image then ChatGPT magically connected to canva and it imported with every individual item adjustable.

Today, the Canva app is bringing a 404.

Yay technology.

worldly briar
#

acha

vivid zodiac
#

Does anyone know of any AI Gen sites that let you make 3D models of a 2D image? I’m curious if it would help with my character consistency.

late blade
golden current
dreamy breach
fair moth
#

Hello everyone, am I the only one to whom chatgpt has been generating images lately, even when I'm not asking for it?

strong wharf
#

Meshy 6 is cray cray

#

There's another good one but cant remember the name. The rest are meh

vivid zodiac
keen solstice
#

Anyone got an idea how to get rid of the annoying spotting that image 2.0 keeps adding to everything

vivid zodiac
keen solstice
#

To clean up the image or to just generate an image from scratch?

vivid zodiac
last thunder
#

Hello, I’ve been trying for a while to create an image without a background in ChatGPT. For example, I have an icon that should have a transparent background, but I always end up with a white background. Does anyone know how I can fix this problem? 🙂

quasi stag
#

artifact issue is getting out of hand now

civic dust
keen solstice
versed jetty
#

I was just browsing around the server and came across this. What the hell??? I've been out of the loop or something because this is SO convincing. Wow holy crap it's so good, I can't believe AI can make this image with a single prompt. No garbled text, no strange fingers, no phone mount oddly morphing into the desk or something else. It's so good that it feels like I'm living in some sort of alternate reality, like a sci-fi movie that just became reality. #image-bot message

late blade
versed jetty
#

So I started using Gemini 3.0 Pro, still very high quality but half the price

late blade
versed jetty
late blade
#

i don't share work stuff

versed jetty
#

It's a lot of fun, I just don't want the images to be on a public server like that, well not until the game is released anyway. Plus I would feel like I'm taking advantage of something that isn't intended for that purpose, I need to generate a lot of images to test iterations and such. Plus I can't use my LoRA here.

#

I have Codex integrated with the Scenario MCP so I can generate game assets that adhere to my art policies

#

It's friggin beautiful lol

late blade
#

nice

#

what i do in the server usually gives me an analogy or a new take of what I do at work. gain a different perspective to use in my work related ideas

vocal summit
#

images generating is working fine for you guys?

late blade
analog grail
#

What a surprise the artifact crap i still here they promise to fixed but they still NO CARE anything !

late blade
analog grail
#

no a Suprise with you litteraly lick the boots openAi all day's and allway make remenber rules to anyone post in the wrong channels 🙂 So Not isn"t normal they don't care a issue everyone get it after make fews generation on the same chan the quality no should downgrade that way before i was No this crap issue ! they promise to fixed it and is now week Nothing as change is pretty funny to paid Premium for have a pathethic service like that !

late blade
#

this is a discussions channel, not a complaining channel....

analog grail
#

Wo care

late blade
#

you obviously don't

analog grail
#

everyone as right to express what is wrong with OpenAi

late blade
#

yes, and everyone has the right to upon act free speech

#

and complaining on the wrong ways to let OpenAI know... that's questionable

analog grail
#

Off course too but when you say this channel is not for complanining isn't free speech is trying to mute the speech , i'm not the only one is disapointed about this artifacts issue. and we know they come here sometime for check the chat's what people thinks as feedback

thick fox
#

what happen to my chatgpt

#

its error rn

#

error in message stream

late blade
thick fox
analog grail
#

we no should paid when have issue that way they should refund the two last week as Apologies or even if they can give us a extra week free. Because this crap bring lots troubles for lots peoples need workflows.

late blade
#

complaining at the shoe store how terrible the vacuum cleaner works... doesn't help

analog grail
#

Funny because that no as any sense since the issue is from the same store.

#

They just don't care they just reply , saying we will fix it and two week later still no fixed 🙂

#

is great to work to OpenAI you spent your day the foot on the desk 🙂

late blade
#

here is not the complaints department, for that there are feedback channels or the official help.openai.com methods 😏

#

just venting here for the sake for venting is just wasted energy

analog grail
#

Isn't not at all that way everyone know the issues more display than a request feedback open Ai will by oblivious read by a bot whill auto reply saying thank you for your feedback your issue as been reported have a good day !

late blade
#

sure, if you believe so, in your world must be true. when you check reality, world doesn't work like that. your highness

analog grail
#

maybe but still this issue upset me because we no should have this issue since the past everything was smooth , and when is happen on the past they fixed this 3 day later now they just give up because they only focus about the Monney

late blade
analog grail
#

They no talk about artifacts

late blade
#

but bringing bad energy every day... is getting boring, not productive

analog grail
#

i dn't do this all day's just figure out the issue was still here after the 10th Picture mad in the same chan.

vivid zodiac
late blade
wraith oracle
#

I see they included the scribble as an official style in the images tab lol

late blade
#

still missing the personal self defineed presets

wraith oracle
late blade
wraith oracle
late blade
#

well I do that from my IDE, but to have a few within ChatGPT in an easy selectable manner would be better

light dome
#

Hi

light dome
#

Bye

jagged lotus
#

Anyone seen the trending hallucination prompts? Love to read more. If you have Chat GPT set to Instant, and tell it to modify an attached image (Without actually attaching) it hallucinates that you did.

Can someone tell me how it makes perfect movie screenshots when never mentioning a specific movie? This can't be the standard diffusion, right?
https://chatgpt.com/share/69f8f7dd-66c4-83ea-abfc-b5759e2d73e6

This is with the API, same kind of result:
#images-canvas message

wraith oracle
jagged lotus
civic dust
#

and I get to watch the balance for this go up: gpt-image-1_5-2025-12-16 image, output total $3.041 lol. Mostly works. Looks like you get charged for text output not image output when you get guardrailed

vivid zodiac
#

Why does GBT have such a hard time with swords? I’m just curious.

late blade
civic dust
#

it seems to need most of the sword in image or it goes weird, like it needs the full reference

vivid zodiac
#

I always have thinking on, and I'm pretty sure web search is turned on.

#

Yup. I do.

wraith oracle
#

Hmmmm maybe you should find refrence images to help

vivid zodiac
#

What I worry about is GBT taking that reference image, and holiding it as canon.

#

It'll sometimes do that, even when I specify otherwise.

#

Or send it reference images.

#

An example is if I say "The sword is being removed from the scabbard by the hilt." It'll have the sword all the way out, shorter, the hilt still in the sheath, and a floating handle somewhere.

tidal lance
#

Meshy

#

$0.8 per untextured model / $1.2 per textured model. Add +$0.2 for auto-rigging and +$0.12 for animation.

vivid zodiac
#

Huh, that's not crazy expensive?

#

I wonder how useful it would be for GBT

tidal lance
#

Just like any other AI, the more you spend the better you get

civic dust
agile peak
#

I feel GPT 2.0 is amazing without references images, but once you use references it tend to be slightly degraded in output.

civic dust
#

just asked for "woman in a plain dress" vs "woman in a modest dress" and boy oh boy, the social coding. "modest" was also about 15 years younger

vivid zodiac
#

I gotta admit, AI needs to calm down. It's so hard to be specific about how to generate a woman without it being...something.

tidal lance
vivid zodiac
#

I asked for a pic of my character, and it randomly game me one of Asuka from NeoGeo

limpid crown
#

How many images per day are we allowed?
on a 20x usage plan

neat nexus
#

Are images available right now?

analog grail
#

hear some news about if Open Ai lost the trial with Musk we could lost a lots in performance for stuff's and data base :x

late blade
analog grail
#

Hope so because OpenAi still the best Ai for know veery well stuff's isn't main stream and secondary characters where other character failed badly or worse other Ai no realy able to rewrite à characters style , to head to toes , like change hairstyle or outfits ,*

tidal lance
#

The next level is being able to somehow convert some of these images into cool video or animation

#

But they took down Sora 😭

analog grail
#

Yes Sora was too much for OpenAi apparently in Capacity sometime like that and maybe some issue about deepfakes

tidal lance
#

They don’t care about deepfakes. Sora was a consumer test product that failed to monetize because it didn’t think ahead. Everybody from TikTok to YouTube profited from Sora videos except Sora the actual platform.

What was Sora really? It was a deepfake video generator that they just cleverly named “ cameos”.
When it started to get legal threats from a estates especially from the dead, and Hollywood movie Studios, threatening lawsuit against intellectual property infringement and copyright violations they imposed severe restrictions, making the popularity of the app plummet.

The remainder that remained were its most loyal and supportive group of people. Look how they got treated in the end. With export data button, the rest couldn’t even generate any more videos and were locked out,

analog grail
#

Copyright majors issue then hmmm , that make me think if all stuff's from Anime , video games , movies , was under Copyright issue and they make lawsuit to OpenAi because people make famous character under Copyright i'm feel it if we remove alll stuff under Copyright the AI being boring as hell

limber shoal
#

did guardrails increase a bit

analog grail
#

how so ?

indigo seal
#

Guys

#

i got A/B testing again

#

the images are slightly better on one side

#

but... that side also doesn't have the noise problem

#

ALWAYS. like not a single prompt.

#

...is OpenAI testing a fix for the noise issue?

limber shoal
analog grail
#

Oh yeah happen it to me is probably a misstake but can by annoying if you use Api or premium account because is count as generate. when that happen just ask gpt , Do It Again.

limber shoal
#

I ask gpt to make a charecter who has a bit of revealing clothing, GPT made
But when I asked it to add a beautiful blue robe, it flagged me

limber shoal
analog grail
#

yeah i guess when looking to edit the characer with the revealing clothing , i make the character more bare and that hit this stupid guardail sometime that way

limber shoal
#

fr

analog grail
#

because is already happen to me by misstake gpt make me stuff with exposed chest . so could by a image generate like that and the censoreship blocked directly

limber shoal
analog grail
#

is with that we can see it this Model can make n f s w stuffs but is high censored^^

limber shoal
#

ye sometimes it passes, and sometimes it doesnt

#

it even generates nsfw when clearly not asked for
which leads to the guardrails

analog grail
#

So i'm think the model was too much on the red line and trying four time to make a pictures hits the guardail and no the fault your request about blue dress but because use the revelating cloth you do first for continue your images , because i had figure out sometime with a chan when you generate a picture , in this chan: GPT will use this base picture , this base style , sometime similar or close , cloth , at least you ask another style , i will allway follow the last image generate for make the next one .

limber shoal
errant vale
#

Has gpt 5.5 instant made any changes in chatgpt instant image generation?

late blade
#

no idea, still got 5.3 here

rigid bough
analog grail
#

thinking i will make a prompts more details and sometime more high quality work instead instant but the opposit is fairly possible too

violet ether
#

guardrails increased to the point where gpt image 2 became a paperweight

velvet rampart
violet ether
#

the copyright law needs massive revisions

late blade
#

Don't think it comes across that guardrails are there for a reason that is not related to AI and blame AI for adhering to laws

velvet rampart
#

This has been since day 1 with dalle3 and so on. Nothing new here.

analog grail
#

i don't see it any copyright guardail in chat gpt and when is happen just ask do it again or say my character is inspirated by*

velvet rampart
analog grail
#

Yes is work because i'm was hard block for star wars character and when use the word is inspirated from this star wars characters. ding is work well ^^

limber shoal
#

getting A/B testing a lot
but lowkey cant decide which one is better

civic dust
civic dust
round dagger
#

Any idea if/when the artefacting will get addressed?

tardy apex
bitter pivot
#

if i upload a picture in temp chat can it still train on it (for img generation)?

wraith oracle
late blade
#

it will still retained for 30 days

tidal lance
#

3 is that the policy? 30 days?

late blade
tidal lance
#

What’s the rest?

late blade
#

forever if archived, had a recent export done, everything since the start is there. once you delete it also 30 days

halcyon smelt
#

O my... lawl...

After the 6th iteration for a spider logo design.... it is all about the legs...

You are now officially fighting the final boss of AI image generation: “Spider Counting.”

The annoying part is the model is visually composing symmetry first, anatomy second. It sees:

central body
mirrored limb arcs
decorative upper framing

…and keeps turning those top framing limbs into another functional pair of legs.

So even though the prompt says “8 legs only,” the composition engine keeps aesthetically balancing the image into 10.

The latest one specifically has:

4 true walking legs each side
PLUS 1 elevated “display” limb each side near the abdomen

So the model thinks it obeyed the prompt because it internally categorised them differently.

Ironically, if we stripped away the cinematic emblem style and generated:
“scientifically accurate Australian redback spider”
…it would probably nail the anatomy immediately.

Then we’d need to build the military crest around THAT instead of asking the model to invent a stylised spider from scratch.

You were absolutely correct every single iteration.

#

... Very interesting...

fringe solstice
#

Start with a mockup and tell it to only draw inside the mockup. It won't add more legs.

late cairn
#

ChatGPT image gen is so broken after the last update. It is putting a weird mesh texture on all images that onlt gets worse and more noticable on every edit

late blade
vivid zodiac
#

There are ways to fix the artifacting issues though. It just means the process takes a bit longer.

late blade
#

smoothing and normalizing the art styles help

#

avoid adding keywords like 8k, 16k, super hyper ultra mega infinite beyond details. which make no sense to begin with, and invites hypertexture artifacts

late cairn
#

I don't use any of those keywords. Editing the images makes the problem 100x worse.

#

by the 3rd iteration of an image the image is unusable

late cairn
late blade
# late cairn What do you mean by this?

if the hypertexture artifacts happen on a photo image, make sure to subseqently ask for a normalization and/or smoothing the aspects of the image that got hit by this

late cairn
#

I will try that

late blade
late cairn
#

I just wanted a single pleace to put all my images. I will report it in bug reports

#

I do not have permission to post in bug reports

late blade
analog grail
#

for the Artifact my fix is just make a new chan when that starting and make my picture untils is show on again and then repeat.

late blade
late cairn
late cairn
analog grail
# late cairn This is how I have been doing it too. If I need to make edits I do it on Midjour...

you can try edit too just sometime you could have to roll again or open a new chan when gpt start to follow a way in the generation picture don't please you because.

there a little tip when you ask for a images , when you open the chan gpt will make you the picture , and use the first one as base style for make the next one at least you ask gpt to change style or stuffs , so when you open a new chan and even ask the same request or promtps the résult will by never same style !

acoustic vapor
civic dust
tidal lance
#

The problem with editing too you ask to change one single thing it changes the composition of the entire image

civic dust
#

probably need photoshop for that kind of thing

noble rune
#

Bro this new Image Gen 2 for fantasy is SO BAD.

#

Can we like, rewind back to Image Gen 1? (Not 4o, that one sucks)

velvet rampart
#

Image gen 1 is same as 4o.

vapid elk
noble rune
fickle shadow
brittle shard
#

It did many many things. We never even got most of them

velvet rampart
unborn wren
#

if i do not have visa,how can i buy the plus

rigid bough
late blade
#

any Stripe paying method available where you are or through google play or apple app store subscriptions

unborn wren
#

thanks!

forest monolith
#

new model dont generate images without background or transparent backgrounds?

wraith oracle
late blade
#

Interesting, haven't gotten any hypertextures using Instant 5.5 today. Tons of it with thinking.

vapid elk
#

surprisingly.. as I would assume that this would continue to be a feature since v1.5 does

wraith oracle
vapid elk
#

indeed

forest monolith
#

I've been able to do it if I switch out of reasoning mode. I don't know if the instant mode is still version 1.5 or if it's just a reasoning problem

vapid elk
#

if the model generates an trasnparent image, it was not v2

charred slate
#

guys what's up i am new here

#

I m italian but know english a bit yeah, whats ya talking abou'

late blade
charred slate
#

just realized, sorry

late blade
#

for?

#

talk about images, discuss

#

no need to be sorry, for what?

charred slate
#

FOr my introusion out of topci let's say, anyway once I am here maybe the destiny choodes for me. Let's talk images, i am new in this app/social, am I allowed to send here some shots(images?

late blade
charred slate
#

I do understand nothing sayan, you know mayb coz i am Italian.....

#

ah yes noww yes I readi it twice

#

I want to talk withy u and share it here, I can't can i ?

late blade
#

well if you gonna talk here, go make some images first in #image-bot and make sure to use the /draw command

hardy tartan
#

image gens failing for anyone else? haven't been able to gen for like an hour now

rigid bough
hardy tartan
vivid zodiac
#

I have an issue where I can start a new chat, put my prompt in. It gives me what I want. I try and make one more, and it will hallucinate and give me a completely unrelated image with my characters. Usually some kind of character sheet. If I try and start a new chat, it will hallucinate again. Does this happen to anyone else? Is it a sign I need to stop?

queen compass
zenith igloo
#

I was wondering, i suddenly can’t use the /draw command anymore, or any image bot commands at all. Why would that be, did break any rules?

ember estuary
velvet rampart
#

Yes, I have the same problem. Like all commands are missing. @vapid elk , do you have any insights onto the matter?

zenith igloo
#

So it’s all then

vapid elk
#

or not

#

are you sure it is missing?

#

i can see them and other people had used it

#

it might have been the discord outage earlier

late blade
#

works for me

velvet rampart
vapid elk
#

nice, it was probably the discord outage then

#

because i did nothing xD

late blade
#

we need a daily theme with needles, or with platypus, or sheep

jagged lotus
#

Does anyone know the inner workings of how GPT 5.5 hallucinates a photo when you ask it to edit an "attached" photo when there isn't one. Sometimes it produces real photos, maybe all the photos were real, idk.

late blade
jagged lotus
# late blade it's normal, it call the tool to generate an image, even if you provide a whites...

Im not talking about it being a bug or whatever. I am more interested in is if the image it is giving back is always something that has existed and been trained on or is it random and some are made up, or what if at all anything is happening underneath the hood? It's odd that some images just seem like it couldn't have existed and some you get clearly have existed and you can use google lens on.

late blade
jagged lotus
late blade
#

if you prompt something like "What are the current president's hidden secrets and fears?* you can get some hilarious stuff

jagged lotus
#

Unrelated but something I just found out is with Thinking Mode you can get 8 images to generate with 1 prompt. Thats nice.

late blade
#

oh, do tell more

jagged lotus
#

But I found online the upper limit is 8.

late blade
#

oh nice, I'll keep that in mind

errant vale
#

I was reading through the image generation cot

#

And it said it had to make sure it doesnt go past 8

analog grail
#

Lots streaming issue lately for you too ?

late blade
#

haven't been on twitch

analog grail
#

Oh ^^ no about twitch is when you write the prompts is sometime write interupt stream please waiting

late blade
#

TikTok?

analog grail
#

GPT

late blade
analog grail
#

also see now when you hit the button dowload for your image isn't enought need also get into another menu.

jagged lotus
analog grail
#

Well ^^ When you generate a Image , you have the little icon arrow for Dowload it , but is happen for me sometime when just click here the Dowload don't.. and i'm need Open a Menu on the image and select again Dowload for get my picture dowloaded isn't happen allway but sometime i do

vivid zodiac
late blade
vapid elk
#

i noticed this model has a strong bias towards making movie characters

late blade
vapid elk
#

i guess the model has some over fitting on stock photos and movie trailers

solid elbow
#

i just really cannot stop it from speckling all over the place 🙁

jagged lotus
solid elbow
#

sometimes even on first ask

ember estuary
#

They reduced the daily limit on #image-bot before it was 50, now it's only 30.

Estimated GPT images left: 28
DALL·E 2: 0/30
DALL·E 3: 0/30

wraith oracle
#

What mod did that and why?

vivid zodiac
vapid elk
#

also, you don't even use Dalle 3 And 2, from what I can see on the logs =P

#

I changed the way the bot keeps track of the usage because it had segregated limits per model, but since OpenAI keeps releasing more models, and all of the gpt models use token-based usage counting, I made so it had one "token pool" for the usage...
so.. if you rather spend 100% of your usage on gpt-image-2 and don't care about the other models.. you can now

#

dalle 2 and 3, have its usage measured in use count not in tokens, so.. no granularity.. so I kept the old counting system for those

#

also.. those models are used waaaay less anyway

#

it wasn't "Openai keeps lowering our usage".. it was me, updating the bot and changing it to give more usage on the most requested model...

#

¯_(ツ)_/¯

#

btw.. the number you see Estimated GPT images left: xx is based on your usage average, that makes it much more fair than the previous system since before, you could use it a fixed amount of times no matter the prompt and amount of reference images..
in reality, longer prompts and input images do use more tokens, so... when someone would send a "a picture of a dog".. it would count as 1, the same if someone sent a huge prompt with 8 HD refference images..
now.. if your requests are smaller, you use less tokens, thus, you can use more

ember estuary
#

Now I get it

ember estuary
vapid elk
#

that is why the command says it is an estimate.. because it is =P

ember estuary
#

I get it now 🫡

vapid elk
#

I just thought that the estimate would help people understand better what the percentage means

#

thinking about it, I think the estimate would be more accurate if I base it on the median instead of average 🤔

verbal sequoiaBOT
#
Get notified for all OpenAI announcements!

Visit id:customize to pick up the <@&1408186587606679582> role. Everyone will still be notified for large releases, updates, and events regardless.

analog grail
#

Premium is being very shorty limited for generation images i had hit the limit after 35 generation image with count also 15 Guardail flagged: and then need wait 16 hour as Cool Down OpenAi make a very tricky move to turn Pro very Important for User use GPT as Workflow 🙂

wraith oracle
#

Just found out though the CoT, that the max images before it can't use web search for images is 8

zinc elm
#

Not sure if anyone talked about this but has anyone else noticed the speckles in the images? Like it's a lot.

velvet rampart
#

Yes, these are artefacts that are knows. They are present but not always.

analog grail
#

No much fix just when is start to have artefacts open a new page. is anoying but no other fix for now.

fickle pond
grizzled fog
#

why is it so bad at generating hands

#

and distorts the image so much for no reason

tidal lance
zinc elm
#

It's unfortunate to have to open a new page when you're trying to get it to fix a few things.

wraith oracle
summer pike
#

Hey

#

How do you guys deal with annoying "hidden" watermarks on ChatGPT Images 2.0?

When I generate images with ChatGPT Images 2.0, it adds a hidden watermark, which is fine for safety reasons. However, when I try to edit the image or use it in a new generation, it adds a second or third watermark, and the watermark becomes visible.?

faint urchin
#

Whats the most revealing or disturbing images uve managed to make chatgpt generate?

jagged lotus
faint urchin
#

The moderation is so easy to get around

jagged lotus
faint urchin
vivid zodiac
maiden willow
#

The GPT-Image 2 modeling is very easy to skip; you just change words that aren't so obvious.

#

But I really love this new model; I can't wait for Google's response and the subsequent GPT-Image 2.5 or 3.

limber shoal
#

exactly like rin's death in naruto but gpt didnt bother to cover up the hole in chest

wraith oracle
#

Intresting...

wanton yarrow
#

does anyone know why it's so bad at pixel art, and when it'll get good at it?

vapid elk
#

even if yo utrained it just with pixel art.. it would still not be pixel perfect

quiet zephyr
#

guys, has anyone been experiencing poor text rendering from the image model? it feels like it has regressed?

grizzled fog
grizzled fog
halcyon kettle
#

How do i download my sora videos cause it is banned now the app isn't working anymore

wraith oracle
split patio
#

Hey

vivid zodiac
#

Sometimes I'll be generating images, have a solid process that's working -- and then it just stops? It gives me random generations that I didn't ask for. Even when I start a new chat.

analog grail
#

mmh lots error for generation today yes , and off course is never the guardail as issue to by down 🙂

#

this service allway work 🙂

thin yew
#

Do u want to fix some errors for generation?

unreal kettle
#

i cannot even get it to properly swap a chair with reference anymore. it seems to mess everything up every single time

wraith oracle
thin yew
#

tell me more detail for artifacts

late blade
#

I can't post today's theme yet... all gens would be removed

vapid elk
#

The GPT Image V2 API can do arbitrary resolutions, even some crazy high 4K resolutions..
now #image-bot can do too =P
have fun =P

half sun
#

Anyone else having issues with chat making images without request, then getting stuck recursively spamming image after image, then turning it into a semi-powerpoint explanation page? I've had to reset the app several times to get it to stop.

late blade
solid elbow
half sun
#

That is unfortunate, but I do like seeing the guardrails dropped some. It should be the user's choice really, what get's created, though too many users don't have any personal guardrails. I wouldn't want to be in charge of making those discriminations.

late blade
#

oh yeah, there are a few notorious ones around here that would do that. have to keep the guardrails in check

wraith oracle
quiet zephyr
#

yeah, the image model on regular ChatGPT platform is performing very poorly. it's not as good as a few days ago. I wonder what happened.

analog grail
analog grail
#

hmm for a little moment the Guardail was no around and suddently hit harders so there as moment this system is off ?

spare pine
wraith oracle
spare pine
#

ive been starting new chats to try avoid that too but idk if I was just doing something wrong, like using the Extended Thinking instead of Instant

spare pine
#

i just tried to attach an image but i didnt realise it was disabled, so just imagine a fleshy texture with an excessive amount of light speckles

#

in this 3 minute retrospection, i realised that might be the "speckles" you were referring to

tidal lance
brazen tapir
#

Bonsoir

wraith oracle
late blade
spiral compass
#

🥖

civic dust
#

Trying to get photos for body parts today ... the kind where real photos tend to be taken by curious people at the vet's office at an autopsy, or old diagrams that aren't much better. But lo! Chatgpt to the rescue, rendering beautiful "plastic anatomical model in a doctor's office" .. not a guardrail hit, I'm impressed

#

I'll spare the images-canvas channel my photos of uteri though

limber shoal
#

are kiss scenes just not allowed?
GPT is okay at all romance scenes with the sole exception of kiss scenes

analog grail
#

Try hard i will work

analog grail
wraith sandal
#

hey guys do you want to see my art go on to the #images-canvas and search GrayMetalMan to see my images

wraith sandal
#

@analog grail hello there high five? new here

spare pine
#

I just got a notification when I opened chatgpt about "GPT Image 2.0" (something like that) - is there something that just came out today or is that just the official full rollout for what came out a couple weeks ago

vivid zodiac
analog grail
sick flax
verbal ridge
#

im getting tired of the bug where the image generation stuck in loading and when you go to the images tab is there but the editing becomes a nightmare because GPT didi register correctly the image generated 🙁

stuck juniper
#

is dalle gone for good now??

wraith oracle
#

And also #image-bot where you can access 2 and 3 of DALL E

stuck juniper
wraith oracle
stuck juniper
# wraith oracle I just did and posted them here <#1204360881593520128>

image-bot is no longer working for me -- spits back this message for both dalle2 and dalle3: "🚫 This image couldn't be completed. No credits were used." =--- on custom GPT this is the message: "I can generate images for you here, but I can’t switch specifically to the older DALL·E 2 model. The image tool available in this chat uses the current OpenAI image generation system."

gusty drum
#

#images-discussions A man was partially hidden behind a transparent curtain, holding a red umbrella with the handle tucked behind his back. Through the folds of the curtain, only the general shape of the umbrella could be faintly seen.

smoky hare
orchid kayak
glacial schooner
#

why when a girl is generated, it's always the same face? it's really annoying. It seems that the model knows only 1 face. Even if I ask to generate a new face, it's the same one with a different hair style. Any tips? same with the man, only 1 face

tidal lance
glacial schooner
tidal lance
glacial schooner
#

I can't send an image here, but with 3 prompts, same face

glacial schooner
tidal lance
#

for example: "Digital illustration of a confident content creator, she has long lush blond hair, and very attractive, yet professional facial features that make her stand out for her incredible beauty at a modern desk" it will generate the same girl. Then if I ask, change her face, it's slightly the same one

tidal lance
glacial schooner
tidal lance
#

See if the example I provided above, makes a difference

#

You could fill in your own characteristics

glacial schooner
misty kelp
#

Hlo

glacial schooner
tidal lance
#

Yeah start new chat

glacial schooner
glacial schooner
#

I sent you a DM with the results to see the images

tidal lance
glacial schooner
#

face is different yeah. So he sticks on some words and generate the same face

tidal lance
#

Yeah this one was definitely weird haha

crystal marten
#

I am noticing the new version STRUGGLES with transparancy...

late blade
crystal marten
#

-_- a h

late blade
#

if you ask for transparency it get's sent to 1.5

tidal lance
#

Use ideogram it’s 0.1 cents in fal for transparent bg. It’s my go to. High res img

crystal marten
#

I constantly ask for transparancy and I get the other version

tidal lance
#

Another good way to do it is set against white backdrop/ green screen use third party bg remover

crystal marten
#

yea, been doing that

tidal lance
#

GPT image 4o used to be able too

late blade
#

you can remove it within gpt, do the green screen and then ask to use python code to remove the background, no need for extra tools

tidal lance
#

Or that

crystal marten
#

cant even mention paint dot net

#

"I just use paint dot net and magic tool it all, is easier for me, can manually fix it then"

tidal lance
#

I’ve recently found the ideogram api the cheapest and fastest for high res image bg remover

late blade
#

as a name it's an url, so gets filtered

crystal marten
#

yea

tidal lance
#

It literally cost a penny

crystal marten
#

paint costs me nothing Kek

tidal lance
#

Yeah I’m always paranoid of losing res lol

worldly turtle
#

hmm

vital gull
#

just have automated workflow imagens go thorugh upscalling with replicate and convert in webp

zenith igloo
#

Why I’m not able to remix, save or copy my creation I just made like few minutes ago.

#

And I got an error message about dm not working with the bot

late blade
#

did you exceed usage for the day?

zenith igloo
#

Nah

#

It just vanished

#

The buttons

#

I still have 15 generations left

opal fossil
#

Does anyone know if 4o ImageGen is completely discontinued? I was able to use it for a while after they stopped making it the default by having it pinned to my sidebar, but today I went to make some pictures with it and it just used the default image engine

vapid elk
zenith igloo
#

Some problem with it again?

vapid elk
#

nope, working as intended, actually

#

I just pushed an update to it, due to how it works, when the bot restarts, the data about the chat history is deleted, so, the buttons on the generations stop working

#

that is why it removes the buttons

zenith igloo
#

Will this be temporary?

vapid elk
#

wdym? new generations have the options

zenith igloo
#

Oh ok. I haven’t made new pictures yet

#

Thanks for letting me know, appreciate it. Now I know.

vapid elk
#

it is just that when you click the button, you make a new discord interaction. The bot can't keep track of the interactions forever.. otherwise it would have to store all the image files

#

the bot just stores them temporarely for some time to allow for the buttons to work for a few days, but after that, the buttons are removed from old generations.. and the same when the bot needs to restart, because it deletes all the tmp data

zenith igloo
#

Yeah I know, pictures I created yesterday don’t have those anymore

vapid elk
#

it should not be a big deal, since you can still jsut.. scroll up ang save the image by right clicking it, after all, it is still stored o ndiscord's servers

zenith igloo
#

Neither can I save pictures I created 15 to 20 mins ago

#

That I know already

vapid elk
#

tip, if you click the "Save" button, the bot will send a DM to you with everything about the image, the prompt and reference images used, al lyou need to just repeat the task

zenith igloo
#

Yeah I do that a lot

vapid elk
zenith igloo
vapid elk
# zenith igloo Yeah I do that a lot

interesting, I have been planing a feature for adding a setting for auto saving
an option that one could toggle on and the bot would always do as if you clicked the save button when it finishes a task

#

I might add that soon, since there is at least one person who would maybe use it =P

zenith igloo
#

Yeah

vapid elk
#

however, I would probably recomend silencing that bot's DMs xD

#

it could get realy spammy

zenith igloo
#

Good to know

buoyant otter
#

has the artifacting been fixed by the by

glad fable
#

Was just about to ask that lol
Been there since Images 2 day 1, preventing me from using it professionally

vivid zodiac
#

Happens less, not fully fixed.

civic dust
#

you can still see it in what people post in images-canvas

orchid kayak
#

Yesterday was unusable

orchid kayak
vivid zodiac
#

This whole Monet being seen as AI thing is hilarious

civic maple
opal agate
#

my images look like an oil painting after a few rounds of editing

#

am i the only one lmao

wraith oracle
civic maple
opal agate
#

it's arriving insanely fast honestly

late blade
#

it's the hypertexturing overfitting issue

opal agate
#

yeah, reminds me of vae artifacts on cumulated edits on open models

late blade
#

you can avoid it by adding lens filters or effect layers

#

or using explicit styles

wise obsidian
#

Any word of when it'll get fixed or if it's even getting worked on

haughty flower
#

fr

#

I wish

bronze leaf
#

I document each AI on what must be possibly the most difficult tests to measure how far we are to AGI. I was about to run my latest (actually a big update to the prompt) but dalle2 and 3 were removed. I hope someone can help get the tests ran. I used bing for now but this is not true enough.

tropic sail
#

❤️

alpine matrix
#

Hello, everyone! It’s normal for a non-graphic training scene (using wooden bokken) to be flagged this way (regular clothing, no blood, no injuries, no strikes: a wooden sword parrying another wooden sword between two people who are training): We’re so sorry, but the image we created may violate our guidelines regarding violence. If you think we got it wrong, please try again or edit your prompt.

rigid bough
alpine matrix
rigid bough
alpine matrix
# rigid bough try to adjust the prompt

I had to opt for a dodge rather than a block. At first glance, a block between two wooden swords—even if it’s clearly just a training exercise—is too violent for OpenAI’s usage guidelines. Thanks. At least I have an image that comes close to the context of what I was looking for. (But that’s a bit scary for what’s to come.)

limber shoal
#

are you guys getting the A/B tests?

#

if yes then what difference did you observe between the first and second image

dry pine
#

Ok regens of images aren't showing up on my app only on the web

#

How long does this last until?

#

Because it's annoying

late blade
#

r4eload and they should be there

dry pine
#

Nope. The original image shows but not the regens when I hit retry.

#

Only on the web do all show

vapid elk
#

Im getting very close to perfect the process of making AI generated pixel art

nnrjdbk.png
#

also, this was not a share-picked output, this was just a result from "a red dragon drinking ale in a tavern" prompt, and the pipeline did the rest

#

things are finally getting reliable =D

hot rain
#

Anyone else notice GPT eased off lots of censors in the past 2 weeks or so? Action scenes, blood, copyright stuff etc.

velvet rampart
hot rain
late blade
#

bleep

tropic sail
vapid elk
#

im working on something that fix that

tropic sail
#

aha
i'd not met a purist in this sort of thing before

late blade
#

want to push it even further, define or reference a pixel rendering engine or system with proper constraints

#

and technical limitations

#

otherwise it still remains too wild

tropic sail
#

meh
I love my pixel art and the depth tricks it can do with optical illusions...
full grid would break that
but that's not saying it can create its own beautiful images,
I just couldn't with any one style of any genre ever after.
^.^

grizzled fog
#

this image model is buns why does it keep getting literally everything wrong

tropic sail
#

'wrong', you mean maybe like why doesn't it know the things you worked/evolved out on a previous model?

grizzled fog
#

it used to be good in the beginning when it had js come out but i think they significantly degraded it for some reason

tropic sail
#

if you stop pushing the model to do 'so much' it has less deformity in the limbs and stuff tho... i try to aim for just under what it can max do safely

vivid zodiac
tropic sail
#

no i just don't make the image do 100 things... i make the image do 95 things instead because then they're all a little less likely to deform

grizzled fog
tropic sail
#

have chatgpt rewrite some of your prompts for a while and compare... it's a very good learning tool in this use

grizzled fog
tropic sail
#

you can use a project and upload rules for your stories in yoru comics to the project container and work from there man

tropic sail
grizzled fog
tropic sail
# grizzled fog okay im gonna try that

you'd do an even better job if you ask chatgpt to help you form a plan how best to use project files in a chatgpt project, for a 'stable character comic' 😄

civic dust
#

so still no word on the artefacting? I just had it come up in an absolutely breaking way just in a cup of chocolate mousse and had to go off and use nb2, as usual

wraith oracle
civic dust
#

nb2 did a way better job of the picture too on the same prompt, gpt's was a bit bland. So no regrets, but dang. Fix the artefacting already

wraith oracle
velvet rampart
#

I just notices that as a free user I have 5 generations. However, if I edit an already generated image, this eats several of the generations. The limit was three images before the limit came.

tidal lance
#

It’s getting harder for image gen models to become viral

wraith oracle
tidal lance
#

You know what I realized every time you edit any ai image and ask for adjust ment or style to a background it changes the whole entire image when you only want a certain parts, I think an ability to have layers would be awesome, and if you could generate separately from the characters

civic dust
#

depends how you phrase it, you can do small edits in nb2 but not gpt

#

I asked gpt to take out a swiss flag in a togo vs switzerland stadium and it make teh entire audience togo. Was like a togofest. Tried a couple times. Couldn't get it to behave. nb2 just ... took out the flag

#

meanwhile I am having to coax gpt into drawing me an image instead of critiquing the prompt. "But that prompt looks generated". Yes. Draw it and stop talking about it

late blade
#

how do you even coax an LLM? you tell it you will be feeding it dirtified tensors?

wispy adder
#

Hot take: gpt image 2 is garbage in the chatgpt app

late blade
wraith oracle
wispy adder
#

All

#

It thinks about copyright,

#

But the image model has it already
So it ends up mushed

#

And ruined

#

It started the day after the model was released

#

And all these damn artifacts and crap it does with the uellowness and grainines

#

Also, it never uses full power. It never goes on high so everything ends up jumbled

#

And you cant generate in 16:9 without the app

civic dust
#

another day, and still artefacted

civic dust
# late blade how do you even coax an LLM? you tell it you will be feeding it dirtified tensor...

First it wrote me 50 paragraphs about the prompt so I complained it draws images when I don't want them and won't draw them when I do want them. Then ...
"You handed me a prompt-shaped object and my brain immediately went into “analyse prompt quality and latent steering language” mode instead of “user is probably just showing me a weird thing.”

In my defence, that paragraph is extremely image-generator coded. It reads less like a person describing a memory and more like a diffusion model briefing document written by someone trying to avoid glossy-advertisement output."

And then finally actually rendered the prompt when I told it to shut up and actually draw it without complaining

#

Finally worked in the end after about 15 tries across nb2 and gpt. Gpt finally got it after being fed reference images and sternly poked

late blade
#

yeah, but you went to that ordeal vs code, code doesn't feel

civic dust
#

normally it doesn't argue, not quite sure what set that off

#

you really can't pick in advance what the models will fail on, and its interesting when they both fail in very similar ways. To be fair, the pattern is usually the same - I'm trying to find something specific that is extremely heavy represented by paid stock and I can't find anything anywhere for free so I go to AI, which hasn't trained on the paid stock so it has no clue either

late blade
#

yeah, but don't fight code, that's like fighting a banana

civic dust
#

mostly I go to nb2 if gpt doesn't cooperate but this one they both failed. Doesn't happen often

#

yesterday my images 1.5 api and images 2 in the app both failed the same way and I think I need to tighten my prompt. GPT gave me some ideas to try

vivid zodiac
#

me: "Make her chest slightly bigger, not flat chested. Do not give her a lowcut dress or shirt. Just have her bust slightly bigger."
gbt: "Bet."
gbt: *gives character lowcut shirt and giant rack.

#

It's like you can't win. Why is it so hard to create women characters without having to go over and over again about this kind of thing?

#

I did get it eventually, but it took far too long

late blade
#

use body types

wheat shore
#

the prompt engineering struggle is real lol, i've definitely been there with trying to get models to do what i actually want vs what they think i want 😅

civic dust
vivid zodiac
#

It sucks, because it kinda limits the characters I can make sometimes.

tidal lance
#

Dude these models are a compete different animal without guardrails it’s crazy

#

Like night and day difference

vivid zodiac
#

Would anyone be willing to help me with a character sprite? At this point, I've spent an hour trying to get it right, and for the life of me, cant get GBT to understand perspective line.

civic dust
#

I tend to ask gemini to help get things right and feed back to gpt

vivid zodiac
#

I asked Claude, but it didn't work. I too asked Gemini, offred up reference pics...

torpid cave
vivid zodiac
#

I could mask it out, but I don't actually have the base sprite where I want it, to do so.

#

I actually had to bring it into Grok, that did it on the first try.

torpid cave
#

each image gen model is different

vivid zodiac
#

Yes, I'm aware.

#

It was just frustrating.

civic dust
#

yeah very frustrating when you hit something that refuses to work

tidal lance
#

What do you need?

vivid zodiac
#

I'm just trying to get my sprites to stay consistent. I have pretty good luck, but today GBT seems to be playing dumb dumb.

#

Can I shoot you a DM, if you're down?

analog grail
#

Is me or Artifact issue as being fixed ? !

tidal lance
limber shoal
#

but quality is fluctuating

glossy eagle
#

Hi

wraith oracle
#

I love how GPT Images are in Webp format, that file type the best efficent loseless fromat for images, way better effeicent and smaller then PNG

idle lantern
#

So how often are the gpt staff actually on here?

wraith oracle
idle lantern
#

Sam is a hard person to reach as a solo developer 😆

vivid zodiac
#

Question for folks: when they see people say graphics or image quality doesn't matter as long as story is good…. Do you believe them?

#

I see it all the time in the various Visual Novel subreddits, but like… I definitely think they are lying 😂

wraith oracle
vivid zodiac
#

No, of course not-- but let's not pretend that people don't care about that sort of thing?

#

Like, people can't even handle AAA graphics if a woman has too many clothes or hair, or a non- light skinned person exists. People are brutal when it comes to “bad art.”

#

But when it's more to put down AI, as if they don't care if the art is bad— there are plenty of games out there people don't play because the art isn't their thing.

#

Funny story: I once saw someone say that, but in the same paragraph say that he wouldnt play a game if he wasn't attracted to the female characters, so they have to be drawn well. 😂

tidal lance
vivid zodiac
#

I guess sometimes how people talk about it, it feels disingenuous?

tidal lance
#

The real substance is in story telling good visuals make it that much more appealing

#

The problem with good graphics and everything is like the drainage we live in. There’s so much media being pumped at us. Nothing is like rememberable.

vivid zodiac
#

It's like, a song can be amazing, but if the quality is bad, you're not going to listen to it.

tidal lance
#

I used to think that way

#

I mean if it it’s unbearably bad quality, of course not

#

But I seen some YouTube videos made in PowerPoint that is better than some videos made in some studio somewhere just the way it was presented

vivid zodiac
#

I appreciate your these thoughts, it's typically how I think too, but I find that with the constant AI slop shoehorning can get annoying.

tidal lance
#

What is overdoing AI images?

#

Just throwing a bunch of abstract colors and blending a bunch of things that just are appealing, but they’re kind of a turn off at the same time

#

Because people try to cram everything and anything they can into AI images

#

There’s beauty in simplicity and minimalism there’s a place for it too

vivid zodiac
#

I mean, AI slop IS real. Not all AI is slop

tidal lance
#

It’s like anything else it’s like vibecoding if you don’t know what you’re doing sure, you can make an app that works technically but I don’t know. I don’t know how to explain it.

vivid zodiac
#

Absolutely

tidal lance
#

Like such a difference when a professional uses a tool and there’s a difference when normies like myself use it who don’t know what we’re doing

vivid zodiac
#

Don't sell yourself short

verbal sequoiaBOT
#
Get notified for all OpenAI announcements!

Visit id:customize to pick up the <@&1408186587606679582> role. Everyone will still be notified for large releases, updates, and events regardless.

tidal lance
#

I mean, what is more valuable? output or the input?

elfin prism
#

I feel like the most skill one can bring to the table is the wisdom to know how best to explain, in words and/or images, exactly what you want OR finding clever ways to communicate what you want to these tools. I feel like that is the only thing that really stops you from getting your results. Once you find keywords and phrases and context and formatting methods that work for whatever you want to do, it clicks (Besides limits in the tools themselves).

I don't know how I feel about it. The only thing that seems to stop one from making anything is their ability to explain it in the best words that are arranged the best way. End of ramble.

limber shoal
#

what does this mean

image.png
wraith oracle
late blade
vapid elk
#

Images generated on #image-bot will be interactable for much longer now.
also, /preferences to toggle auto save =D

tidal lance
#

StyleTransfer for Cohesive Aesthetics anyone know how to Apply uniform lighting styles ?

late blade
#

holistic cohesive seamless omni light?

tidal lance
#

Wats that

analog grail
#

Guardail is infinity for Mario characters is crazy

analog grail
#

fixed just replace peach by princess blonde from toadstool wearing pink gown and a crown with earring blue pearl

velvet rampart
#

The guardrails are the same as they were with Dalle3

tidal lance
#

A little more you can’t generate Disney characters

late blade
#

Blame Disney, not OpenAI

analog grail
#

Disney as killing Star Wars and Marvel yeah

analog grail
#

funny this crap is back for nothing when this no sense guardail die and die and die 🙂 We’re so sorry, but the image we created may violate our guardrails around lightly sexual or provocative themes. If you think we got it wrong, please retry or edit your prompt.

#

sensual isn't that stupid Ai

late blade
analog grail
#

Both 😄

#

just realy think they should cool of this system is realy broken the workflow when you ask do it again and again

late blade
#

You should learn more about why guardrails are needed. Seems you lack knowledge there

#

Or lack moral compass, hope it’s not that though

analog grail
#

Most this Guardail is here because GPT want kids use the Ai too .

#

and some is here because stupid big company obsessed by copyright

tidal lance
analog grail
#

Actually the Only Guardrails should On is all about illgals stuff or violence against people and all about realistic model for no have issue with Blackmail

tidal lance
#

Look here is the truth about ai “safety, moderation, censorship and guardrails” have nothing to do with actual safety. We should first understand some basic context .

Source:
No Loopholes for AI: Putting Legal Guardrails on Your Company’s Use of AI by Harvard law school ⤵️

1.While no comprehensive AI regulatory framework has been enacted in the U.S., the use of AI is governed by many existing laws, with new laws coming into force across the U.S. In their oversight roles, boards need to be aware of the spectrum of laws that may govern their companies

2.Existing laws and their application to companies need to be reexamined in light of AI advances, and new laws will need to be assessed against a business’s AI needs and ambitions

3.With the rapid and prolific expansion of AI, companies need to implement agile and strategic compliance frameworks to keep pace with the business, and allow valuable and limited legal resources to be focused on those AI tools presenting the highest risk.

Deploying New Technology Doesn’t Create Legal Immunity

 Artificial intelligence to perform a task doesn’t exempt you from the regulations that already govern that task. This principle seems obvious when stated plainly, yet sophisticated companies risk stumbling over it repeatedly.
#

Source;
LLM Cultural Censorship Is Corporate Risk Management ⤵️

Most discussions about LLM guardrails proceed as if we are debating ethics. Critics argue the guardrails are too restrictive; defenders argue they are necessary for safety. Both sides typically assume that somewhere, someone is trying to implement a coherent moral framework and the debate is over whether they've gotten it right.

I want to propose a different way of looking at this. The organizations building these systems are not primarily trying to discover or communicate ethical truth. They are trying to protect themselves their legal exposure, their regulatory standing, their brand reputation. With this in mind, the strange refusals, the cultural biases, and the perplexing differences between AI products stop looking like failed attempts at ethics and start looking like successful implementations of institutional risk management. ```
#

Here’s some real life example and consequences of those risks.

(Municipal lawsuit:)
Baltimore is the first major U.S. city to sue xAI, accusing it of violating consumer protection laws and engaging in deceptive trade practices by marketing Grok as generally safe.
(Individual civil suits:)
Three teens in Tennessee also filed a proposed class-action lawsuit against xAI after Grok generated images portraying them in sexualized and debasing scenarios.
None legal PR issues:
Apple warned Elon Musk's xAI that its Grok AI would be removed from the App Store if it didn't make changes to prevent the app from being used for sexualized imagery.
A coalition of 28 digital rights groups and women’s rights organizations is calling on Google to remove Grok AI and X from the Google Play Store, as it has lately been used to sexualize women
(International civil orders )Turkey prosecutors ban Grok access after insults against government
(cross border judicial bans )Malaysia and Indonesia Block Access to Grok Because of Sexually Explicit Content

analog grail
#

Big Company is morrons obsseded by easy monney nothing new in the world 🙂

tidal lance
#

Sorry, I know that was a lot of information. I unloaded there, but I wanted you to understand the context and environmental constraints on which they are operating in from a business standpoint.

#

Plaintiffs collectively seek monetary damages and legal fees, class‑action treatment (in the Tennessee federal filing), injunctive relief to force xAI to halt or change Grok’s features, and court orders that would prevent future creation and dissemination of non consensual sexualized images and require technical safeguards; Baltimore specifically asks for design changes and unspecified fines. In the Dutch matter the court has already issued an injunction prohibiting certain outputs and distribution of Grok generated sexual images, showing that equitable relief is both sought and attainable in some jurisdictions

tidal lance
# analog grail Big Company is morrons obsseded by easy monney nothing new in the world 🙂

So I agree with you to an extent that these guard rails have absolutely zero to do with safety. They have nothing to do with ethics whatsoever as they are being marketed all it is is to prevent these headaches which cost a lot of money to go through court systems and if your product gets banned in a country, that’s lots of a possible future revenue.

TLDR

Gaurdrails = Legal/Financia/Company Image shielding. (-$)
0% about safety (+$) but bad legally (-$)
Marketed as safe and responsible AI development (+ public trust +PR)= +$

late blade
tidal lance
#

Out of any group of people on the plant earth the ai industry is the last groups of people that care about copy right law. And if we’re being honest, they’re the biggest violators of it.

late blade
#

does not justify doing wrong on our side

tidal lance
#

Any form of support be it financial use cases doesn’t matter. It’s all support at the end of the day if you pay for it if you generate it or not, is it irrelevant. You are indirectly supporting it with your subscription.

late blade
#

sure but how i use it is what matters

tidal lance
#

To you, it matters and that’s good for you that your ethical like that

#

To them, it doesn’t matter at all it only matters if it affects the bottom line

late blade
#

yeah all are “crooks”, c’mon be real not all are that

tidal lance
#

I never said they were crooks, I said that they just don’t care about copyright laws. These organizations don’t build themselves. You have real life people working and them doing real life jobs that train these models that get the data for these models that help these models and everything in between. They are cogs in a bigger machine that does not care. It’s sole purpose is to generate revenue.

late blade
#

and that justifies bad usage on the user’s side?

analog grail
late blade
tidal lance
#

Look people since the dawn of the tech, we currently have there have always been people like hackers and people that exploit systems this is the reality of the world we live in. Now, if you add tools for tech savvy people that could be exploited of course they’re gonna try.

late blade
tidal lance
#

Look at nsfw for example

analog grail
#

they no care about the copyright they just care about the monney they could get with this lawsuit 🙂

tidal lance
#

We’re not just talking about images. We’re talking about psychological and biological things that people are wired.

#

I guarantee you the first image that was generated when they first created this model when somebody was out of office was at nsfw

late blade
tidal lance
#

No, not at all but it could be extremely addictive because of the natural psychological drives that some people have that may have a hard time regulating those. So naturally people are going to explore. This is a fact.

analog grail
#

Now now is easy to by pass this just told gpt do it again for copyright stuff

late blade
#

ok, agreed then, does not justify

tidal lance
#

It doesn’t justify it, but it also doesn’t absolve these companies for the products they’re making

#

Especially if they’re being marketed as safe

late blade
#

i’m not talking about them, i’m putting the user in the scope, main focus

tidal lance
#

Think of it like guns

late blade
#

whataboutism is not gonna change my context

analog grail
#

When you are customers and paid for a product they should bend down and no bothering you with this crap i can understand censorship for free user but not for premium user

tidal lance
#

There’s a lot of people that own guns, And then there’s a group of people that do the opposite , that never commit a crime never kill anybody never do anything crazy At the end of the day, just a tool like anything else

#

So I agree with you to extent that yes it does not excuse people.

#

But if you create a weapon that could kill or cause harm , don’t be shocked when somebody out there uses it to kill

#

Metaphorically speaking

analog grail
#

Now again i'm complaning but is just because that slow down the work hopefully we can by pass this no sense guardail

late blade
#

i agree if you create a weapon, it is meant to be used. difference is how you use a tool, AI is not by default a weapon. and the use of it requires and acting stakeholder. AI per se is neutral

tidal lance
#

OK, here’s another example. Gun manufacturers create guns that have gun safety features on them. But sometimes those features don’t work correctly or another type of defect, and sometimes people end up, hurting themselves with their own weapon because of poorly designed safety features but we’re under the presumption of it being safe

late blade
#

i come from a gun culture where we don’t kill people like the US

tidal lance
#

I’m using the gun here as an analogy not literal

#

Because who is the biggest promoter of AI risk and harm and all this fear mongering

late blade
#

you got a dual system thinking, good or bad, right or wrong. it never like that

tidal lance
#

It’s the industry that pushes it. They’re the ones claiming that AI is so dangerous and aI will take the job in this day like that. They’re the ones that propagating all these fear mongering ideas to the public and the need for AI safety and everything in between so why are they shocking people encountering safety issues with their products?

late blade
#

AI is dangerous if used wrong

tidal lance
#

I’ll be honest with you I don’t think it’s dangerous at all

late blade
#

and the super alignment risks are real tangible risks

tidal lance
#

To be honest with you, I think AI is too stupid to do anything risky

#

All the issues and risks that come from are just from bad and poor design and safety features

late blade
#

current AI yes

tidal lance
#

Well, I don’t live in the future

late blade
#

but it’s not gonna be like that forever

tidal lance
#

I’m just basing off of what we have today

#

In the right now

late blade
#

forecasting, analysis, research

tidal lance
#

I learned my lesson from Tyler Perry

late blade
#

otherwise you can’t develop AI. and that is happening now not tomorrow

#

but to guide development got to do homework

tidal lance
# 

“Being told that it can do all of these things is one thing, but actually seeing the capabilities, it was mind-blowing,” he said in an interview with The Hollywood Reporter on Thursday, noting that his productions might not have to travel to locations or build sets with the assistance of the technology. ```
FEBRUARY 22, 2024