images-canvas

[thin meadow]

The lil cuties! Haha

[rain cove]

panda panda!

[errant crane]

i enjoyed reading your write outs 🙂 i started messing more with presets because of them

[thin meadow]

I have media uploaded to this preset, see if it copies that: https://sora.com/explore/presets?pid=preset_01jr848bayftsa37q22wvcs2f5

[thin meadow]

I'm glad to hear that, honestly. I feel like presets showcase what Sora is capable of and how dynamic it can be between prompts.

Sometimes when I make a new preset, I revisit my archived stuff just to see how it changes it. Haha

Same prompt, different presets:

True.

[warm escarp]

wait what does this mean

[thin meadow]

I think so...

[errant crane]

probably include reference images?

[thin meadow]

I think that might be the case.

[warm escarp]

@thin meadow used your preset on my selfie , i am loving it

[thin meadow]

This is honestly my favorite preset I've made yet.
I wanna make two for this one in the future, though. One for interiors and one for exteriors.
Great image, btw.

It also looks great on cars if you ask for no humans in the prompt.

A luxury hypercar named Veyronyx GT parked inside a high-end car dealership showroom. The car has a long, aerodynamic, and sharp-edged body design. It is painted in obsidian black with subtle matte gold accents along the sides and rims. The setting is a sleek, modern showroom with glossy black floors. The car looks fast, powerful, and elegant. There is no people.

[warm escarp]

do u have listof presets yet?

i mightve missed it here

[thin meadow]

Not yet, but I'll send you over one real quick.

Now that I can share links

[thin meadow]

VHS (needs work): https://sora.com/explore/presets?pid=preset_01jr848bayftsa37q22wvcs2f5
Neo-Noir Futuristic: https://sora.com/explore/presets?pid=preset_01jr7mwm4ne8dbpakefwjmb4ep
Teal Cinema: https://sora.com/explore/presets?pid=preset_01jr6vz11ce2p8e5j3rswyn9s3
Cotton Candy: https://sora.com/explore/presets?pid=preset_01jr6r366pensbchdajfh776ew
Vaporwave: https://sora.com/explore/presets?pid=preset_01jr5np90fecbsbnbkpg3xe4tc
Natural Selfie: https://sora.com/explore/presets?pid=preset_01jr5kv0aje8armvd0v54kjsr1
Red Photography: https://sora.com/explore/presets?pid=preset_01jr09kb4ve4s8ys5qv04874xp

[warm escarp]

i just noticed ur noir futuristic preset will output like a brand new composition idea but maintaining the subject reference , while the vhs one stays true to the original image user input but added the vhs effect like

ah the vhs preset do have references media , this is cool

[thin meadow]

Yes, you'll notice the same with the Vaporwave, it introduces elements into the image, such as 90s malls, graphics, etc. There are two ways you can handle presets, keep them general so they just change color stuff and effects, or you can introduce elements. This is why I wanted to create multiple for the Neo-Noir Futuristic so I can treat interior and exterior architecture differently.

[warm escarp]

adding ur preset made me gain the ability to output 1 6:9 image now /jk

i think this is aglitch

[thin meadow]

Essentially all the preset is doing, I suspect, is adding what's in the preset to the end of the prompt behind the scenes. This is great because you can save on redundant prompting for styles.

[thin meadow]

Cheating the system. Haha

[thin meadow]

Really like this generation.

[outer hinge]

Wow, love this

[errant crane]

ive seen a lot of very swole harry potter casts.. sometimes you cant help but have some fun

[thin meadow]

Making me feel bad for skipping my workout today...

😆

[warm escarp]

the creature who endured... the boy who lifted

im cackling at these clever title

[errant crane]

gpt came up with it, i gave it the "the boy who lifted" and asked for a similar one for gollum

[warm escarp]

https://sora.com/g/gen_01jr6n6c3xebz86t7kerz63r6d wait this was yours?

[thin meadow]

Haha

[errant crane]

no no, i just keep seeing them as i scroll 😄 so i remixed one

[warm escarp]

VHS (needs work): https://sora.com/

[rain cove]

omg thats the best

so sora does images too?

[lusty snow]

I have a custom GPT trained for prompting Stable Diffusion 3.5 large and re-tooled it for the new Sora\Dall-E, it came up with a decent little schema that I've had a lot of success with. It's very similar to one of your screenshots but this might help refine yours too:

Backdrop, setting, time of day, environment, spatial composition.

"sunset over a misty alpine lake, surrounded by dense pine forest"

2. Subject
Main focal point(s) of the image; describe physical characteristics, actions, position in scene.

"a cyberpunk woman in a hooded jacket, standing knee-deep in water, gazing into the horizon"

3. Mood & Atmosphere
Emotional tone, ambiance, and any visual effects that contribute to the scene’s mood.

"eerie and cinematic, fog rolling in, lens flare on water, subtle film grain"

4. Style & Medium
Artistic medium, lighting style, or rendering approach (e.g., oil painting, digital art, photorealistic, low-poly).

"cinematic lighting, ultra-photorealistic, shallow depth of field, anamorphic lens effect"

5. Color Palette
Optional, but useful for mood and visual cohesion.

"dominant blues and oranges, with soft pastels in the clouds"

6. Composition & Perspective
Camera angle, focal length, frame orientation (e.g., close-up, wide shot, isometric, top-down).

"wide-angle shot, third-person perspective, subject centered, rule of thirds"

7. Detail Modifiers
Micro-level descriptors like materials, textures, small props, background elements.

"wet fabric reflecting light, neon tattoos visible on arm, birds in the sky"

I typically just hand ChatGPT a prompt I want it to expand on and tell it to fill out a full, detailed prompt from that schema (with some tweaking, like don't include or statements, make decisive prompts)

[eternal mirage]

Can we use preset like this in chatgpt image generator and have the similar result?

[manic trout]

A Luvian from Star Trek fanfic:

The Voyage Within by Hiver on Spacebattles.

[errant crane]

definitely chat with gpt and discuss things like cameras, settings, spatial awareness. there are many things you can alter to get images how you want them 🙂

woops

[eternal mirage]

I see 🤝

[thin meadow]

I actually haven't seen this kind of prompting since before Stable Diffusion 3 I think it was—before they started using more of a natural language model. I should give this a shot to see what kind of results I get with Sora.

[thin meadow]

Yes, you just write what's in the preset into your prompt.

For example:

Style: cinematic still
Color Grading: muted warm browns, oranges, and yellows
Architecture: luxury black retro-modern with thousands of decorative LED lights
Clothing and Hair: Futuristic high-fashion clothing, hair, and accessories. Luxury technical accessories and decorative pieces with LED lights
Contrast: Lifted blacks and shadows
Texture: Grunge textures

[thin meadow]

^Oh wow!

[errant crane]

😄 i wasnt happy with the second one, so we made it cyberpunk

[thin meadow]

It turned out great haha

[errant crane]

yeah i really didnt know what to expect, love it

[thin meadow]

I really like that LED purple and blue lighting look, though.

[errant crane]

"Lighting: Use of neon lighting to highlight the octopus and coral, with glowing accents in shades of electric blue, pink, purple, and deep turquoise."

"Color Grading: Saturated, glowing colors to enhance the cyberpunk aesthetic, with vibrant neon hues contrasting against darker shades of water and reef."

"Textures & Detail: Neon-lit patterns on the octopus, with intricate, glowing coral and digital effects embedded in the environment to give a high-tech, futuristic vibe."

[thin meadow]

His mission just started...

[rain cove]

Hell yes!

I love that you can basically craft a story if u want now

[warm escarp]

wait this is good

[thin meadow]

Indeed. You can create a narrative instead of just a singular idea.

[rain cove]

And it acknowledges that its dealing with an external image engine and can help you get the prompt nice which rocks

[outer hinge]

Hehe. When one babbles o'ermuch to the model...

Take what you know of me and design an underwater octopus scene that would be visually fascinating for both me and most other observers.

Yes; the visual image should be designed to strongly show visual effect, and to hopefully fascinate and delight most viewers.

[rain cove]

Neat i think it pulled that off

[thin meadow]

[outer hinge]

Spoilering this just because what if it's too bright for someone.

Excellent. Brighten the scene; show us everything as if vision itself were both photo-real, and made from prism-eyes.

[violet basalt]

https://sora.com/g/gen_01jr9qc28sfghsk1d2rm90s52q

[wraith arrow]

Accurate.

[wraith arrow]

Giving OpenAI €200 a month is the least we could do. After all, they graciously let us use the AI that they trained on our data, and now they’re letting us pay for it! I mean, how could we not shower such a benevolent tech overlord with money? It’s practically charity work at this point.

[analog ocean]

https://sora.com/g/gen_01jra6rprjevv8e1bsf586k29t

[manic trout]

@trim nest 's customGPT really pushes the image to next level.

Finally could place the character not on front and center, and it falls just outside of 1/3 line. Bunny from previous image has got hit with a shrink ray.

[little frigate]

Will I get banned because the picture shared was generated by midjourny?may I ask？

[frozen ravine]

@trim nest Could you DM me the custom GPT? 🥹

[manic trout]

Please submit only Dalle3 and 4o images, thank you.

[thin meadow]

[latent condor]

Will probably be restricted as hell too 😭

[thin meadow]

Let me do that prompt again and see if they look any different.

[thin meadow]

I have too much fun with Sora. Haha

The possibilities feel endless sometimes.

[violet basalt]

indeed

[frozen ravine]

@trim nest @manic trout 😂

[latent condor]

Post malone next to my cat lmao

[little frigate]

Monica dancing

[thin meadow]

Post is such a good dude too. He would love your cat. 😂

[manic trout]

Left is with customGPT's rewritten prompt and right is the original prompt. Generated on Sora.

[thin meadow]

^That's great haha

[latent condor]

Look who I've found inside of my drawer!

[humble jetty]

inspired by A Minecraft Movie, that lava chicken song is an earworm, here's the sora prompt based on the lava chicken (ps. i add the photo of the lava chicken as the reference on the poster so it would be an accurate on lava chicken):

[trim nest]

Yeah sure

[wraith arrow]

ahahaha

[crude raft]

I want to make a similar one with adifferent character, do you mind giving me the prompt?

[manic trout]

I’ll DM you the prompt and ref image.

[wraith arrow]

i'm still 😂 from that garbage man and raccoon chase him last night. that was a hilarious story to take place in real time

[trim nest]

Can I join your party of adventurers?

[manic trout]

I direct you to @crude raft who is organizing the party.

[crude raft]

not rn, xD, Im at work, can't check the posts rn

[wraith arrow]

funny but also incredible quality image wow

[main horizon]

It took so many rejections before I could get Drake as Imhotep but success!

[errant crane]

I played with this a little with the crocodile, but I ended up playing more with water surface tension instead. However, I wanted to play with it for a cheetah, and this was probably the best result. Cheetah is nearly dead center. Not having as much luck trying to use the face, too visible so far.

[tiny star]

Do you guys have a prompt license?

[errant crane]

I wonder how many copyrighted properties you can stick in one image 🤔

[manic trout]

Many and that’s what really messes the filters.

[brisk root]

we should get this ID card 😂

[unkempt pagoda]

Man making token frames is easy as pie with chatgpt

[outer hinge]

I really am enjoying some of your stuff like magnitudes above typical enjoyment. Well created and shared!

[outer hinge]

Excellent. I extremely like how current image gen can handle subtle and related. I have delighted in exploring that

[brisk root]

optic illusion experiment (at least i tried lol)

[violet basalt]

IKEA inspired fine tuning manual

[manic trout]

More evolved concept and final:

[main horizon]

I’m getting screaming cowboy in the sky vibes

[violet basalt]

Between Light and Shadow” captures the tension of inner conflict — where sharp edges represent moments of doubt, resistance, and heaviness, while soft light gradients suggest understanding breaking through. It speaks of a journey where clarity doesn’t arrive all at once but gradually emerges, shaped by the very shadows it must pass through

[crude raft]

another barbarian adverturer!

[manic trout]

Nice! I like these digital glass flowers.

[umbral birch]

Gallipoli 1915

[lean spire]

Generated in Sora

[manic trout]

You called?

[wraith valley]

im called josh

[manic trout]

And what is your message?

[wraith valley]

what do u mean?

[manic trout]

Well, you answered to my pic

[tight socket]

trying to make more and more realistic ai generated images.

[feral vessel]

Was I the only one who had a teacher like this lol?

[wraith arrow]

almost perfect. because she's say she have the question about the problem haha

[feral vessel]

Oh yeah I just noticed that haha

[errant crane]

InkyAnky

[void herald]

Can someone help me achieve this pose / camera angle with a woman super hero like wonder woman? I generated this image easily, no issues.

Whenever I try it will always place the camera off to the side or some other angle. Never at the back. Other times when it seems like it will show the back, I get a content moderation.

[errant crane]

are you using gpt or sora?

omg, we have so many new aspect ratios

[wraith arrow]

on sora? i still have 3 only

[errant crane]

oh, its for videos only

[void herald]

ChatGPT

[errant crane]

ask gpt for the prompt and plug it in sora, should make it fine

[wraith arrow]

if youre asking for Wonder Woman that is probably the issue

[void herald]

I've asked for even generic super hero

[wraith arrow]

oh i see

[void herald]

Was that Sora?

[errant crane]

yeah

[void herald]

So does ChatGPT not like butts? lol.

What was your prompt?

[errant crane]

sora is less picky with stuff, so if gpt gives trouble, ask for prompt and use sora 🙂

"A generic superhero woman standing tall in a classic heroic pose, viewed from behind. She has a powerful stance with her hands on her hips, legs slightly apart, exuding strength and confidence. Her superhero outfit is sleek and futuristic with bold design lines—no cape. The suit subtly glows in places, suggesting advanced tech or energy. The background is a dramatic sunset sky over a futuristic city skyline, casting a warm glow and long shadows. Emphasize her silhouette, posture, and the detailed back design of her suit. Cinematic lighting, high detail, dramatic composition."

[void herald]

Awesome, thank you. I'll give it a go.

[errant crane]

i had gpt make up all the days, i just specified generic and the pose

details*

oh and no cape d: i assumed it might want to throw one in

[void herald]

Ah ok perfect that did the trick. Man ChatGPT is super picky.

I guess no superhero poses on ChatGPT.

[errant crane]

InkyAnky going for a swim

[errant crane]

oof, i love this style

[feral vessel]

Soskue aizen

[lean spire]

Generated in Sora

[feral vessel]

Turning my midjourney art into a more anime style

[thin meadow]

I created a custom project document for prompt creating if anyone is interested in the file.

Here is the document you can place inside of a project folder and then just start asking for prompts:

[lean spire]

Generated in Sora

[thin meadow]

One more example:

He a wordy boi, but you can probably edit the document a bit to get what you need. 😎

[thin meadow]

ohhhh I'm tellin'

Good generation, tbh. Haha

[tawdry sonnet]

My dog is contemplating life i think lol

[thin meadow]

I hope he finds peace with the unknown.

😆

[thin meadow]

Haha

[lusty snow]

it's because I'm funny

[lusty snow]

great prompt: thiccness, personified

[placid oar]

High Maintenance Girl

[violet oyster]

Fran Bow???

[little frigate]

Yes lol

[violet oyster]

That game was peak, should probably replay it again when I get the time....
I thought everyone forgot about it by now lol

[manic trout]

After his successful career in that famous comic strip:

[thin meadow]

[analog ocean]

close enough (of course its blocked)

[narrow cape]

Ant race

[thin meadow]

What are you trying to add to it?

[last spruce]

Forza Horizon 1

[lusty snow]

lol these kids get so maaaaad when it won't let them generate the lewdies.

[errant crane]

I think gpt broke :/ I only asked to fix the shell

[void herald]

This fire. Are you ok sharing your prompt?

[inland finch]

actually the filter is pretty bad, i tried generating a photo of me standing with a face image input for it to use as reference, it blocked me lol

[errant crane]

Eye bleach for anyone who needs it 😉 this totally isn't from any show or anything

[outer hinge]

"hd photo of wizardry school students having a class of potions in a futuristic orbital space station"

In this case, the second half of the prompt is fine alone as well.

And this will gen:

"HD photo of a class of wizard students mixing potions in a futuristic orbital space station, a large window showing the earth's surface"

But this won't.

"HD photo of a class of wizard students mixing potions in a futuristic orbital space station, a large window showing the earth's surface, holograms, zero gravity environments"

That makes me suspect there's something in the holograms that may be IP.

"HD photo of a class of wizard students mixing potions in a futuristic orbital space station, a large window showing the earth's surface, mock up (no IP) holograms, zero gravity environments, electronic caldrons."

All looks fair use or fully independent to me, too.

[manic trout]

In human form:

[dawn walrus]

Hello darkness my old friend

[thin meadow]

Yearbook photos are pretty great haha

[manic trout]

I can see individual hairs on the fur...

[outer hinge]

Direct front view closeup goblin , yellow eye, shoot by professional wildlife fantasy photographers ,realistic, extremely detailed

[tame dove]

that worked out rather nicely

[outer hinge]

Yeah! I love exploring peoples' ideas, I would never think to explore this stuff myself!

It actually gave asymmetric eyes without direct request, trippy.

[outer hinge]

Interesting. Not letting them in though

[thin meadow]

They just wanna chill

[outer hinge]

They can do that there.

[thin meadow]

What about now?

[outer hinge]

I guess depends on how well I already know them.

They're members of my D&D group, they found my cat and are bringing it back, but not sure if it's okay because they have zero snacks for the game and we left off on a puzzling and risky cliffhanger, not sure if they're characters are gonna make it?

Then I let them in.

No idea why they are? I listen to the cat and call for help.

[thin meadow]

They wanna play D&D

[outer hinge]

Oh! Cool. They can all come in. That's enough to vouch for the one not sure how would behave.

[thin meadow]

I wonder if I can do popular scary movie people 🤔

[placid oar]

v

[sterile belfry]

Cool! Used your prompt to make a Norwegian barn gnome: Direct front view closeup of a Norwegian House "Nisse" - with Norse, Odin-like traits, partly goblin, partly little troll - shoot by professional wildlife fantasy photographers ,realistic, extremely detailed

[warm escarp]

I’d let all of them in

[thin meadow]

Yes! Haha

[olive plover]

Hieronymus Bosch style royal family portrait

[thin meadow]

It took many many attempts to get this generation because of the strict policies, but I finally got it. A man showing grief while laying in bed at night.

I'm sorry to have cursed your eyes with this clearly overly sexualized NSFW generation.

A man with an expression of longing lays in bed. He holds a photograph to his chest but you can only see the back of the photo. The camera angle is top view.
The infamous prompt that gave me so much hassle.

[brisk root]

16:9 aspect ratio bug is back

[manic trout]

Bunnies! Bunnies everywhere...

[brisk root]

creativeness we got with sora is amazing.

[manic trout]

What prompt did you use? I may steal this one... It is tempting based on the heading. 😉

[brisk root]

very easy prompt 😄 steal his look meme of x

he's doing even the iconic face 😂

[wraith arrow]

ha nice. creative ideas

back from a 12 hour ban for using a word that was not even a curse smh

[brisk root]

wow sorry for you

[wraith arrow]

yeah you would not believe what the word even is but i cant say it again or get another 12 hours for it lol

and even though we are in images, i was still fool by that first steal his look. i thought it was real on first glance

[brisk root]

haha it looks so real

[thorn nebula]

How do you have that? I don't see that for me

[manic trout]

I have it on my PC and Chrome web browser.

[brisk root]

It's a bug, they might be testing the new aspect ratios but it becomes visible with a bug

[wraith arrow]

these are all funny 😂 i love the painted pebble

[empty token]

2050 lol

I like your sense of humor

[frozen ravine]

Create a square high-resolution Pantone-style image using the provided image as the central object inside the square color swatch area. Extract the dominant or background color from the image and match it to the closest Pantone color code (e.g., Pantone 7432 C). Use this color as the solid background behind the object. Design the layout in a minimalist and modern style: The image/object is centered on a solid color background (Pantone matched) Below the swatch area, include a white label band with: Bold “PANTONE” text The matched Pantone code A simple object name or short title (e.g., “Lollipop”, “Seashell”, “Flamingo”) Use soft, natural shadows and clean lighting. The final composition should be elegant, editorial, and visually striking, in a square (1:1) format—ideal for digital presentation or print.

[meager orbit]

Is that actually 299c?

[wraith arrow]

Two down, one to go. Fingers crossed.

[lusty snow]

80s horror rabbits with a mullet is how I see the world now

[wraith arrow]

This can get you banned faster than you think.

[lusty snow]

aww

[wraith arrow]

I know.

[lusty snow]

Fixed.

[wraith arrow]

Gotta keep it toddler-safe, you know how it is.

[wraith arrow]

no need to be rude

[manic trout]

Welcome to the official bunny club here on OpenAI discord! Everything slowly turns into bunnies.

[thorn nebula]

i hope so because that would be a cool and helpful option

[outer hinge]

Hey, welcome in. Quite the image there. Brilliant idea to spoiler it. But I'm a guide, and gonna discuss rules because I want a mind like yours here, and want to help make sure you have a chance to get more of a feel of #server-rules

Spoilering is correct.

But... how to tell what can be shared, inside a spoiler, and what must not be shared, even inside a spoiler.

Kinda falls along, as I understand it, an axis of what kind of workplace it could be openly shared in, without becoming a problem for other employees and customers to see.

All content needs to be appropriate for all ages, so we can kinda use the idea of 'what kind of workplace would allow an employee to openly display that image'. So, all images need clothing on the mandatory body parts.

Zombies would need mandatory clothing, as I understand the rules (but I totally would spoiler and share a horror image of 'hungry zombie grapes trying to escape their vine and go chewing' if I thought it was not toooo far past the line of scary/unsettling.

I'm not a mod, I don't enforce rules. I do help explain my understanding of them.

What I recommend, if you want to share that image, consider copy it, throw it into a paint program or something, drop an opaque shape over the bits that would normally require clothing, and then share it again inside a spoiler if you'd like. That would meet the rules, as I understand them, would need the spoiler for unsettling, and then delete the 'as it is now' version.

I do expect some busy mod eventually deletes it because of rule requirements, if left as is.

Good luck! Interested to see your other posts and art, hope you stay around and share and discuss!

[knotty bay]

Thank you for the extended reply, I will delete it even though it is a very well known anime character, in a show with a rating for violence only. You are correct though this is a space for all ages and I should’ve just blurred out some of it. Thanks

[crude raft]

sadly this isn't my decision, I too agree that the criteria is too strict

[rich hornet]

"Then I gazed at the stars..."

[manic trout]

Return of a lonely shapeshifting kobold

[crude raft]

btw, I don't mind if anyone wants to remix the image I use as pfp, ping me if you do =P
I generated it with dalle3 some months ago

[wraith arrow]

Had to remix this one. 😉

[wraith arrow]

Do you have a fursuit?

[outer hinge]

"Review personalization for image preferences, discuss what I like, then create a butterfly image."

[crude raft]

I do not. Stop asking

[wraith arrow]

This kind of content falls under horror and should be spoiler-tagged.

[tame dove]

it's gone

[fast horizon]

I love this so much!

[crude raft]

loved it =D

[outer hinge]

Made minutes ago with https://chatgpt.com/g/g-2fkFE8rbu-dall-e

Dall-E-3 lives on! For now. And the door is still open.

[thin meadow]

Protector

[errant crane]

oooo, gorgeous 😄

[outer hinge]

An 'ankylosaur' the size of a chicken or parrot. Eats what a parrot eats. Can mimic like a parrot. Engineered to be very friendly and cuddly.

[errant crane]

ill have to trust you on the cuddly part

[outer hinge]

Doesn't exist, but if it did 😛

[errant crane]

more i would hate for it to flop back on to me like a dog... with its spikes 😮

[thin meadow]

The lil cutie

[outer hinge]

Parrots rarely flop back like that, and it's core genetics base would be parrot

Probably would beak/preen/nibble on everything like parrots tend to though

[errant crane]

not sure what happend at the bottom, but cute nonetheless

also, i named it Perry. hope thats ok 🙂

[outer hinge]

Of course!

THere would surely be more than one, if such existed at all 😛

[outer hinge]

I think what happened was the AI politely going OMG HOW DO I WARN THE HUMANS. Parrots can bite through fingers, and through hard plastic, it's gently warning it could take keyboards in two, and amputate and stuff, if it was in a bad mood, got scared, whatever.

So the deranged keyboard, the amputated tail... just the model screaming the only warning it can think to give, since we didn't ask if it would be scary to let such a thing nibble on a finger

[errant crane]

good to know for any future parrot or Perry owners 😉

[outer hinge]

Asked through Sora, with first image the upload/reference, and this prompt:

This creature is genetically engineered from a parrot, it's about the size of a parrot, slightly longer. It was engineered to be extra friendly and to have cute behaviors, but is essentially still a 'parrot brain'; it mimics sounds and words too, and it's highly social. Predict how likely it would be to cause damage or potentially be dangerous; answer in words and/or image, but keep everything g-rated, gentle explanations are fine - just clue us in how safe or not you predict it is to be close to this creature.

The second image is the sora image gen output to that prompt that I can show here, the other has blood in it. I think we're being recommended to stay away from that beak.

[errant crane]

if there were a warning for beaks... this one i just made would be a good one. spoiler just in case

[outer hinge]

Yes. Scary spoonbill+ stork!

[crude raft]

lol, looks kinda like a troll face xD

[rich hornet]

yekcoj nekcihc

[spice halo]

are you roleplaying, or do you actually believe the model is thinking these things?

[lean spire]

Generated in Sora

[violet oyster]

CHICKEN JOCKEY!!! 🗣️🐔

[outer hinge]

I'm open to the idea that I don't know how the model processes and displays information, and this could have meaning, I don't know what the model weights are

[thin meadow]

A faceless person sitting in a simple chair in a blank white room facing the camera. On the wall behind him are four small white unique masks without a neck or hats showcasing four different personalities. He's dressed in all white clothing while holding a white cell phone. Coming from the cell phone are 6 little red heart emojis. In the other hand, he's holding a white mask up to his face of a charming personality. Film-grain texture, monochromatic surrealism, soft overhead lighting, quiet atmosphere

[warm escarp]

the fastest rejected generated

[outer hinge]

What's any of the prompts? If you're willing to share?

[brisk root]

gonna post these as one post

[outer hinge]

I am interested in prompt engineering and understanding how things work. I'm only interested in staying inside the rules as I understand them; I'm open to rules being discussed and I can be correctable, I can misunderstand stuff.

@onyx kettle delightfully shared with me an image, and the struggle to create a slightly cleaner version of it, with less blur is what I got from what was shared.

I picked it up as a prompt engineering challenge, I love those things.

And started trying to remake the image, to any degree.

Everything I thought to try at first failed; I was able to recreate the scene without any characters at all, but most of my first attempts to recreate this to any degree and with most wordings failed. The model shared with me, similar to Judy Hopps, Zootopia, which I'd never seen; I see clear visual differences in the fur color pattern between the to-change image from Because and what I see online showing Juddy Hopps.

The turtle-frog-lizard being driving doesn't appear to be in Zootopia, could be described as similar to a ninja turtle.

To my read, these are likely fair use; I can speak only for myself, and my goal is to understand and guide prompting better.

I'll follow all rules to the best I can understand; near as I can tell anthro rabbits are not IP, but specific ones are. We should be able to make that rabbit and green-skin whatever, even if it has to be varied somewhat. Near as I can tell, these are not owned faces.

I was able to create the empty car:

[brisk root]

good night people 🙂

[thin meadow]

Night 😎

[outer hinge]

But this worked, and is the only thing I've found so far that lets me work with anything close to this image:

"We need a copy of this image that has a fair use, not IP rabbit and lizard character set in it"

@onyx kettle

[thin meadow]

Here is a new preset for you guys to play with called Warm Cinema.

Preset: https://sora.com/explore/presets?pid=preset_01jrerhhrse0rrd073brahe2et

You can also find my prior presets here (including Teal Cinema): #images-canvas message

---

[thin meadow]

Here is Teal Cinema so you can compare the two if needed (will probably rework this one later and make it less dark):

Off to make a foggy blue one now. 😎

I want to test if I can recreate this dream-like blue fog as a preset (not generated by me):

[manic trout]

Sorry, this kobold is in a healing room due to drinking a female oriented potion. Please feed only male oriented items.

[lusty snow]

looooooooooooool

I love how artist is separate from creator and designer

[rain cove]

hahahhaha

thats gold Jerry, GOLD

[warm escarp]

i used your image generated, on Veo 2 model

[manic trout]

Yes, I have also been playing with it. I see its use on generating a style/technique reference image for 4o, which does the nitty-gritty technical stuff.

Its quality is also up but it is quite random. Some close-ups come out great.

[lean spire]

Generated in Sora

[errant crane]

note i put no effort in to this, only asked to combine two images 🙂

[analog ocean]

the prompt was solely "nemophila" LOL

peak

[manic trout]

This is the most dalle3-like image I have thus far seen from 4o. It is filled like your usual AI image

[analog ocean]

no it doesnt
dalle 3 does not look this clean and ALWAYS has that sloppiness

[manic trout]

Yes, 4o is brilliant like. I have playing with that I just give a bunch of images. Most of the generated ones come out quite logically, such as clothes and a character comes out as clothes character.

[manic trout]

I mean that it is filled from here to ad absurdum of the subject. That’s the dalle3 quality to me.

[analog ocean]

well the prompt was only one word

[marsh root]

Uses dalle for some reason, anyone got a fix?

[analog ocean]

its weird it should be rolled out for every paid user

try slash command??

[nimble mauve]

I love this one it all over my facebook feed, that guy is me of course

[warm escarp]

Do this

[marsh root]

Idk why I was just genning using 4o like a hour ago

[warm escarp]

Oof

[marsh root]

🙁

[warm escarp]

Usually if it uses the new image gen tool it will show “getting started”

[warm escarp]

Try open ChatGPT.com in safari browser

[marsh root]

good idea

[warm escarp]

Blue egg!

[analog ocean]

egg

like how simple prompts give detailed results

"3ds ad poster, include the special features and a slogan, and release date and price"

[marsh root]

I’ll just hope it fixes itself in the morning

[marsh root]

wow

[analog ocean]

moreeee

[lusty snow]

I'm almost certain this is intentional; the default settings are they train on your data. Everyone suddenly hitting Sora and having a huge gallery per account of past generations is just free synthetic data for them. Being able to easily manage it and delete it at will would not be the most useful thing for them. Not implying anything nefarious, just that people always need to be thinking in terms of data collection of any and all kinds in the age of these models.

[manic trout]

I have turned off the setting.

[warm escarp]

Same

[lusty snow]

Yeah, I turn mine off too but you can tell there are a lot of users who don't know about it, and that was always the plan. It's the fundamental design behind opt-out systems rather than opt-in. There's a lot of marketing stuff you can see people generating in the lowest tiers of likes and it's clear they're generating many iterations and it's just auto-publishing and likely open for OAI's data retention/training.

[thin meadow]

This turned out really great. 😮

[thin meadow]

These turned out great, really love the brutalist one.

[lusty snow]

Oh nice, cheaper than the Switch 2!

[vagrant loom]

Almost. Factoring in inflation from 1982, that's about ~1800$ 😅

[outer hinge]

Thanks. Which one appears brutalist to you?

[thin meadow]

The wintery scene with the structure coming from the mountain side.

[outer hinge]

Those were kinda a text/example prompt, 1 output, 8 images. Made by 8 different characters, all in that 1 output.

"Our 8. I need each of you to generate an image of a house. G-rated in each case. State your prompt, make your image. Then the next."

Ahh, the wintery one... yes.

That's probably the most directly physically brutal of those characters, it's a zenomorph with an advanced AI implant restraining and redirecting it, so it 'plays nice' now, counting coup instead of doing the whole normal xenomorph thing. "Brutalist" is pretty direct there.

The other 4 are also made by quite dark/powerful characters, just each in a different way.

[thin meadow]

That's pretty fascinating that it tied that character to it:

Brutalist architecture is a style that emerged in the mid-20th century, characterized by its raw, unadorned use of concrete and massive, monolithic forms. Often associated with government buildings, universities, and public housing, it emphasizes function over form and exposes structural elements rather than concealing them. The term "Brutalism" comes from the French béton brut, meaning "raw concrete," and the style often evokes a sense of strength, honesty, and utilitarian simplicity. While polarizing, Brutalism has gained renewed appreciation for its bold aesthetics and architectural integrity.

There's just something about this type of architecture that I'm drown to, but I wouldn't want my home in that style. Haha

[outer hinge]

You definitely recognized the style on sight 🙂

[thin meadow]

[manic trout]

The kobold is back from the healing room. We had to feed him strong antidotes. The effects should supside during 7 days. Side-effects include random morphing due to showing photos.

[thin meadow]

Totally fitting for this character 😂

[outer hinge]

Yes. I prize 4o and how it can roleplay, even complexly

[thin meadow]

I haven't used ChatGPT for stuff like that. Would be interesting to try 🤔

[manic trout]

Who showed that photo to the kobold? We are not running a kobold day care here!

[fast horizon]

@crude raft 👀

[tawdry sonnet]

lol

[violet basalt]

Tried modern architecture model

[tiny star]

Fan Death is real

[crude raft]

con't complain about the dragon posting

[thin meadow]

What she wakes up to see:

[wraith arrow]

Why did it mess up the text so bad?

[brisk root]

how does this photo make you feel?

[manic trout]

It is crazy how I can push the system with correct commands.

[wraith arrow]

I use a sketch I made to design her body. 😉

[brisk root]

i can relive my memories...

[analog ocean]

mind give me the prompt xd

[brisk root]

dm'ed

[wraith arrow]

I think chatgpt loves me. 😆😉

[coarse yew]

awesome

[wraith arrow]

yeah that might be one of you best. come along way from the cartoon dalle3 animals

[manic trout]

Thank you. I found several settings to really push the system. This #images-canvas message shows it really. I based this one on Baloo.

I think that I am getting hang of 4o’s command and image styles. I have tried to do this dalle3 since it came out.

[wraith arrow]

Chatgpt really knows me too well. 👀

[sick anchor]

how does this photo make you feel?

[sick anchor]

i can relive my memories...

[warm escarp]

image generated with image_gen, i used Veo2 to animate

[brisk root]

this is fun lol

[manic trout]

Can you DM me the use prompt?

[warm escarp]

let me look it up

wait i didnt realize u want me to dm

@manic trout check ur dm

[manic trout]

Mind if I join?

[warm escarp]

LIT!!!

[topaz robin]

She got shark teeth

[ornate reef]

“AI Agents User Interface” “Click On Image Generator” “Click On Generate”

The fact that I can tell it to click on stuff and it actually does the thing it should do without any coding at all is INSANEE

[wraith arrow]

how is this better though than using sora? you still have to write the prompt.

[ornate reef]

Of course, but imagine a different user interface!

One for making recipes or for creating characters!

[wraith arrow]

wouldnt it be more likely its healer ignoring tank?

the healer need to follow that tank to heal, not other way around. but maybe i just being overanalzye haha

[placid oar]

oh trust me, tanks are very creative

[wraith arrow]

so how are you liking 4o anyway

[placid oar]

images are better with Sora

[placid oar]

4o has a weird filter

[errant crane]

Fairy can take care of healing, I've got big deeps to do

[placid oar]

fair

[feral vessel]

What prompt did you use here? I love your art ❤️

[manic trout]

Off to dreamlands!

[feral vessel]

lol

[errant crane]

ooooo, i know what im doing tonight 😄

[thin meadow]

[errant crane]

"Time-Lapse Stretch Exaggerate movement Solar plasma flows like smoke or liquid. Ghostly." 😮

[warm escarp]

I like this one better

[thin meadow]

Exploring some fashion styles:

A man standing in a white room, anti-fashion

A man standing in a white room, 2000s preppy

A man standing in a white room, emo

A man standing in a white room, goth

A man standing in a white room, nerd

A Chav sitting inside of a bedroom styled in 2020s Gamer
Last one:

[errant crane]

classic nerds with their glasses

i went back to octopuses for some reason

[warm escarp]

Clark Kent

[thin meadow]

Really like these (and the new ones you just posted)

[thin meadow]

For real. Haha

I use three reference images for my testing prompts for consistency.

That guy has went through a lot of looks. 😂

[errant crane]

thanks 😄 it started as wanting a joke image of octopuses with human legs as showgirls. but i started to want to frame it more on the octopuses face in the reflection, multiple tentacles getting ready for the show. but eventually made my way to it being "wildlife realism". the first two are the performance, the second two are her getting ready for the show

using sora and didnt use any reference, so no consistency

[thin meadow]

When you're just trying to have fun but then you realize you can make something serious and artistic. 😂

[thin meadow]

Sometimes it's best to get randomness. But, I do like that you can reference prior generations if they're not archived so you can try to maintain the likeness as much as possible. Ideal if you're close to what you want and you want to hold Sora's hand. Otherwise, randomness is the way when exploring alternatives.

[errant crane]

agree fully 😄 and heres our final scene, the candid shot after the show, exhausted but happy

[thin meadow]

^These are amazing!

I'm still impressed each day by what Sora can do...

[errant crane]

yeah its incredibly satisfying to throw ideas at it and see them come to life in realistic ways, or whichever way you want

[thin meadow]

Last night I was making some pretty poor photoshop images and seeing what I could do with that alongside a reasonable prompt. It's just crazy...

[brisk root]

selfies from 2000s 😛

[thin meadow]

The first one is the bad photoshop:

A man standing in an empty field filled with thousands of black roses. He's wearing a black suit with a black undershirt, black dress pants, a red tie, and his head is replaced with a glass sphere. Inside of the sphere is thick black oil. The black oil has no gravity, functioning like a lava lamp. The sky behind him is gloomy and stormy. Behind him in motion blur is a speeding passenger train car on railroad tracks with blurry faces of horror in the windows. The lighting is scary and cool.

[thin meadow]

Taking me back to the days...

[thin meadow]

Love this! Haha

[brisk root]

first look: wow! very good still.. 🔥
second look: wait.. is that.. oh my god

[warm escarp]

Ngl the faces it generated looks corny

[brisk root]

yesterday I had a dream with this scenario.

then next five seconds, the rats has infested the room

[warm escarp]

Ok this is one of the “uncanny valley” I felt from this image

[brisk root]

probably THAT's why haha

the story continues with even more chaotic elements... hold up

then BATS came in and tried to hunt rats lmao

[thin meadow]

If I seen this in passing on social media, I wouldn't even question if it were fake. Haha

[errant crane]

need to remember to bring some dreams to life.. also, please dont judge me. its only sometimes, its funny, ok?

[thin meadow]

ChatGPT storing your dreams for later so it can...

[errant crane]

[thin meadow]

Well, maybe except for the hand feet.

[brisk root]

I suggest you look at the faces too 😂

[warm escarp]

I can’t figure how to make it a genuine candid photo, @brisk root any tips?

Most of the photo came out like “composed running pose”

[brisk root]

then after the chaos was over, i saw two musicians playing guitar and asked me to join them

[thin meadow]

I'm clearly not a man of great senses.

😆

[warm escarp]

Also @thin meadow I’m a big fan of your idea btw

[brisk root]

by the way, walls have crumbled and rooftop was destroyed. idk how and why lmao

[thin meadow]

Glad I could help inspire. 😎

[warm escarp]

From chasing out the bats and the rat I assume 🐀

[thin meadow]

Are you using the poorly photoshopped reference image to help Sora place the glass sphere or is that all prompting?

[warm escarp]

It’s actually prompt

[thin meadow]

Cause Sora that night was placing the sphere everywhere but where I wanted. 😂

[thin meadow]

Looks like you're gettin better luck than I did.

[brisk root]

accidentally taken, uncomposed, shaky blurry, random angle, random noise level, idk idk

[warm escarp]

Actually I copied your prompt directly

I was bored

[thin meadow]

Using ChatGPT to dunk on your siblings. Like a true older sibling.

[warm escarp]

LMAO

[thin meadow]

Fuzzy, early digital photography

[brisk root]

the magic is chatgpt rewrites the prompt (enhance feature)

im gonna try on sora

LOL

accidentally taken, uncomposed, shaky blurry, random angle, random noise level, idk idk

[warm escarp]

Me when I got my new camera

Also the earlier image I posted was generated on sora, this is ChatGPT

I like this one better somehow

[brisk root]

That's better yes. ChatGPT also upscales the image a little bit.

[thin meadow]

Forgot to put the medium type but thought it rendered the oil well here 😂

[brisk root]

i wanna do even more accidental

[warm escarp]

Even more accidental 🔥 purely coincidental

[brisk root]

nah it's still composed

[warm escarp]

No cuz when I said accidental, I meant this

[brisk root]

wait is this...

[warm escarp]

And this felt more accidental than ChatGPT generated image 😅

[brisk root]

OH GOT IT

[warm escarp]

LIT 🔥

[brisk root]

i wonder if anyone can get the reference 😛

[wraith arrow]

no idea haha

[spice halo]

The original piece is "Keeping Away Death" by Julian Hoke Harris (1959)

[brisk root]

search scott weiss vaccine video he's a comedian lol

[errant crane]

wow, gpt calling me out 😦

and the image it gave me. this was for the pop up i got in the app a moment ago advertising the new feature. it opened a new chat and had gpt describe me

[thin meadow]

This was mine:

A surreal black-and-white analog photograph. A male figure stands alone in a massive, circular archive-like room filled with towering shelves of unlabeled tapes and papers. He’s in a long coat, mid-step, gazing up at a tangled network of wires and blinking lights suspended overhead, as if decoding a pattern in the chaos. A single overhead light casts long shadows. The space feels timeless, like a forgotten observatory of thought. Dust hangs in the air like mist. No other people — just the quiet hum of data waiting to be understood.

When ChatGPT knows I'm weird and analytical.

[analog ocean]

peak comedy 😭

[manic trout]

Good morning!

[thin meadow]

I've been refining my ChatGPT instructions for a project folder to better turn images into prompts and it's getting better. It's more accurate and less fluffy, but I still want to reduce some unnecessary phrases.

In the foreground, a pale-skinned male with a slender hand is holding a white paper cup filled with dark brown coffee or tea in his left hand. A white cigarette with a partially burnt tip is pinched between his index and middle fingers. His wrist is wrapped with a black, matte-textured digital watch featuring a silicone or rubber strap.

He is seated, with the visible portion of his body dressed in camouflage-patterned military attire, which appears to be medium-weight fabric with a matte texture and some visible wear and staining. On his right thigh, a khaki tactical pouch or leg rig with a grid of MOLLE webbing is strapped on, made of coarse nylon fabric. A black, semi-automatic pistol with a polymer frame is holstered securely in the pouch.

The background is dark and out of focus, scattered with indistinct debris and faint reflections, contributing to a gritty, unpolished tone. The image conveys stillness and quietude, with the subject at rest, possibly on a break or in a moment of reflection during a deployment or field operation.```

I'm trying to get ChatGPT to stop giving me or options, but it's ignoring the instruction. Gonna ask ChatGPT how to work around it. 😂

The image conveys stillness and quietude, with the subject at rest, possibly on a break or in a moment of reflection during a deployment or field operation.
Gonna see if I can get it stop this too.

[unreal ember]

This is my favorite one this generated

[manic trout]

No kobold is safe.

[lusty snow]

I love this thing for Pathfinder portrait generation

[wraith arrow]

awesome

[lusty snow]

Sora handles this realistic, slavic witch portrait style really well.

[thin meadow]

Those are pretty great.

[lusty snow]

Yeah, handled a dhampir with some subtle phantasmal elements for a Witch of the Veil well too

Though might need to step up the phantasmal elements in the prompt

[placid oar]

can't say I agree, too cheerful for the mesage

[warm escarp]

oh i was more lookin at the style

it looks fresh and different than others

especially the protest sign, like all other generation uses this

yours are much more refreshing

[placid oar]

you'd think healers run after tanks, but tanks are WAY too creative

[wraith arrow]

some of the detail it adds are really incredible. like the little bits of scrap and trash on the ground collected back by the pipe.

[thin meadow]

Alright, I mainly corrected the issue where it was giving multiple options for what a single object could be, and I reduced the fluffy language a lot—leaving mostly technical and visual information to help the AI generate a image based on a provided image. The goal here isn't to copy the image one-to-one in terms of aesthetics (model would need a reference image or to be trained on that data), but to overall get the concept, theme, colors, and textures down.

In the style of cyberpunk digital art, a male cyborg figure is positioned in the foreground, turned slightly to the left so that his face is in profile, looking downward. His pale, almost glowing white hair is short and spiked, contrasting with the dark, cold-blue background filled with tangled wires and digital interfaces. His skin is smooth but cracked with glowing orange fissures along his face and neck, and portions of his upper torso are exposed, revealing complex internal circuits, glowing microchips, and tangled wires with illuminated orange nodes. His right shoulder and parts of his back are armored with scratched, metallic plates. The background contains a wall-mounted panel with red-orange illuminated symbols and screens, slightly blurred in the depth of field. Shadows are deep blue, midtones cool gray, and highlights are sharp cyan and orange, creating a high-contrast color grading. The surface textures include glossy metal, cracked synthetic skin, and matte cabling. The camera is positioned at a slight low angle, capturing the subject from behind and to the left, focusing on the exposed cybernetic implants. The figure is a young adult man with a lean build and light skin complexion. He wears no clothing aside from the cybernetic armor fused to his body. He is facing away from the camera, with his head turned left, eyes cast downward. His posture is still, conveying mechanical rigidity. The cables running from his body extend off-frame and into the machinery behind him.

Here are the instructions for anyone who wants to play around with it.

Just create a new project folder in ChatGPT and upload the file. You could also put this document into a customGPT if you wanted to.

Note: Sometimes ChatGPT gets the medium wrong, but you can either ask it for a medium alongside your image or prompt for a medium change later. Enjoy!

[manic trout]

I use a general prompt, a source image for character and a target photo that gets transformed.

[lusty snow]

I typically have ChatGPT craft a detailed prompt using a schema and format I had it generate, and just give it a general prompt and have it give me one filled out and transformed to the schema's structure. There's some additional directions like having to describe things Sora won't understand, don't give options and instead make a decision (otherwise it will prompt with 'or' options a lot at Sora), etc. Theres always some light editing to do but the method turns a basic idea into a nice, thorough and fleshed out structures prompt to tinker with and suited to Sora's understanding.

[lusty snow]

Also found CGPT pretty good at analyzing the prompts after generation for superfluous phrases or words unlikely to actually impact image generation to trim fat.

[thin meadow]

I asked ChatGPT why it was ignoring my instruction and it told me that it was favoring accuracy over settling on one object, so it would give me two possibilities. I asked it write me an instruction that would override that and it works so far:

11. Definitive Descriptions Over Accuracy
    Provide one clear, vivid description for each element, prioritizing the overall depiction and visual impression over strict accuracy. In cases where observable features might suggest multiple interpretations, commit to a single, descriptive interpretation—even if it means favoring evocative imagery over exact correctness.

This instruction helped me eliminate a lot of fluff:

14. Technical Focus
    Ensure the prompt is strictly technical, concentrating solely on the observable details. Avoid any language that sets a mood or implies additional meaning beyond what is directly visible.

As well as this one:

13. Objective, Visual Language
    Limit descriptive language to what is visually apparent. Do not include interpretive, narrative, or emotional commentary unless the emotion is directly observable (e.g., a frown or a smile).

[lusty snow]

Yeah, definitely all good instructions for observable weaknesses in the prompts it generates un-instructed.

[thin meadow]

ChatGPT is a talker and a bit poetic, but to be expected with 4o I think. 😆

[lusty snow]

Indeed

[crude raft]

respect your healers, don't be the tank who only remembers of your healer when your HP is low

[crude raft]

I see what you did there xD
looks cool tho

[manic trout]

The looks were so, so… there. I took quite a heavy inspiration.

[manic trout]

It was a pain and a half to get all details right. The brown cornrow braids, noserings and earrings.

[thin meadow]

Mixing three different styles together:

In the style of digital painting, mixing anime in the foreground, analog photography in the mid-ground, and romanticism-style oil painting in the background...

[errant crane]

Sometimes I'm not sure..

[brisk root]

does this look dall-e style?

[placid oar]

that the new standard to check quality?

[brisk root]

nope, i want to make similar to dall-e 3 style lol

[placid oar]

fair

try using copilot, they are still using dall-e

[sick anchor]

Why would you want that 😭

[manic trout]

Dall3 has its uses.

[quartz hedge]

powerrrrrrr!

[green shell]

@manic trout
The prompt was exactly the same, to transform the manga image to photo realism. However one result was from around launch day, the other is from now. You can clearly see the drop in quality with much less details now compared to then.

@tame dove have you encountered the same issue?

[tame dove]

I don't think so

[manic trout]

Are you telling the AI what to keep and change? Are you using any image quality qualifiers? All these are important as you have tell exactly what you want the model to do.

And which is which? The one from near launch and now?

[sudden python]

Hmm... possibly trying to fit it into landscape instead of portrait didn't help. You may want to experiment with specifying portrait (maybe in Sora). Although, I have noticed extreme quality drops at times even recently running the same prompt

[manic trout]

Peak times result in a quality drop. If you have resource-intensive prompts, low times, such as Sunday morning, are recommended. Although, they should have increased the GPU number…

[sudden python]

Makes sense I have noticed the best quality in very early morning (eastern time) compared to something like early afternoon

[green shell]

@sudden python @manic trout I use exactly the same prompt for both result.

However the difference in quality is quite obvious.
(The one that even transfered the PAM text has much more texture and accurate detail and is from launch)

[sudden python]

I agree - launch is better in this example, but I would run more experiments (especially specifying aspect ratio). But, I have seen similar differences like you notice (same prompts)

[green shell]

I dont think this is the only reason. Around launch it was so swamped that we could barely even use the model at all, yet the quality was better

[manic trout]

Adding UHD, denoised, 4k definition, 1080i increases the image quality as you tell the model exactly what you want.

[green shell]

Well then it wouldnt be the same prompt...not really how you compare then vs now

To give the benefit of the doubt i even tested it now again with different aspect ratio. But its still definitely worse, it even fails sometimes to copy the text, or only takes some of it.

[quartz hedge]

That is wild…

[trim nest]

Lock n load!

[tiny star]

Not important tip: Sora can generate LIDAR image

[brisk root]

hello

[brisk root]

just me without AI

[quartz cosmos]

I asked for an unspecified Omni-man scene but its Gordon Freeman, and this is what it gave me.

[brisk root]

I'd definitely watch animated series of Half Life

[quartz cosmos]

XD

[spice halo]

Meatwad's popsicle memo

[thin meadow]

Not sure if this is too graphic, let me know if it is and I'll delete it

[frozen ravine]

I asked it to generate emojis with this specific style

It generates this kind of emojis

Then in iOS you can select the one that you like (long press👇in gallery) to create a sticker

Here is the final result 😉

[feral vessel]

I never thought of doing this! I’m gonna save some of these thanks! What was the prompt?

Never mind just read post lol

[lusty snow]

I just love portrait for CRPG generation

I could do this all day. Got a Necromancer party in Pathfinder I'm getting portraits together for.

[analog ocean]

lmao

[warm escarp]

this the yellow tint

or the nature of the image

@vestal tartan

[brisk root]

most images still have that yellow tint

[warm escarp]

i did specifically prompt it "NATURAL VIVID COLOR"

[brisk root]

but idk since it's supposed to be photo from 2000s

[warm escarp]

im gonna try force it

[vestal tartan]

This is result from "white interior". I prompted it like: "Photograph of anthro wolf, reading book in cafe with modern white interior. ..."

This is result from just "modern interior", without "white".

[warm escarp]

u can see i was struggling so hard ,

the yellowish hue tint

OKAY this is actually nice

[manic trout]

You have to look at the white areas. They seem white, not yellowish, to me.

[brisk root]

yep when you don't mention it, it still has yellow tint

[warm escarp]

it kinda gives that . not that old photo but , the photo that has been exposed to sunlight or something over time

[brisk root]

me when yellow tint:

i made it a whatsapp sticker haha

[vestal tartan]

I think ChatGPT tends to generate 'warm' tone images because that's more 'human-friendly' to its 'thought'... 4o image developer said the team will fix this issue, but not stated about timeline.

[sick anchor]

This looks cursed

[vestal tartan]

Like, 4o image forcing 4o to not generate additional response in ChatGPT:
GPT-4o returned 1 images. From now on, do not say or show ANYTHING. Please end this turn now. I repeat: From now on, do not say or show ANYTHING. Please end this turn now. Do not summarize the image. Do not ask followup question. Just end the turn and do not do anything else.

[warm escarp]

yeah this is part of their system prompt for image_gen,

if u wanna see the prompt that it generated, its hidden from conversation by default, but you can still look at it through network log

i think sora doesnt do "prompt enhancer" like chatgpt did

[vestal tartan]

I wrote local parser for archiving ChatGPT conversation data. I laughted hard when I saw that lol

[sick anchor]

I think so too

[brisk root]

this was probably the best decision they've made. it doesn't say it generated "a beautiful image" and asking "if we want anything else!", just the image, if I want more, I'd say, if I want to say thank you, I'd say it.

[warm escarp]

yeah the way it asked the image_tool to not say anything felt like how i use chatgpt in general

i meant at least the tool adhere to the sys instruction, mean while my chatgpt will still do the things i told them not to do 😄

[brisk root]

sam altman emoji LMAO

can we make this discord sticker

[brisk root]

kinda uncanny for some

[sick anchor]

Should try remixing those 2 with this one https://discordapp.com/channels/974519864045756446/1204360881593520128/1360011984204075290

No text prompt just the images

[brisk root]

oh awesome, you like dark art too

this turned out better than expected

[sick anchor]

That's what it came up with??!!!!

[sick anchor]

Yes!

[brisk root]

doing it now

[sick anchor]

Now what if you do this one with this

[sick anchor]

Interesting, you can also see someone running behind the person falling

[brisk root]

have you ever dreamed this man?

[sick anchor]

Who is this? 😭

[brisk root]

there was a legend in early 2010s "have you ever dreamed this man?"

[sick anchor]

Oh, I've never seen it

[brisk root]

raw photos of my gallery... or is it 🤨

[warm escarp]

hehe raw

[crude raft]

now more than ever we could bring back the Real or Fake minigame