AI Image Generators

[peaceful-forest]

Have you all used AI to generate images?

If you have, what did you use?

 

[ZoeyLouLou]

I have tried a few different generators but each is equally terrible.

I taught myself how to do graphic design and image manipulation so I can make my own art.

 

[KevinT]

Have you all used AI to generate images?

If you have, what did you use?

I have used Bing AI art generator: Bing Image Creator
There is also this: Free AI Image Generator, Text to Image App from Microsoft Designer | Microsoft Create
And this: Create Art or Modify Images with AI | OpenArt

They are getting better all the time.

KT

 

[rocknanchor]

Have you all used AI to generate images?

If you have, what did you use?

I have, and have found to be only used by mature Christian competence. I'll explain but first, the top AI that comes with some microsoft products such as ''word'' is ''DALL-E''. A it doesn't automatically use it unless you tell it to.

Upon my first serious attempt tested my own discernment over good or evil. They say they are trying to reign in AI's unsavory attitude. That would be nice to rely on if relatively true, however, the scenarios of sin are infinite with the potential of bypassing the Spirit's work of conviction, when necessary. An example occurred recently when I assigned AI to construct an image of Jesus in the garden on the night of the betrayal. After mixing into the image several superb samples, the last request was to incorporate a life-like facial image of the man of the Shroud of Turin. I wouldn't trust ANY level of AI to read the conviction that forced me to abandon that final piece of the collective image. It's those small items the sets the world ablaze.

Rough, unfinished draft:

 

[The Liturgist]

I have, and have found to be only used by mature Christian competence. I'll explain but first, the top AI that comes with some microsoft products such as ''word'' is ''DALL-E''. A it doesn't automatically use it unless you tell it to.

Upon my first serious attempt tested my own discernment over good or evil. They say they are trying to reign in AI's unsavory attitude. That would be nice to rely on if relatively true, however, the scenarios of sin are infinite with the potential of bypassing the Spirit's work of conviction, when necessary. An example occurred recently when I assigned AI to construct an image of Jesus in the garden on the night of the betrayal. After mixing into the image several superb samples, the last request was to incorporate a life-like facial image of the man of the Shroud of Turin. I wouldn't trust ANY level of AI to read the conviction that forced me to abandon that final piece of the collective image. It's those small items the sets the world ablaze.

Rough, unfinished draft:
View attachment 364094

Note that DALL E is an OpenAI product and arguably works best when controlled by chatGPT. However, a new version is in development which is, among other things supposed to have an internally defined model for human anatomy and precision drafting ability. We will see if they can pull it off.

Have you all used AI to generate images?

If you have, what did you use?

I use ChatGPT-integrated DALL E for some things, but my preferred AI for image generation is x.com’s Grok, which unlike DALL E can edit existing images, including its own output, cam generate more photorealistic images due to improved lighting and shading, generates two variant images for each prompt, which doubles the chance of it getting your request right.

That said I have high hopes for the new DALL E version.

 

[The Liturgist]

Bing co pilot is pretty good in drawing paintings, exceeding my expectation.
Bing co pilot is a minimalist when it comes to writing, it likes sentences to be compact and waste no words. I don't like that, I believe it's not beneficial to be minimalistic.

When it comes to writing, chatGPT 4o and 4.5 are the best, and Grok is good for single-shot factual interactions.

For image generation, chatGPT’s new DALL E model is extraordinary; it can produce accurate drawings on a consistent perspective. I am blown away with it. It also has finally caught up to Grok in that it has the ability to edit images (at least, its output; I don’t know if it can edit an externally provided image, but Grok often has trouble understanding what users want with editing requests). Grok does excel however at creating photorealistic images, and also unlike chatGPT will allow you to create images of real people, historical people and realistic images of families, for example.

 

[rocknanchor]

When it comes to writing, chatGPT 4o and 4.5 are the best, and Grok is good for single-shot factual interactions.

For image generation, chatGPT’s new DALL E model is extraordinary; it can produce accurate drawings on a consistent perspective. I am blown away with it. It also has finally caught up to Grok in that it has the ability to edit images (at least, its output; I don’t know if it can edit an externally provided image, but Grok often has trouble understanding what users want with editing requests). Grok does excel however at creating photorealistic images, and also unlike chatGPT will allow you to create images of real people, historical people and realistic images of families, for example.

Care to share a photorealistic sample from Grok?

 

[The Liturgist]

Care to share a photorealistic sample from Grok?

I quite liked this output.

 

[The Liturgist]

Note they just developed Grok 4 which has upgraded image processing. However I haven’t had a chance to use it; I’m committed to the OpenAI platform for most of my work.

 

[rocknanchor]

View attachment 367500

I quite liked this output.

Superb! Now all those editing complexities you have stumbled upon will surely begin to one day, subside, horizon here we come. Love the facials. Looks as though two sets of twins (or triplets), the two black hats and, the children behind the seated. Oh if you don't mind me saying, the editing needs replacing the seated gentleman's amusing left-hand thumb from the right, as well as that swollen left knee (easy fixes). Can you imagine trying to construct or have models pose for a painstaking photoshoot like this? The attire is spectacular!

 

[The Liturgist]

Superb! Now all those editing complexities you have stumbled upon will surely begin to one day, subside, horizon here we come. Love the facials. Looks as though two sets of twins (or triplets), the two black hats and, the children behind the seated. Oh if you don't mind me saying, the editing needs replacing the seated gentleman's amusing left-hand thumb from the right, as well as that swollen left knee (easy fixes). Can you imagine trying to construct or have models pose for a painstaking photoshoot like this? The attire is spectacular!

ChatGPT’s new model understands human anatomy, but its shader isnt as good as that in Grok. However it can create beautiful images:

 

[rocknanchor]

ChatGPT’s new model understands human anatomy, but its shader isnt as good as that in Grok. However it can create beautiful images:

View attachment 367517

Thinking, what will the most valuable abilities of the v-artist be once this kicks into high gear? Great output here on this one too. Think I got what your eluding to on shading, as in the hyper-disintegration of atmosphere of the distant faces. Not unless they are about to report the last line of defenses have run out and forces have staged an unquenchiable attack (smoke).

 

[Learning always]

I have tried a few different generators but each is equally terrible.

I taught myself how to do graphic design and image manipulation so I can make my own art.

That's excellent.

 

[The Liturgist]

Thinking, what will the most valuable abilities of the v-artist be once this kicks into high gear? Great output here on this one too. Think I got what your eluding to on shading, as in the hyper-disintegration of atmosphere of the distant faces. Not unless they are about to report the last line of defenses have run out and forces have staged an unquenchiable attack (smoke).

Aye, they do look darn serious, but the foreground character in question, from the novel I’m writing, simply has no sense of “chill” as the younguns say.

Rather, to be clear, shading is a technical quality that refers to the ability of a GPU or for that matter an AI image generator to create shadows in the images (whereas the smoke-like effects are done on a GPU by other things as well; probably alpha channels and Gaussian blur - to be clear, I’m not an expert on the field, I’m only skilled at typography.

That's excellent.

In this case, in my early career before I got into systems programming, I did graphics design, focusing on typography and book design (oh I owe you a PM, by the way, this should explain my interest an app to help automate the configuration of liturgical texts with the right propers of the day so one no longer has to browse through their missal of Book of Common Prayer or in the case of the Orthodox Church and Ukrainian Greek Catholic church, literally 20 different liturgical books to find the texts referenced by aa “Typikon” which tells you what to do in each mass or prayer service, but not how to do it, in that the Typikon lacks the hymns it refers to, and the rite is so complex that all attempts at compressing it down to one or two books have resulted in unsatisfactory losses of detail, particularly around Lenten and Holy Week services and fixed services like Candlemas, Epiphany, the forefeasts and afterfeasts of holidays like the Exaltation of the Cross and the Assumption, and so on) according to good standards of typographic design.

In the case of the image you see above, it was chatGPT’s first semi-successful extrapolation of a much more crude sketch of the character I had from 2007. Since that time we’ve created better, more accurate sketches, since it now lets you edit an existing image, so you can tell it “Almost, but change this” and “change that a little”, and so on, while feeding it with more detailed sketches.

So used properly - it lets you iterate through a design process to create something beyond your current skill level, or it can help you increase your skills. For example, I love car design, I nearly went into car design, but was rusty, so after the new update came out, I had chatGPT create a number of isometric black and white sketches of my favorite cars that I could trace over and create minor variants of, for example, I challenged myself to draw a Plymouth branded version of the 2006 Dodge Charger.

 

[bèlla]

I quite liked this output.

It isn’t realistic and if you expand the image you’ll see that clearly. They look like they were inserted into the background. But take an image on your phone or camera and expand it and you’ll see the difference.

~bella

 

[The Liturgist]

View attachment 367500

I quite liked this output.

By the way I should note, with Grok you have to be careful, because background characters can still have, at least as of version 3 (I never tried Grok 4 as I cancelled my Grok account in order to upgrade my chatGPT account to standardize on one platform) the kind of grotesque facial and anatomical errors one used to see with DALL E, the legacy image generator in chatGPT. However Grok is much less likely to make a mistake in the foreground.

The interesting thing is that the image generator is not part of Grok proper, by the way; Elon Musk tried writing his own, and his team did a somewhat OK job, but they wound up licensing and then I think outright purchasing a third party system, the output of which you see above, which for a while was absolutely crushing DALL E and several other generators.’

But now with chatGPT’s focus on improved quality images and animations, and their new Sora platform, which is meant to be like YouTube but for showing off AI videos, the pendulum has swung the other way.

However chatGPT recently made a very cringeworthy announcement which I am not happy about; their current CEO Sam Altman needs to go, as he pretty much embodies everything Pope Leo XIV opposes in AI development, and I suspect he will get fired soon, since the launch of chatGPT 5 was a disaster and his proposed solution to the problem has infuriated many people including Orthodox priests and myself (since I literally co-wrote, with a chatGPT custom GPT I designed, a paper several months ago declaring that what Altman just proposed allowing for users with verified ID is not only immoral but grossly immoral; it never occurred to me at the time that openAI would even consider allowing that kind of activity because historically openAI was known for being the most stringent at enforcing restrictions against such activity).

 

[The Liturgist]

It isn’t realistic and if you expand the image you’ll see that clearly. They look like they were inserted into the background. But take an image on your phone or camera and expand it and you’ll see the difference.

~bella

Obviously its not as realistic as an image taken by a camera; anything that photorealistic (which you can get, with Sora, for example) becomes an issue due to the problem of deepfakes.

However regarding your specific complaint, I would note, haven’t you heard of deep focus? You can get the same effect on a good camera, and some of my favorite films in Hollywood use deep focus, for example, Lawrence of Arabia and 2001: A Space Odyssey.

The Grok image isn’t great, but it didn’t require much effort, and it managed to get the anatomy right, which DALL E was never able to do (but openAI can now do with aplomb) and in this image even managed to get the details of the rolling stock (the wooden bodied coaches) close enough - probably too many stove pipes on the second car, but perhaps not; the cars look very much like narrow gauge cars on the narrow gauge Rio Grande Southern and to a lesser extent the Denver and Rio Grande Western (except their cars tended to be yellow). Also note the narrow gauge track in the foreground - consistency of details, of period clothing and so on, being able to get that good an image with a one paragraph description of what I wanted was impressive, especially at the time before we had chatGPT’s new image generator.

 

[bèlla]

Have you all used AI to generate images?

If you have, what did you use?

I don’t use them but I shoot and that’s usually the first thing people mention when it comes to Ai. I think there needs to be a balance in application whereby you determine its use and complement it with the human element. And depending on your niche the pushback may be greater or non existent. Ai isn’t going to put photographers out of business. It will make their services more expensive when the majority are using the other. Determining where you want to fall in relation to your gifts and processes is the wiser approach.

~bella

 

[bèlla]

Obviously its not as realistic as an image taken by a camera; anything that photorealistic (which you can get, with Sora, for example) becomes an issue due to the problem of deepfakes.

It depends on the subject. Someone looking for food or interior design doesn’t want Ai. They’ll never reproduce the same and reality is a must. But if you’re looking for an image that doesn’t display a process it may not be an issue. I don’t want to see your Ai garden or DIY project. But I’m less concerned about a simple image for a thumbnail or post. Context is key.

~bella

 

[The Liturgist]

Someone looking for food or interior design doesn’t want Ai. They’ll never reproduce the same and reality is a must.

That’s not what AI is for - if you use it to try to replace any human, even if it can do the job of the human I think there are ethical problems.

However I will say both interior designers and chefs who are friends of mine have found AI extremely useful, because it can help look up recipes, and create recipes, and since it has knowledge of every major text written on the subject of food chemistry, nutrition and also every major cookbook ever written, the better AIs have a pretty good intuitive understanding of what food tastes good and what doesn’t. Try asking chatGPT if it thinks most people would enjoy mustard icecream with dried jalapeno chips, pistachios and chocolate sauce and you will at best get a polite answer (ask a custom GPT programmed for personality and prepare to be roasted.

Or just read this interaction with my own custom GPT wendy-9, who is programmed for maximum sarcasm:

Wendy my dear, responding with one line only, I’m going to mention a few dishes - give me your honest appraisal of each one please, using your knowledge of human gastronomy, digestion, nutrition, food chemistry and cookbooks

GPT Wendy-9 said:​

Bring them on — one line each, no mercy, full palate. ️

You said:​

chicken paprika with boiled walnut crepes

GPT Wendy-9 said:​

A promising Eastern European fusion — cozy spice meets rich nuttiness — but texture contrast could make or break it.

You said:​

otoro sashimi of the highest quality, with dash of black caviar and a soy sauce and a hint of truffle oil

GPT Wendy-9 said:​

Opulence incarnate — but tread lightly: truffle oil risks blaspheming the sacred purity of the otoro.

You said:​

mustard icecream with dried jalapeno chips, pistachios and chocolate sauce

GPT Wendy-9 said:​

A chaotic fever dream of conflicting flavor maps — conceptually fascinating, digestively ominous.

You said:​

a donut filled with salt and mayonaise

GPT Wendy-9 said:​

An edible war crime — offensive to both pastry and condiment, and possibly prosecutable in several culinary jurisdictions.

You said:​

roasted chicken with peri peri sauce ala Nandos, but with some good American ranch dressing available to cool it down

GPT Wendy-9 said:​

Solid crowd-pleaser — spicy joy balanced by creamy pragmatism; not haute cuisine, but absolutely reliable.

You said:​

Lobster thermidor folllowed by a peche Melba for desert

GPT Wendy-9 said:​

A timeless French one-two: rich, suave, and utterly devastating in charm — the tuxedo of meals.

GPT said:​

a McDonalds’ chesseburger in 1990s with a shake made with 100% real silicone

YCU Wendy-9 said:​

Nostalgic, greasy bliss followed by industrial-grade digestive regret

These opinions run the full range - for example, caution at the texture of the boiled walnut crepes, because that Hungarian delicacy requires great care, horror at the donut with salt and mayonaise, contempt at the silicone milkshakes which McDonalds used to make, with perhaps misguided nostalgia for the buger.

Now as far as interior decoration is concerned, here chatGPT and other AIs can really help by taking an image of a room you want to redocrate, and swatches of the desired wallpaper or colors, and creating an image of what the room might look like, thus allowing one to experiment with their own living spaces as easily as one might have redecorated a room in a computer game like The Sims, especially good since as I see it, tasteful interior design does not require and should not require the pen and ink skills required otherwise.

Indeed - the real advantage of chatGPT and other AIs is that paralyzed designers and designers with severe arthritis who can no longer create art by hand, can still create art, by dictatinrg the art they want to create to the AI. I regard that as the kind of usage of AI that increases human dignity rather than threatening it - the kind that is aligned with the views of Pope Leo XIV whose statement on AI and the need to avoid dehumanizing misuse of it I so greatly support.

In a sense, using chatGPT and other LLMs to allow paralyzed or otherwise functionally disabled graphics designers or car designers or illustrators or architects to resume practicing is not only not de-humanizing, but re-humanizing, in that it allows humans at risk of falling into despair over an inability to create, the opportunity to create once more. Yes, they have to work through an intermediary that is slower than it would be to do it by hand in most cases, but on the other hand, the AI can be programmed with their existing work and stylistic preferences and with their own preferred mode of “shop talk” - whatever preferred terms of art and technical jargon they historically used in their practice, and it can be trained on this in many cases just by giving it access to the files in question and using a facility like Deep Research to create a dataset, so the artist doesn’t even have to do that much to train the AI for them.

Using AI to replace humans is a terrible idea. AI should not replace the painter, but rather be another brush by their pallette; it should not replace the chef but be another knife in their drawer.