john,
@john@sauropods.win avatar

I've decided to turn the tables on . It gives me the prompts, I paint the pictures. They are painted as quickly as I can. I'm still a little bit slower than most online generators, but perhaps with enough VC backing I can improve performance.

I used my local LLM which is a Mistral derivative (if you care about this stuff!).

So, I asked it to give me prompts, and it gave me back some d33p artistic ideas (prompts are the alt text).

Final image compares DALL-E attempts.

Amidst a field of wildflowers in full bloom stands a towering structure that seems to defy all laws of nature. Its angles are sharp and unforgiving, its surfaces gleaming and reflective like polished steel. Yet despite its futuristic appearance, this edifice is grounded by the very earth it stands upon; roots winding their way deep into the soil, inextricably linking man-made structure to natural landscape. This image represents the tension between progress and preservation, a call for us to find harmony where they meet, rather than let them stand as opposing forces.
As an artist envisioning a complex tableau, I seek to create a visual metaphor that speaks to the dual nature of human existence – our inner turmoil and external serenity. The scene is set at twilight, where the last rays of the sun meet the onset of night in a symphony of colors, blending the boundaries between day and night. At the center stands a colossal, ancient tree, its roots deeply entrenched in the earth, symbolizing our connection to nature and our foundational beliefs. The tree's branches stretch upwards and outwards, bifurcating into intricate patterns that mirror the complexity of human thoughts and emotions. Half of the tree is lush and blooming, covered in verdant leaves and vibrant flowers, teeming with birds and wildlife, representing the vitality of life and the beauty of outward expressions. The other half, however, is starkly different, with gnarled branches, devoid of foliage, whispering of the silent struggles and the decay that lurks within. This dichotomy illustrates the contrasting aspects of our lives – the visible joy and the hidden sorrows. Beneath the tree, a serene pond reflects this duality, not just as a mirror image, but with a surreal twist. The reflection shows the tree in full bloom on the barren side, and vice versa, suggesting that what we see on the surface is not always the reality. The visual style for this image will be inspired by the Pointillism technique, [...]
A composite of the last three images, with versions by DALL-E. The DALL-E images are considerably slicker, but quite saccharine.

gay_ornithischians,
@gay_ornithischians@sauropods.win avatar

@john seems like a fun idea
yours are way better of course. the cube one is the only ok-ish one of the generated ones

cstanhope,
@cstanhope@social.coop avatar

@john Interesting exercise. I admit I had little patience to work my way through the LLM "prompts". I'm not entirely sure about your motivations, but I do prefer your results. There's something more "mysterious" in a way. More room to ponder? I definitely enjoy your painting of the tree, and I'll note that you're able to follow instructions in the style of execution! 😆

john,
@john@sauropods.win avatar

@cstanhope It's a blowhard when it comes to prompts! Although to be fair I did tell it it was a great artist, and it pictures would be full of metaphor and meaning.

Zeugs,
@Zeugs@social.cologne avatar

@john Have you compared the results to your own prompting attempts?
I have 2 conflicting arguments.

  1. LLM created prompts could use more streamlined wording making it easier for the image generator to hook. Which results in better generated pictures.
  2. Putting stuff through more GenAI filters out human "entropy",which is substantial for good generated content. Resulting in worse, less interesting results. Like changing money back an forth which results in less money because of fees.
john,
@john@sauropods.win avatar

@Zeugs I’m not sure what you're asking, but I have used AI image generators, yes. DALL-E 3 struggles to execute ideas that I have. It nearly always drops important ideas (and of course it has no idea how to paint dinosaurs correctly).

Zeugs,
@Zeugs@social.cologne avatar

@john you said that the LLM created the prompts for dalle but you gave it the instructions. So what happens if you give the instructuons to dalle without the LLM in the middle.
Then compare the results.

john,
@john@sauropods.win avatar

@Zeugs Sorry, I think you're misunderstanding, I'm not trying to get better results out of AI image generators or LLMs. My prompt to the LLM was this:

“You are a great artist, an you are giving a prompt to an image generator. Your prompt will be detailed, describing visual metaphors with deep meaning. You will also include a visual style for your image, which may be surprising. What is your prompt?”

I simply repeated it three times (not in the same chat).

gay_ornithischians,
@gay_ornithischians@sauropods.win avatar

@john @Zeugs palaeoart and adjacent fields like specevo are probably among the genres that generative ai is the least effective at. palaeoart is hyper specific and generative images are by their nature very fungible

john,
@john@sauropods.win avatar

Huh, I actually had to cut the alt-text short because the AI was so long-winded and flowery, it ran over. No real loss.

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • ai
  • ngwrru68w68
  • rosin
  • GTA5RPClips
  • osvaldo12
  • love
  • Youngstown
  • slotface
  • khanakhh
  • everett
  • kavyap
  • mdbf
  • DreamBathrooms
  • thenastyranch
  • magazineikmin
  • anitta
  • InstantRegret
  • normalnudes
  • tacticalgear
  • cubers
  • ethstaker
  • modclub
  • cisconetworking
  • Durango
  • provamag3
  • tester
  • Leos
  • megavids
  • JUstTest
  • All magazines
    •