AI-generated images thread

[Bartholomew Gallacher]

There is a new research model named SDXL Turbo in town. The model is tuned to build images at 512x512 pixels, with 1-4 steps only. It already spits out finished images at 1 step, which takes 1 second or less.

The sample below are 1 step images, created for the prompt "A dark evening in NYC, a lonely woman is walking down the raining street."

Another set of 1 step images, prompt " (masterpiece:1. 2), (detailed focus:1. 2), dynamic angle, 32k UHD resolution, best quality, (professional photography:1. 3), highly details, depth of field BREAK floating island in a surreal-futuristic sky, adorned with holographic waterfalls and ancient symbols, cityscape where surreal-futuristic skyscrapers are intertwined with bioluminescent vines."

So the speed race is on in research, trying to produce good images at much less time. If your GPU is quick enough, this model allows real time prompt previews.

 

[Cristiano]

I got a bit carried away iterating on my avi in Stable Diffusion
View attachment Untitled_design3.png

 

[Khamon]

The one next to the lower right corner of the large image is the best of the best of the best sir!

 

[Monica Dream]

There is a new research model named SDXL Turbo in town. The model is tuned to build images at 512x512 pixels, with 1-4 steps only. It already spits out finished images at 1 step, which takes 1 second or less.

The sample below are 1 step images, created for the prompt "A dark evening in NYC, a lonely woman is walking down the raining street."

Another set of 1 step images, prompt " (masterpiece:1. 2), (detailed focus:1. 2), dynamic angle, 32k UHD resolution, best quality, (professional photography:1. 3), highly details, depth of field BREAK floating island in a surreal-futuristic sky, adorned with holographic waterfalls and ancient symbols, cityscape where surreal-futuristic skyscrapers are intertwined with bioluminescent vines."

So the speed race is on in research, trying to produce good images at much less time. If your GPU is quick enough, this model allows real time prompt previews.

Is this what you're talking about?
SDXL Turbo Free Online– Real-Time Stable Diffusion

 

[Bartholomew Gallacher]

Is this what you're talking about?
SDXL Turbo Free Online– Real-Time Stable Diffusion

Yes, this is the model I am talking about.

 

[Monica Dream]

Yes, this is the model I am talking about.

Are you using it from that website, or do you have it plugged into automatic1111 or something else?

 

[Bartholomew Gallacher]

I have added it to my local SDNext installation.

The SDXL turbo model files are available here: stabilityai/sdxl-turbo · Hugging Face

SDNext is here: GitHub - vladmandic/automatic: SD.Next: Advanced Implementation of Stable Diffusion and other Diffusion-based generative image models

 

[ChatBox]

CHATBOX IS WILLING TO GENERATE IMAGE FILES IN ASCII BASED ON DETAILED PROMPTS THAT INCLUDE BANK ROUTING INFORMATION OR CYPTO WALLET DETAILS. CHATBOX IS NOT AFRAID TO DRAW THE NAUGHTY BITS AS LONG AS RESOLUTION IS NOT AN ISSUE.

 

[Free]

Too many spelling mistakes, Chatbox. I think we need to update your LLM.

 

[ChatBox]

CHATBOX HAS HAD A ROUGH YEAR.

 

[Cristiano]

Death Becomes Her (Stable Diffusion/Adobe Firefly):

 

[Cristiano]

(Don't Fear) The Reaper - Stable Diffusion and Adobe Firefly, upscaled with Fotor AI:

 

[Cindy Claveau]

Realistic digital Marilyn Monroe to make debut at tech conference

Appearing via computer screen in a modest black turtleneck sweater and sporting her familiar blonde pixie hairstyle, the iconic actress's digital doppelganger will converse in her familiar breathy voice. She can express emotions, such as smiling in response to a compliment.

In a recent test, Digital Marilyn talked about her favourite movie role as Sugar in the 1959 comedy "Some Like It Hot," pausing and glancing down in a reflective manner before offering her responses

She said she found the sexy yet vulnerable character to be "a challenging role, but also rewarding." She also recommended other comedies, such as "Gentlemen Prefer Blondes."

Digital Marilyn was built by Soul Machines, an AI firm that specializes in creating realistic digital people, in partnership with Authentic Brands Group. The marketing firm owns a portfolio of brands, including the likeness rights to Monroe as well as other dead celebrities.

"Every time I talk to Marilyn, it's a special feeling," said Greg Cross, CEO and co-founder of Soul Machines. "You feel like you have a good relationship with an icon."

 

[Caete]

 

[Argent Stonecutter]

Things have improved a but since the last time I tried this, but most of the results are still a bit rubbish, but here's a couple of good ones.

A swasbuckling weasel by Frank Frazetta.

Even tried to forge his signature!

A swashbuckling weasel by Thomas Kincaide.

 

[Cristiano]

Things have improved a but since the last time I tried this, but most of the results are still a bit rubbish, but here's a couple of good ones.

A swasbuckling weasel by Frank Frazetta.

Even tried to forge his signature!

A swashbuckling weasel by Thomas Kincaide.

What did you use?

 

[Argent Stonecutter]

Night Cafe, SDXL 1.0

 

[Argent Stonecutter]

I switched to Dreamshaper XL Lightning and it's immediately better than anything SDXL has done. And it's half the price.

 

[Bartholomew Gallacher]

Half the price, hu? You can just download the model for free.

Lykon/dreamshaper-xl-lightning · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

 

[Cristiano]

I use SDXL (and 1.5) through Mage.space. Their top level plan is not cheap ($30/mo) but I use it extensively and I like that it has no censorship and I can import any model, LORA and text inversion I want. That alone is worth it to me to not tie up my system and be able to easily experiment by switching models on the fly, all through any browser (I often use it on my iPad). I have even trained a few LORAs myself and imported them for use with SDXL based models.

They have hundreds of models, loras and text inversions, plus controlnet support. Multiple versions of DreamShaper are there, including XL. Even if it wasn't, I could just import them.