metadata

language:
  - en
license: creativeml-openrail-m
thumbnail: https://huggingface.co/Guizmus/SD_PoW_Collection/resolve/main/showcase.jpg
tags:
  - stable-diffusion
  - text-to-image
  - image-to-image
library_name: EveryDream
inference: false

Intro

This is a collection of models related to the "Picture of the Week" contest on Stable Diffusion discord.

I try to make a model out of all the submission for people to continue enjoy the theme after the even, and see a little of their designs in other people's creations. The token stays "PoW Style" and I balance the learning on the low side, so that it doesn't just replicate creations.

I also make smaller quality models to help make pictures for the contest itself, based on the theme.

14 novembre 2022, "The Never-Ending Loop"

Theme : The Never-Ending Loop

It is a passed-down proverb that lines represent the flow of time itself. They converge and take shape. They twist, tangle, sometimes unravel, break, and then connect again.

Without words, how are we able to accurately represent this flow of time with only lines? geometrically, intricately, asymmetricaly, seamlessly, ornately...
Think of a never-ending pattern, texture, or shape– looping on and on for what feels infinite.
Just how detailed are you able to get with your patterns?

Our prompt for this event was created by @Asukii !

"the fractal flow of time stretches towards the horizon, surreal fractal intertwined looping pathways, dramatic cinematic perspective, detailed delicate intricate ornate linework, geometric abstract masterwork digital art, quantum wavetracing, ink drawing, optical illusion"

Models

based on all the submissions to the PoW
training: 101 pictures, 9000 steps on batch 6, 1e-6 polynomial LR.
balance : a little strong on the style but it made it possible to differentiate each participants
Activation token : PoW Style
Other noticable tokens : Your Discord username, if you participated. Also Rick Roll and "fullbody shot"
CKPT
Diffusers : Guizmus/SD_PoW_Collection/141122/diffusers
Dataset

based on the suggested prompt and theme
training: 50 pictures, 1750 steps on batch 6, 1e-6 polynomial LR.
balance : correct, but the style doesn't apply to every subject
Activation token : Fractime Style
Other noticable tokens : intricate, nebula, illusion, person, road, tree, boat
CKPT
Dataset

09 novembre 2022, "Abstralities"

Theme : Abstract Realities

Glitch, warp, static, shape, flicker, break, bend, mend

Have you ever felt your reality shift out from under your feet? Our perception falters and repairs itself in the blink of an eye. Just how much do our brains influence what we perceive? How much control do we have over molding these realities?

With the introduction of AI and its rapid pace taking the world by storm, we are seeing single-handedly just how these realities can bring worlds into fruition.

Can you show us your altered reality?
Are these realities truly broken, or only bent?

Our example prompt for this event was created by @Aether !

"household objects floating in space, bedroom, furniture, home living, warped reality, cosmic horror, nightmare, retrofuturism, surrealism, abstract, illustrations by alan nasmith"

Models

Main model based on all the results from the PoW
training: 51 pictures, 3000 steps on 1e-6 polynomial LR.
balanced on the light side, add attention/weight on the activation token
Activation token : PoW Style
CKPT
Diffusers : Guizmus/SD_PoW_Collection/091122/diffusers
Dataset

based on the suggested prompt
training: 100 pictures, 7500 steps on 1e-6 polynomial LR. overtrained
Activation token : Bendstract Style
CKPT

based on the suggested prompt
training: 68 pictures, 6000 steps on 1e-6 polynomial LR. overtrained
Activation token : BendingReality Style
CKPT

based on the first few submissions
training: 24 pictures, 2400 steps on 1e-6 polynomial LR. a little too trained
Activation token : PoW Style
CKPT

License

These models are open access and available to all, with a CreativeML OpenRAIL-M license further specifying rights and usage. The CreativeML OpenRAIL License specifies:

You can't use the model to deliberately produce nor share illegal or harmful outputs or content
The authors claims no rights on the outputs you generate, you are free to use them and are accountable for their use which must not go against the provisions set in the license
You may re-distribute the weights and use the model commercially and/or as a service. If you do, please be aware you have to include the same use restrictions as the ones in the license and share a copy of the CreativeML OpenRAIL-M to all your users (please read the license entirely and carefully) Please read the full license here