Demonetized Jan. 11th, 2024 @ 18:50 EDT
Is This NightCafe Guide Becoming A Disaster?
Share your feedback on this piece and be thanked later in the official release
The Magical Introduction
Approximately one year ago, I promised several readers and writers on Medium I’d craft a guide for generating beautiful artworks in NightCafe.
This was a serious undertaking for me, so, of course, my perfectionism set in, and I never began until I felt confident it’d be everything it could ever be.
I also play sudoku this way, and beat 90-99% of scores almost every time.
But as you might know, sudoku isn’t real life!!! I mean, there are plenty of fair comparisons to make, such as “where you focus your attention will determine your direction,” or “always follow through,” but in the end, it’s just a game.
In real life, not taking action until everything’s perfect is how we lose out on the very best opportunities available to us.
Well, this week, I gave the guide another go!!
But even incomplete, this is already a long piece, so I’ll leave you to it.
I’m mainly curious to know if I’m rambling too much, or if the additional information actually helps. You might also have questions I could answer in the guide as well.
This whole thing might get scrapped again.
And then we’ll start anew!!!
Your time and energy are deeply appreciated.
Sincerely,
G
P.S. The story begins here below. 👇👇
Be forewarned, there's no more
AI art to see until the end, but I did
share a real life
drawing and notes just a little
farther
down from here.
Please enjoy, and good or bad,
I look forward to your
feedback.AI Artworks as Theft
Before we begin, I think it’s important we cover this one thing, about whether or not AI steals and regurgitates images or not. Personally, for the AI, I see it as thieving as myself having consumed much art and being inspired by it.
I think retraining AI on free images would ultimately change nothing, because it’s all human output, and we’re all capable of amazing things when given the chance.
But not everyone likes that explanation.
Especially often those who had the privilege to cultivate such skills, to find the right teachers, and to spend the time mastering their craft, and who perhaps should be grateful countless other members of society made that possible by taking care of the necessities first, probably for generations—
I myself, as a new amateur artist and a writer currently on social assistance since last March, require this level of gratitude for others daily.
But we’re focusing on whether AI art is theft, specifically.
NightCafe themselves puts it like this—
The first misconception to address is that AI steals art. Yes, an algorithm can scour every digital file in the world to source shapes, images, texture, and colours–for example, if you input ‘Puppy wearing sunglasses,’ it would need to analyse data sets to discover:
> What a puppy looks like
> How big should it be compared to the glasses
> What glasses are and how they are positioned on a face
> What textures each item has–how fur looks and moves, how sunglasses hinge
Perhaps then, it’s the fact that AI artists take credit for the machine’s work that makes it wrong?
Luckily, I don’t do that.
In fact, I’m incredibly self-conscious of how to present my AI generated artworks to the public, because I don’t want to take credit. Shoot— half the time, I don’t have a clue what the AI will make with my prompts. That’s a large part of the fun!!!
And it could make practically anything, completely without warning.
There are safeguards against recreating certain public figures, gore, child porn, and other big no-nos, but otherwise, the sky ceased being a limit long ago.
In any case, I don’t personally regard any creativity as theft.
Want me to put my money where my mouth is?
Here’s a list of works I set to no rights reserved.
Or you can comment on any of my pieces and ask I add it to the list. I might be giving away my signature too this way!!
Who knows.
Anyway, let’s move on—
Stable Diffusion
For the purpose of this guide, we’ll be focusing on the Stable Diffusion algorithm, the most commonly used algorithm for generating AI art in NightCafe studio to date. Past algorithms aren’t yet obsolete, but they are slower and more expensive, and perhaps best relegated to purely artistic uses.
If you appreciate this guide, please consider letting me know with a tip to cover experimentation costs and the time I put into it.
Either way, I hope you enjoy this guide to generating art in NightCafe.
Let’s get right into it!!
Getting Started
I’m incredibly tired of this guide hovering over me. Promised many months ago, it’s only today, on the 6th of January, 2024, that I resolve to get it out of the way— by any means necessary!!
You’ll have to make do with what I give you.
Luckily, it’s not hard at all!!!!
If you can type a short description, then NightCafe will make it easy to bring it to life.
Is it “real art”?
Is this?
TK [falling sand buckets post from Instagram]
The machine is trained on information 37 billion years in the making, from gases expanding in space to single cell life forms to rats and the first human ever privileged enough to focus on art as a calling, their students, and so forth—
As with the rise of photography, many are freaked out that this will be the end of real art.
Truthfully, there are countless talented artists still today, and for as long as we’re doing anything as a species, there’s no real reason to believe that’ll ever change.
You yourself might be interested in learning the skills, if time and other resources aren’t an issue.
It’s healthy, so I’d recommend you go ahead.
Whether we’re discussing paint, photography, a variety of rendering software, of making art on NightCafe itself, your mind will benefit from partaking in creative activity.
The more of yourself goes into it, the better.
I present you this guide, because most of us don’t have the time or energy to become full fledged artists, but today especially, there are still innumerable reasons to create.
For example, maybe you need—
- a new cover photo for your latest magazine entry, storybook, or Medium article
- a unique banner for your ad
- an interesting logo… for your sock matching business
- a stylistically matching set of photos for your THICK pamphlet on traveling to Greenland during winter months
- a fresh giftwrap design for your niece’s eleventh birthday party, though the spoiled brat is only ten
- a simple moment of silence with colour arrangements and scenes you’ve never seen before
- etc.
Whatever it is that brings you here, I hope this guide will offer everything you need to generate the art you’re looking for today. Please feel free to ask your questions in the responses, and I’ll do my best to respond in a timely manner.
How to Earn and Buy Credits
NightCafe offers five free credits per day to all its users, but more can be earned, or purchased.
Apart from credit rewards for crossing various thresholds along your AI art creation journey, the easiest way to build up your credits for free is to post your art on social media.
Presently, the only platforms in the rewards system are Instagram and TikTok.
At least for the time being, Elon Musk has made X (Twitter) useless to us.
There’s a limit of 10 social media posts per day, which awards you 30 credits to create art on the site. Other popular ways to earn credits include interacting with newbies and voting on challenge results.
Submitting to challenges is free, and another fun way to earn credits, but there are no guarantees you’ll win.
It’s always worth the shot!!
For those with the means, buying credits is the simplest option, as always, but I’d still encourage you to share your creations and engage with the community, for this may be what really makes the platform such a special place.
Credit packs are easy, but you might want to start with a Pro subscription to unlock all benefits on the site first, not to mention the credit prices are better. Especially if you buy them quarterly.
If I ever run out of credits, and money isn’t a problem, then I’ll fast-track my next Pro payment to save again.
For any questions about acquiring credits or managing your Pro account, please let me know in the responses, and I’ll do my best to get back to you soon.
Beginner Presets
NightCafe is a beginner friendly site. If you’ve never even touched or heard of an AI generator before, you can still make beautiful art, albeit with less control over the process.
Presets allow you to describe the general idea you want to see, then select a style to generate the art quickly.
New users to the site will be presented with the option to generate art with presets right away.
Let’s test it out together—
TK [generate some quick concept using three different presets]
As you can see, NightCafe presets are an effective way to make the art you want, even without AI prompting experience or knowledge of terms for art or further rendering effects. The options are quick to choose from, and adequate to most purposes you may have to create.
Indeed, you could stop here, and only use presets, and I’m sure you’ll have loads of fun making all the arts you need, except for those pesky concepts which elude even the best of us—
More on the AI’s creative limitations later.
For now, let’s move on to the next factor in giving you control over your AI artworks, with modifiers!!
The Magical Modifiers List
The modifiers list is where you can find all popular prompt terms on the platform, organized by type for your convenience. It’s also where you can save your own prompts, if that’s your style.
I’d recommend exploring this list to the fullest, researching everything and anything new to you as you peruse.
You can start with the ‘NightCafe favourites’ tab, but it might be more interesting for you to jump straight to where you’d feel most comfortable experimenting.
Personally, in the beginning, I wanted to see works in the mixed styles of various artists, as rendered with modern design tools. Most of my earliest generations involve such artist name combinations as horror comic artist Bernie Wrightson, Italian renaissance painter Agnolo Bronzino, modern fantasy artist Josephine Wall, and abstract sculptor Jean Tinguely.
My earliest chosen lighting effect was ‘ambient occlusion’, as recognized from modding TES V: Skyrim, not long before I discovered ‘volumetric lighting’, then eventually settled on ‘ray tracing’, because emulating real light particles is cool, and the technology’s never had such an easy time as this with it before.
All of these work wonderfully alongside shadow depth.
Lately, I stick to photographs, imagining cameras existed in all my chosen realities.
Lastly, if you take the time to research the full modifier’s list once, then you’ll find yourself better understanding not only what you can achieve, but what you want to achieve!!
Because sometimes we need to know something’s possible to allow ourselves to want it.
Next, we’ll discuss the use of ‘seeds’, whether you need to worry about them, why, and when.
Using Seeds
The seed number exists to add randomness to your creations. Changing it will modify the outcome of your prompt without editing the prompt at all.
There is no set length for seeds, and you may experiment with its character limitations, but NightCafe already makes it easy by offering to set a random seed number for every creation. Or you can use the same one, as many times as you need across different prompts, to repeat its influence.
I don’t worry about it too much.
Changing seeds is most useful while evolving or duplicating creations and you want to mix it up a little.
Keeping the same seed while evolving can help perfect a creation that was already close.
Seeds function differently with other algorithms, but with Stable Diffusion, they are quite this straightforward.
Same seed = similar results.
Different seed = unpredictable results.
Moving on!!!
Duplication vs Evolution
Duplicating is when you regenerate a concept from scratch, using the identical prompt, algorithm and seed information as the original creation. Evolving is similar, except it employs the original creation as a start image as well.
I highly recommend this practice with your own creations, but NightCafe also allows you to duplicate and evolve the works of other users on the platform, granted they left their creation settings unlocked.
Thirty-seven billion years it took us to develop this superpower, and there are still people gatekeeping their techniques.
It’s silly.
Anyway, here’s an exercise for you!!
Here’s a creation of mine.
I want you to duplicate or evolve it to your liking. This can mean tweaking part of the prompt, the full prompt, the seed, or any other setting you want to explore.
Using Start Images
A start image provides a reference for your final piece.
NightCafe has just recently updated start image capabilities with ‘ControlNets’, which give users the ability to further influence their final image by focusing on specific aspects of the start image to draw from, such as character pose, image edges, and depth.
If there’s any feature I’ve underused in my two years generating AI artworks, this would be the one.
You always want to be mindful when choosing start images that you aren’t stealing from anyone. Using stolen images is not allowed by the platform rules, and doing so may result in a permanent ban from the website, if not worse.
Read The NightCafe Terms of Service.This is still a young industry, and I wouldn’t be surprised to see less leniency on this matter over time.
A vital feature to understand about start images, is the noise weight.
Increasing or decreasing the noise weight on your start image will determine how much the new creation will differ from the original, based on the ControlNet you selected.
A lower noise weight often pairs well with a higher prompt weight, but we’ll get more into that later in this guide.
The Advanced Editor — From Default Settings to Personal Setup and Making Personalized Art in Under a Minute
If you’ve enjoyed the NightCafe experience so far, then let me tell you, it only gets better from here. The advanced editor is where you may take full control over your creations, and it’s customizable for the settings you care about the most!!
Just click on ‘All Settings’, by the bottom of the settings column, underneath the ‘Sampling Method’, and you can choose precisely which settings to pin for easy access.
I have them all selected, just in case, but you might choose to omit any of them for a cleaner interface tailored only to you.
Having completed this step, you should feel much more comfortable using the platform, and you should be ready to move on to more refined aspects of the art generation process.
Exploring Multiple Prompts and Weights
In the days of the ‘Coherent’ algorithm, expert NightCafe users would stack numerous prompts with expertly divided prompt weights to generate masterpieces the majority of users on the platform could hardly begin to comprehend.
Back then, users would spend the bulk of their time engineering the prompt, which made the twenty minute generation process all the more satisfying to watch.
With today’s stable diffusion, generating artworks takes only seconds, and it’s recommended not to use more than one prompt at a time.
Multiple prompts are still possible, but the results are harder to predict.
Instead, stable users can now add weights directly within their primary prompt. Negative prompt weights are also possible, for designating what to avoid in your image, but the in-line weight syntax still requires positive weights within the negative prompt to work.
There are also the ‘Overall Prompt Weight’ and ‘Refiner Weight’ settings, nearer the end of the creation steps.
Raising the overall prompt weight is often wise when evolving a piece with low noise weight gain, as it tells the AI to focus more acutely with the few changes it applies to the final piece.
Lowering the overall prompt weight can render your results more unpredictable, if you prefer that.
Refiner weight, which is meant to add vividness and detail to your creations, is picky, depending on the training model you’re using.
TK [experiment with low vs. high settings with default model]
Whatever you do here, a single prompt is usually enough, and you should always start there before adding on. Opening with multiple prompts can otherwise confuse you when determining which elements were caused by what modifiers, making it difficult to adjust as needed.
Start small, and go from there, always!!
Choosing a Model Based on Your Stylistic Needs
While community models — ‘loRAs’ — might be considered cheating by some and are still banned from official NightCafe Challenges, you might find one is exactly what you need to achieve a specific effect that would otherwise be too wordy on top of the concept you wish to portray.
This is essentially how I used artist names in the beginning, referring to their style where a full description might overwhelm the AI and result in a piece less specific to the concept you aimed to achieve.
You can browse through the models, or search them by keywords, or you may explore the tags.
The sorting feature is also packed with cool ways to order the results, from model creation time, to several engagement metrics, as number of likes, comments, tips, and more.
Mixing and matching models is perfectly allowed.
When using additional models, you must add them to the prompt. NightCafe makes this easy with a clickable button that will paste the label directly into your prompt at the cursor location.
Model weights may also be adjusted, with NightCafe suggesting to keep it between 0 and 1.3.
How to Train Your Own Model as a Pro User
Training your own model in NightCafe is simple and fun, but you really want to get it right the first time.
On starting, you’ll get to pick up to 120 images to train your model, and these should all share something common. High quality images that focus on a specific style or theme will perform best.
After this, the AI will do the rest.
The whole process should take less than half an hour.
When you return, there’ll be an option to publish your model for others to use. Many users choose to be notified when anyone uses their algorithm, but you don’t have to. The more popular you get, the less manageable it becomes anyway.
I trained this one on a selection of my previous generations.
You’re free to try it out!!!
These baseline images are already an interesting; some of them are even ready to be used on their own.
Share your own model in the responses, and I’ll try it out myself.
How about it!?
The Truth About Sampling Methods — According to Me, with Examples
Everyone who looks at the list of sampling methods probably thinks the exact same thing— “why!?” You’d think all the Stable Diffusion models would be enough, but here are eight more options for you to play with!!
If you have that kind of time.
Honestly, most users will never touch these, because we don’t understand them.
No information is given about what makes them different, other than to understand which came first, and from looking at my test results, that doesn’t really explain anything.
Here’s an example of what I mean, by comparing creations made from the same prompt and seed, but using a different sampling method—
As you can see, at least with this concept, every option has resulted in a beautiful image which equally fails the assignment.
Working With Tricky Concepts
Some ideas you’ll believe should be a no-brainer for the AI to figure out, but for unknown reasons, it won’t.
I run into these commonly— especially where my friend Charlotte gets involved!!
We’ve failed to generate—
- a man’s face with a big toe for a nose
- a face covered in ears
- TK
- etc.
You might be tempted in such cases to go at it until you run out of credits, as I’ve done. Please realize, if you’re not careful, this mission could eat you up and spit you out.
Be mindful when you’re tiring and your prompts get lazy and repetitive. At this point, you might just be wasting credits.
Organizational Tagging Tips + the Difference with Collections
Tags and collections are two different ways of organizing and displaying your artworks each with unique reasons for doing so.
Above all, tags make your creations easier to search, and it affects who is more likely to come across them in their feed. Collections are more refined, and feature a description of what’s included in the bunch.
As a Pro user, you can also pin a collection to your profile for visitors, if that’s your style.
I’ve tagged all my creations since July, and will never go back.
My collections are in the works.
On The Social Aspects of NightCafe — Comments, Evolution, Messages, Challenges, the Notification System, etc.
What makes NightCafe special, apart from the spectacular AI image generator and all the organizational features, is the social aspect. I adore the vibrant community of people supporting each other in doing something they all love.
In fact,
Link to An Updated List of Medium Writers Making NightCafe Art
Here is a list of Medium writers I know are making art on NightCafe, with their profiles in case you’d like to check them out.
Not on the list, but should be? Considering I can’t find most of you at this time— probably, yeah!!
Let me know your NC handle in the responses and I’ll add you.
Turns out most NC artists on Medium
I can find don't link to their profiles.
More work is to be done.
I could keep adding on, but does anyone
really WANT a forty minute guide to
make art in less than a few seconds?
Let me know in the responses!!
Peace!
GAbout the Author:
🔍 ㅤGustave Deresse Is a Canadian Writer, Editor, & Musical Artist Exploring Themes as Spirituality, Logic, Love, Life, Technology, Philosophy, Nature, Art, Neurocognitive Psychology, Creativity, Writing, Humour, Inspiration, Music, Well-being — and the Weird.
P.S. Subscribe by email to catch my original works!! ✨ All it costs is your heart and soul.
My Introduction Page —
One or two more to satisfy —
— by Séphora Bemba
— another by Gustave Deresse; Writer & AI Artist
I appreciate you,
take care.
Sincerely,
— G
