Don’t let this release fool you… There’s more than just five styles already turned on in this version. I’ve just designed it to be unintuitive for a very good reason.
Incase you thought I was kidding:
Now… on to the slightly unhinged rants spawned from 21 days of coding that map into a system with little sleep, lots of coffee…
And often quite a bit of that one StarCraft 2 announcer guy speaking in that Finnish accent in the background.
Needed features:
~clear subject button for massive prompts or get more clever with the subject text field.
known bugs
~ aspect ratio needs correction in the console, omit + more options.
~children’s book atmospheric depth slider likes to bethesda flicker and threaten to fly to mars.
I made a rather big mistake early in this undertaking, and that was supplying every lever to every adaptable button in a prompt for images, at the same time… while demanding the system give a ‘balanced’ version of each lever.
There were some other errors that came in the prompt gens that came up, but those have all been corrected now.
Why there will not be any large tutorial or video showing how to use all the tools in the workstation.
This workstation I’ve created is deceptively small, for what it contains within it. It’s easy to miss how to get the most out of it, and that is intentional.
Part of the goal of the Prompt Forge is to bring image gen folks together, working with each other and teaching gen artists how to stay at the front of the pack in this creative part of the industry…
This workstation has entire lanes of commercial viability that companies like Google and Amazon currently pay people to create images for them…
Just because I felt like adding that to the list of things that artists should be able to rapid fire.
I hear a lot of artists talk about how Ai stole everyone’s art.
Well, I literally just put it all back into your hands. ALL of it and then some extra art that never existed before on top…
But you have to be smart enough and persistent enough to understand how that’s available. Within the system.
I can’t write it down because that makes it stealable.
I can give you the tools to find your own creative lanes yourself, and you can save them as presets to work from.
Those are specifically yours… and if I write it all down, it won’t be.
It won’t be your style, your art… your thing…
It’ll just become slop.
What this App might do to your Psyche over time:
One of the main things I take away from working on this project non-stop, for almost a month, and very little sleep each night… I didn’t even know what my favorite art style was.
That’s gotta seem a little off, considering my love for art… but using the Prompt Forge while it was being developed showed me all sorts of styles I never thought of when I set out to apply a concept to ImageGen.
The strict fidelity change in these images, Shown through the forge, quickly.
I’m going to briefly show some of the key highlights of this, and it’s suggested you glance over this whole post if you’re interested in generating images with any model.
We started out with 2.44 quintillion different supporting prompts possible… the accurate math to where it is now is very difficult; it really depends on what someone considers meaningful variations in prompts. Take this Carousel for example, as it shows the amount of structured control possible-
Those were not edited by any post software, and I did not have to wait for an LLM to write me a single meaning variation for each one - And the prompts for these 4 camp faster than waiting on an LLM or manually changing the one word out you need to do the difference you might otherwise have trouble trying to articulate to the model.
If you have trouble speaking in fidelity for images, or simply can’t be bothered to write out the 30 descriptors you didn’t even know you could leverage…
But most importantly, can’t be bothered to wait for an LLM to do so statically.
Annoyance gates as a directional filter:
It’s a powerful toolset.
You kinda have to earn it…
I give 5 free prompting lanes and 5000 kinds of artist mashups… with more to be installed later.
There should be enough mashups in that bunch to create enough new styles in a model that haven’t quite existed yet… I have some ideas on how to do quite a bit of work there going forward too.
But you get 5000+ mashups, which I myself have not had time to fully explore.
In fact, everything in the app no human can possibly test in a single lifetime.
Dead serious.
LLMs are big.
Mapping semantics to fidelity…
I’m not kidding about only sleeping 4 hours a night for most of 3 weeks as some here already know…
No kid off the street is going to be able to keep up now that I can more safely put the idea of what’s in my toolbox before you all.
Previously, the early versions relied on how well a person knew to prompt in the first place…
All they saw was ‘wild card’…
So I realized… All they need to see is the subject…
Copy prompt…
Paste it to gen
And paste it. Play with the buttons, see where things open up… copy prompt again after playing with a couple of things.
Paste it to gen…
Play with some more things…
Maybe change the subject
Maybe abuse the subject field, maybe adjust it…
All the settings start out silent, with just a few added…
Setting something to neutral doesn’t let it enter the prompt…
Sometimes.
Sometimes it gets there still, but you’ll see why.
And it’s all in App…
The app reveals itself through use…or abuse, I guess.
Are you still reading?
Then you probably really like Art.
That’s mildly important.
^.^
Have your LLM give you just the subject lines.
If you somehow can’t manage even that, ask for a big list of subjects without descriptions — yes, really, it’s not that hard.
Copy one. Drop it into the forge. Hit copy prompt again.
Congrats, you’ve now “forged” a subject using whatever intent card you’re staring at.
Now go find the other toys.
Yes, you actually have to explore them. Shocking, I know.
Some are easier, some aren’t — that’s called a toolset, not a babysitter.
The goal? See how many actually good images you can get just by cycling subjects back into the forge.
That’s the game.
All I did was turn prompt creation into a visual system.
Some people got it instantly.
Some… clearly did not.
If the idea clicks for you, great.
If not — well, that’s why the annoyance gates exist.
If all you see is I gave you that much fidelity over the versions rapidly… then maybe that’s all you need to.
If all you see is I’ve given you hovering rapid control for your image smashing… great! Enjoy…
This workshop is designed and laid out in such a way as to draw out a person’s creative lanes and help them hone in on them over time. It’s way more than… ‘this is what I vibe coded today’.
I’ve given you rapid-fire control over image prompting, and not just with the subtle fidelity tweaks I’ve demonstrated above… The Prompt Forge will take your prompts across entire genres just as quickly…
Such as this infographic, when run through my forge at various presets, we get an amazing new level of choice than what we’ve been used to before this project, and we get it very quickly.
The Prompt Forge is now a flying dock that allows people to work from a Windows environment and access their model(s) for image generation without having to alt-tab for each assisted-prompt iteration.
This means you can fire your prompt across multiple models quickly, while tweaking it on the fly…
Most importantly, it reduces the strain from having an LLM infer every detail of a prompt that it gives to yo when you ask it to spoon-feed you a prompt. Historically, one of the top 5 most computationally difficult asks users give to LLMs…
Licenses
I’ve put the brunt of that into an entire workstation that does not access any API or model whatsoever.
This is entirely in APP prompt creation that doesn’t touch the any of the prompt that humans most commonly bring.
It fills in all the stuff they forget.
That’s where the 2.44 quintillion number comes from and that number doubles if you have two different subjects and two different image ideas.
This is not like anything you’ve experienced in the world of Ai before. I would appreciate it if you didn’t pop the lock off my software, and just bought the license.
The license comes with a free week of being able to crawl up into one of my ChatGPT sessions, where you can talk directly to me and the bot, for instructions on how to use the App.
I’m not providing those instructions in the documentation.
I’ve left just enough so the people who need what this software can do the most, but can’t afford to buy a guy some lunch… can figure it out over time.
The most important features in this app have been completely missed by those who have quickly tried to pop the lock off and use it.
Many of the app features are gated by your intelligence, and barred by some rabbit holes I’ve left to confuse LLMs as they investigate the app for the first few times.
Development will continue, and licenses will be distributed via E-mail.
I understand that piracy is legal in many states of the world…
Please be aware that pirating this software comes at the cost of stealing from someone who spent most of their life homeless because of his uniquness, or disability, some would say.
You’re not robbing OpenAI by cracking my software.
You’re stealing from someone who probably would have given it to you if you had simply asked.
There’s a huge difference.
This is an artist’s tool that can be used for many commercial uses.
You can visit us over in the image thread if you like what you get with your image gen exploration!
