Hunter Tomson - 1/1/1970
Hunter Tomson - 1/1/1970Gm Warriors,
Hunter here, reporting back to you about some pipe unblocking!
As you can tell, the entire project has gone for a reset. I won't dive in too much into the internals of this, but hopefully you can start to see the new vision coming to life in the recent blogs, discord action and snippets of info released on our socials. Having said that, one key thing we've done as a team is review areas where in the past we've been blocked, slow to act or had imperfect processes.
One blocker from my side was the need for Sylent to create LORE images for each LORE Journal I posted. This wasn't to say that Sylent was blocking me, for sure, I have myself to blame for many of the weeks that went without LORE updates. However, it was an additional hurdle to have to cross. We all love Sylent's art. It's the backbone of this project. But LORE images could often take up to 2 or 3 weeks for Sylent to plan, paint, create and handover a finished piece. The final products were always exceptional. But we must remember, Sylent's time is often spread across other verticals such as asset creation on the Warriors side for things like skins, weapons and much much more.
Ultimately, it's time to take LORE art into my own hands and start to paint the world out for our readers.
Don't worry, I'm not about to pick up a paintbrush and try and make things happen. I learned long ago that I paint better with words than I do with a brush. But thankfully, in this modern age I can apply my words to an AI image generator and a whole lot of cool s*&t starts to happen.
Finding Alfa's look
Finding Alfa's look was really really hard. I knew from previous experiments that starting with the basic principles such as facial structure and features would be a key starting point. Starting small is important as it allows you to re-prompt, reseed and reiterate upon your creation. I've found this is really helpful with AI Image generating so please if you take anything way from this report let it be - think small and take small steps towards your final image.
For facial structures and features, there isn't much else I could do other than think about female characters or actors that I feel have the following qualities:
- Facial features can portray mixed emotions with subtle expression changes
Then I also took into account the following question:
- If I was a perfect AI algorithm and was going to design my own body, what would I want to look like?
My answer to that question was.
- I need to be someone people can believe in and see themselves in, therefore I must look human to an extent.
- I want to be noticed by the male gaze but also slip under the radar. I should be attractive but in a simple refined way
- I want to be admired by the female gaze, being attractive in a simple refined way also works here.
- I want to the world to support me, I need to be lovable, like a puppy, lovable childish features but still obviously a woman of age otherwise the male gaze things start to get super weird.
With the above in mind, I started prompting.
I was so excited.
I was so ready for AI Image Generation to be the future of this world I am creating and dive in and start painting the most amazing Alfa the world has ever seen!!!!!!
FUUUUUUUCK! What the fuck is this. Is this AI shizzle really the future... OMG work needs to be done.
So I watched about 2 days worth of youtube videos and finally. Finally after loads of learning and hours of re-prompting I was able to create!!!
Omg. I want to die.
At this point I closed Stable Diffusion and started work on the next LORE chapter.
Luckily, about 100 words in I got stuck, reopened this image and started to think in a positive light.
I've made progress:
- This is a human figure
- There is a coherent background
- Yes she looks like she is 80 years old and about to die but oh well, we are making progress.
With that in mind, I kept searching through YouTube. During this search I learned about a key ingredient for Image Generation... Models.
Finding the right model
Google says a Stability Diffusion model is: a subset of generative models that aim to create realistic and coherent images from scratch. These models use complex algorithms and deep learning techniques to learn patterns and features from a vast amount of training data.
But in my eyes, I'd call it guiderails that make this confused, unusable technology make sense.
I say this knowing that there are many skilled AI artists that can make dreams happen with image generators, but for me, someone who's never really done this before, I found models to be a saving light.
After 1 hour of playing around with a model, I was able to create this image:
Ok wow.
What have we got here:
- 1. Some kind of mid tier Angelina Jolie look alike but not really
- 2. Gold armour
- 3. Framed setting with background creating a frame
- 4. Shit loada butterflies (I didn't prompt for butterflies)
Point 1 was kinda important here. We've got a hot woman. The only issue is, its not Alfa and it didn't stick to the principles I mentioned earlier that I felt were important for Alfa's facial structure and composition.
But lets continue exploring this as we've made some good progress. How can I make this into Alfa?
I start prompting. I want to make this woman Silver Blonde / Ginger highlights (ideally) and also get rid of those butterflies.
OK great. Hair colour change is complete (almost) but our butterflies are everywhere, getting bigger and now turning into hair accessories FML! I also notice this model isn't photorealistic, it's more of a watercolour style detailed painting.
I tried one more prompt before binning this model and moving on to try another model. I had made some learnings here, time to find a better model that works for what I want as this model made me realise, I don't want a painting style artwork here, I want it to be more lifelike. It can still be animated / cartoon style but not hand painted. That just doesn't feel REAL enough.
I changed a tiny bit of the prompt, and then boom.
Ahhh sheeet. Calm down Alfa.
I stared at this picture for a few minutes. It spoke a lot of things to me. Mainly, this Ai Image Generation stuff is going to blow up a lot of industries but maybe I'm heading down the wrong kind of pipe unblocking task... More on that later but overall... yes! We were making progress. This wasn't Alfa. She was too hot, rough around the edges and actually, too human. Time to find a new model and keep digging.
Day 4 - Sit down to my PC with coffee and a new AI Model installed
*Sips coffee
**Writes huge prompt to include everything I ever wanted from Alfa kinda fed up with life and doubting I would ever get what I wanted
***maybe I am the problem, maybe I am too connected to this character... grumble grumble.
Generates image.
*Spills coffee
Holy shit. It's Alfa, like this is actually starting to tick boxes. Lets revisit criteria.
1. You can relate to her emotionally - yah
2. Male gaze - yah
3. Female gaze - yah
4. Lovable - Yah
But not only this, the image was also in a visual style that seemed suitable for the project. It needed a bit of de-disney-ing. But this style feels fresh, exciting and full of potential.
Lets dive deeper to see if we can find our final ALFA.
It's important we can get one image that I'm really happy with. This way, I can keep the seed phrase and continue to use it for future images we generate. This way, Alfa will always have the same look.
Main things I wanted to try to move away from
1. less disney like
2. no armour, Alfa fights with her head and that's enough armour.
Ok well this isn't out of a Disney film.
It seems that Alfa wants to wear armour. If she doesn't wear armour, she dresses like a hoe that wants extra speed when going down a water slide.
To be clear the prompt for clothing here was: 'Simple survival clothing'.
This is a coherent theme with Stable Diffusion. Pretty much at all times, it tries to undress female characters. I'm not sure if this is due to how the Algo is being trained by humans but its pretty weird. I never used sexual prompts but trust me when I say this. The above image is skimming the surface of what Stable Diffusion made Alfa do.
Alfa onlyfans coming soon. Who said NFT monetisation days were over...
Joking but the point is real. It's pretty hard to keep clothing on women right now in Stability Diffusion and even harder to change the clothes they have with precise instructions. More on this later.
Anyway, lets add a cloak and gown to help cover her up.
**goes for cigarette and a long hard think about life
For real at this point, I moved over to generating Hunter Tomson (me :D) and then also Gaspar the leader of the Human League. I knew I was onto something but I got fed up with how much nudity / how hard it was to dress Alfa appropriately. To be clear, generating those two male characters was easy. There wasn't any nudity or topless men. I wonder why? What is this world haha. Having said that, Hunter's forearms are a bit ripped. But he's been single for some time living on his own in Bali before Alfa found him so I guess he just beat meat and surfed waves all day, buff forearms are approved in this case.
Day 5 - Alfa is alive
Day five is here and I'm finally getting to grips with all these sliders. Now I understand sliders, I can take an image where Alfa is looking like I want and also clothed and then control how much Randomisation is applied to future images generated. By keeping randomisation strict, I'm guaranteed that Alfa doesn't go into slut mode.
Great progress.
After a fair bit of tinkering, this image came about.
I like the background and where the character is situated.
The background tree and snow kinda works and doesn't feel too fake
I don't like the blue eyes (Alfa has green eyes) and I also don't like the medieval armour - Sylent was super against this.
Having said the above, I felt this was almost perfect. I think the image still requires a few tweaks but I think this image in some form or another will be used in marketing material moving forward.
This is an important milestone. It means one thing.
I have found the facial structure and composition for Alfa that I'm happy to move forward with.
Now lets reapply that composition into a frame I prefer?
Ok our background is improving.
Hair needs to be slightly more wavy.
I don't like the metal collar clothing that looks super fake.
But this proves my point even more, this is Alfa, I have found her...I just need to tweak a few things.
Sadly, as previously mentioned changing clothes on characters after using a seed ID in an image is really hard... but with a bit of good luck and lots of clicking BOOM.
Alfa is alive.
Findings
- We now have a new route to create LORE images that unblocks the LORE creation / social media image creation side of the project.
- Out painting is a painstaking process, as you can see, Alfa's final image is a portrait movie poster size image. It would be nice to make this landscape, but when I try to resize the image, I just end up with a Alfa with 1 head, 4 breasts and 8 eyes. Not sure why, more work / exploration needs to be done here!
- Small precise changes are hard, changing colours of exact items, removing exact items. Although it should be easy, you can often find issues here. Best advice is to up the image generation amount to 8 so when you generate, you have more odds of creating an image with the outcome you want. Then hold onto that image with your life.
I hope you enjoyed the journey of creating Alfa.
Love to all
Ya boi Hunter T.
p.s - forgive my grammar / spellings here. I'm writing to you as a human not my in LORE character and I want to be less perfect as my inperfect human self. I hope you understand <3 BUT YES, LORE grammar and spelling will improve YAWN FML MUST TRY HARDER!