Project: Soulstealer
July 2023 Digest / #3
Images created / total: 19 / 61
details:
* Academic Paintings made / total: 9 / 33
* Souls accumulated / total: 1 / 6
* Patchwёrk Souls stitched / total: 9 / 24
---
* Model: Fluffusion: 13 / 55
* Model: YiffyMix: 5 / 5
* Model: FluffyRock: 1 / 1
Views gathered / total: 12498 / 20187
Favorites received / total: 710 / 1335
details:
* Most beloved image of the month: Patchwёrk Soul: Merchant Cat, with 74 views and 12 favorites. Every 6.1 persons liked this image!
* Top 5 beloved images at all times:
1. Patchwёrk Soul: Delphox/Eevee Mix - 124 views, 26 favorites. every 4.8th faved!
2. Patchwёrk Soul: Merchant Cat - 74 views, 12 favorites. every 6.1th faved!
3. Academic Painting: Anthro Otter - 189 views, 28 favorites. every 6.75th faved!
4. Patchwёrk Soul: Goodra/Salazzle Mix - 100 views, 14 favorites. every 7.14th faved!
5. Academic Painting: Avali - 122 views, 17 favorites. every 7.17th faved!
Watchers joined / total: 170 / 239
details:
* Bystanders: 156
* Neuroartists: 4
* Artists 69- level: 10
* Artists 70+ level: 0
Academic Requests received this month / total: 6 / 45
details:
* Fullfilled requests this month / total: 0 / 18
* Unfullfilled requests this month: 10
* Vetoed (indefenitely postponed) requests this month / total: 1 / 5
--- --- --- --- --- --- --- ---
Topic: Experiment of Getting Into Popular Section On Demand
Greatest success, I would say. It took literally 2 days to conquer enough minds on single picture into snowballing, and to do it for 4 images in a row. All you need is a realistic AI model and a couple of nude creatures. Just one day in Popular yielded me up to 10% from my monthly views/favorites, winch is pretty huge. I've usually get 320 favorites PER MONTH, but when you are Popular, you get 40 PER DAY. And that was just after 2 realistic pictures. Imagine what would be if it would be, for example, diapers-themed picture or cub-related one... But it is almost saddening on how easy those sections were conquered. Like, my images were neighbooring images of artists with 15 years, 20 years of experience... Just imagine that literally nobody with 2 months of experience topples your entire empire with 6000+ followers you amassed through years in a 2 days. And now imagine that YOU can be this 'nobody' with absolute ease. I haven't did something extreme or even hard - I just wanted to see more of X, so I asked AI to make X with fluffy fur and neck fluff, and posted it without much preparation. Welcome 'a lot of views'. I am on the same page as Kitsuneyoukai, Rikifox, Kuroodod, Darkmirage and so on. Among titans of art, a simple newborn Neuroartist is located. Me. You. Anyone, actually.
But there is a specific thing. If you want to get into popular with seemingly RANDOM items (not the realistic ones that everyone on this site loves), then you would need a 'starter base' of ~800 Watchers.
Topic: On the verge of popularity...
If you counted my previous digest, you may find amusing details. In first month (May), I got 33 watchers. In second month (June), I've got another 36, effectively doubling my count to 69. In this month (July), I have TRIPLED my Watchers count, to 207!!! Or, if you compare 1st month with 3rd - then SEXTUPLED. This is a lot of Watchers, god fluffin dammit! :D
Of course, I, again, somehow doubled my Favorites count (winch by themselves is a 'mini-popular', especially if they tend to stay for 2-3 weeks and doesn't change much and not hidden from view). Of course, doubling the view count was pretty obvious at this stage.
Topic: New Models And The Watchers
First, I'm somewhat disappointed. This website should be full of furry adorators. Like, the actual animals. Furries are kind of meant to be closer to ones, to copy behavior, parts of anatomy, and so on. But instead of real furries, I kind of see just human porn lovers.
After comparing stats on my submissions, I found out that most popular images do resemble humans, and are mostly anthropomorphic, especially ones that I've made with YiffyMix v3.1 model (they are the most viewed and most favorited). I expected for furries to love pictures of Zoroark and Otter; but instead, much more views directed at Lopunny and Raccoon... making me think, that most of visitors of this website are just looking for humans, but with colored fur over their body. Like Neanderthals, or Apes. And while everyone is free to choose whatever they like; I feel very disappointed that Furries chose damned humans. Why then say that they love animals, if they don't pick them? I dunno.
I tried two new models this month. In addition to a usual Fluffusion Revision 1 Epoch 20; I also used YiffyMix v3.1 and FluffyRock MultiRes Lion Low-LR Epoch 87 with ZeroSNR and V-Prediction. First creates more realistic images, and if you are using 'dfc ubbp updn bwu' negative embeddings, it will actually shift your model towards anthropomorphic forms rather than feral ones. Second is... interesting, to say the least. VPred and SNR is very interesting additions to usual technology stack, and they do increase details a lot. But there is a subtle cost of making you use much more tags than usual, as those two techs take your requests quite literally, and 'if it wasn't named then it shouldn't exist', they literally erase background if you haven't specified one. I do look forward to see something like 3M models with all those nifty things attached.
Topic: Academic Paintings (ID 0000)
Well, while I did some, they were not as informative as usual. I mostly experimented with other models and with Popular Section. Kind of sorry for that. I'll try to get it better next month, perhaps will actually use ControlNets to actually pose some characters together, and teach more about VPred/SNR. I'll also try to teach on how to do negative embeddings. I suppose people do have right and freedom to see how it is actually done.
Topic: Plans on August
* To finally return to stealing souls. There is a lot of artists that should be known, and I can't wait till they all paint green. So August is a named IDs month!
* Maybe, occassionally, create couple of guides on ControlNet. Can try, at least. We do have requests that require them.
* I honestly want more Watchers. Watchers mean fast rise to Popular section, and you can't spread beautiful owlbeasts with Mature rating without Watchers. All hail cuteness! Maybe near the end of a month I'll create more realistic pictures. With dragons, with disney characters, and so on.
--- --- --- --- --- --- --- ---
Academic Requester: Current state.
Vetoed Requests this month
* One with Canine/Feline, and one of them is acting like a toilet.
Hey, whoever wrote this. I can understand your request if you are either hamster, rabbit or a god damned panda. But you are not either of those. Therefore, you are in desperate need to see the psychiatrist. Just believe me, go and tell him about your problems, and he will help you. But don't approach me. I am no shrink.
Currently In Queue Requests
* One that asked for a Male Canine, Hiding Behing Rock.
Hm... another month that I've missed this request. I haven't forgot about it, but it requires more advanced techniques that I currently teach. Perhaps, next month.
* One that asked for a Rocket Raccoon on spaceship bridge.
Yeah, I still haven't forgotten about that. Sorry, I did got carried away with new models.
* One that asked for a Salazzle.
Well, I might as well turn it into YiffyMix v3.1 tutorial. Just you wait.
* One that asked for Lucario lying on back on huge beach ball.
Still same problem as Male Canine, actually. AI doesn't want to make foreground object of this size. But it is fixable, just requires time.
* One that asked for Feral Floatzel-Typhlosion Hybrid
Still can't make this hybrid look GOOD. It makes just shiny (brown) typhlosions, without any other things from Floatzel. But there is a light in the dark - a V-Prediction. Maybe it will help next month.
* One that asked for Feral Female Light Fury/Sisu Hybrid
Don't know what you expect from that, stranger, but I'll try... one day. Dead honest.
* One that asked Lucario in Zygarde armor
Idea is the best, but AI lacks memory for that. I'll try some things, but can't guarantee it in nearest future.
* One that asked for AuroraTabi
I never met a request for OC. However, when I started to create it, I've met an AI restriction. In short words - you can't mix Ideals. In longer explanation - if you ask AI for a 'red leopard', then someone in the furry world drawn at least one red leopard, and at least one orange one, and at least one yellow, etc. So network knows that 'leopard' does have colorcoding ability and can color it's fur differently than usual. BUT when asked for a SNOW LEOPARD (like you did), then every single artist drew it as gray/white colors, and there is no other versions of it. So if you add a blue/black/red/yellow fur to it, then it stops being 'snow leopard' in the eyes of the network, and therefore, AI will make it more like Snow Leopard again, turning it white again. There it some possible solutions, but it would take more time.
* One that asked for Legiana (Monster Hunter), female and Lugia, female
While both are possible, it is not easy for a couple of reasons. First is your specification of how they must interact with each other. AI, especially one trained on tags, cannot easely comprehend what is asked of it, as it is pretty much lobotomized at this point, and there is no exact specifications - so only things like Inpainting are left to experiment with. Second is... however funny it sounds, AI simply don't know what 'legiana' is. This concept is very alien for it cuz artists never drew legiana in same poses/places, and therefore AI considers this MH creature as a mess of scales. I hope it will be fixed in future revisions or I will have more time to train our own LoRA.
* One that asked for Feral Female Absol
Well, while basic Absol by itself is very drawable, what you asked is a different story. You see, when you ask for 'teats', you mostly get ones from Equines. Because equine ones share same tag with Feline/Canine/others. So you can't simply create a tutorial on 'add a word and you are golden' - this theme is pretty advanced with a good amount of inpainting. It will come in future, that I could say for sure - just, not today. Stay tuned, tho. 
