santa hat
deerdeer nosedeer glow
Sign In

Quick Guide to "No Person" style Loras

7

Quick Guide to "No Person" style Loras

Preface

This is a QUICK AND ROUGH guide - and we'll eventually extend it. This isn't a FREE option, although there ARE software options that are free, we'll get to those when we extend the article/rewrite it later.

The requirements YES are vague, but it's a start to get you to building an avatar. You can actually build anything from an ADORABLE WAIFU (no joke you can use anime style attachments) to CENTIPEDE to BARA MALE - to just your regular every day person - to even VAMPIRES!

Quick warning: If you're into actually being artistic and you aren't PERFECT at art itself? This can get really fun.  We do this for our alters, so it's both accommodating and fun at the same time. 

Requirements

  1. Sadly this one costs SOMEWHAT, there ARE free options, but you'll need about 10-50 USD depending on what you're doing to start.

  2. PC/Mac/Linux - doesn't have to be a great computer, honestly -- SL still runs on potato machines.

  3. Second Life viewer of choice, secondlife.com has all the main info, but you CAN research third party viewers - on MAC we personally have used both Firefox and Catznip.

  4. A sense of humor, and willingness to put up with SL's outdated hardware on the server end (rumor has it, their servers were on fire once lol.)

  5. Paypal or a debit card. It's quicker and easier sometimes to use https://marketplace.secondlife.com/ to get your content than roam events. (Lag!)

Shopping Time!

  1. To get a DAMN good look, you'll want to purchase a mesh head, and a mesh body.

    1. Sadly MOST OF THE GOOD ONES are in-world, so you'll have to go surfing around for Lelutka, Catwa, Genus and a few others - those are your HEAD brands, your body brands are way too many to list but we use "LEGACY/The Mesh Project" as well as Lelutka heads.

    2. Get a SKIN for both your head and body that MATCH the head and body you're buying - don't buy a CATWA skin and expect it to work on another head.

  2. Pick hair that you like, SOME HAIR directly is built for certain heads but others aren't.

  3. If you're going for a FEMININE route, do pick some good makeup in your style - you can get facial extensions, otherwise.

  4. Men: Do get a feel for beards both Bakes on Mesh and extensions for the Lelutka huds or other heads.

  5. Get at LEAST 1-3 outfits, doesn't have to be expensive, you can get some good deals on marketplace, or during events - or running around shops getting "Group gifts" - some groups in SL have "FREEBIE HUNTING".

  6. Pick up a "FACELIGHT" if it didn't come with your avatar.

  7. If you're used to SL already, find some fun objects to add to your stuff.

  8. If you're NOT USED TO SL: Don't buy too much up front, just the basics.

I've done all that now what?

Alright, your next job is to BUILD your "NO PERSON" - I'm not going to walk you through USING Second Life, at least of which not right now (i'm about to go in and actually MAKE data for future loras.). Once you have all of your items UNPACKED, you need to do the following:

  1. Attach and TEST said items, hopefully you got demos!

Once you've done your testing, you're going to want to now decide WHAT you're going to do.

(look i know this is vague, and i don't even have tutorials up here for how to do outfits, where to find things, but just CHILL - i'll get to that another day when i'm not going in game myself lol.) 

Demos Tested NOW WHAT!

Ok you've crafted your look, you're done doing the "I HAVE GOTTEN DRESSED MOM NOW WHAT".

TIME TO TAKE YOUR PHOTOS.

There's an actual place in SL called "BACKDROP CITY". I don't have an "SL URL" at the moment because it's been weeks if not months since I went on. You'll be able to choose from MANY different styles of places. This place allows for "REZZING" (it just means you can place/drop objects).

You'll be able to use either your attached ANIMATION OVERRIDE (again i'll make a more in depth tutorial later on) - or a POSE STAND that you've pre-filled with fun animations (hopefully static lol).

Photography Mode

In TIME you'll be able to get around to buying more than the basics, but for now you're just going to use the basics. (If you know how to do more do more ).

If you've already noted most viewers have an automatic "PHOTOGRAPHY" button for snapshots. I'll go more in depth on this another day, but it's fairly straight forward.

Resolutions are in pixels, and it's up to you WHAT SIZE you want just make note what size lora you're doing (Aka 1.5, 2.1 or SDXL).

SL has a built in personal lighting feature if you're using Catznip i think has it? I don't remember what other viewers have it. You CAN PICK UP "WINDLIGHT" packs from marketplace, but if you're unsure what these do, don't worry.

Mostly, just PICK a time of day and make sure you're well lit.

You don't have to use backdrop city, you CAN find some really cool cities in freebie areas to just snap away.

Make sure you get MOST sides of your avatar, as well as good facial shots. Make sure you have at LEAST 50 shots for SD 1.5, and I warn you - you should have upwards of 200+ for SDXL, Mica's BARELY works for the data we gave it. Tori's and Aeron's didn't work, and Kitch and Criss (Matoya not the old school version) - theirs work well, Mavericks' works well too. But those that DID work on SDXL? Had 100+ shots for a dataset.

(Mav = 150+ , Matoya'iivi had 200+ , Roi had 100+, Kitch had 300+) 

Other stuff on SDXL doesn't always take that much data, but for some reason SDXL doesn't pick up well on SL the way it should.

WARNING: SDXL will automatically make it 3D when you start, but if you want it to be REALISTIC  OR ANIME you'll just dial down the lora went testing/using.  If you're smarter than me, and figured out how to combat this LET ME KNOW!

Training Time!

We don't have exact preferential settings here, but this is how we've done it:

  1. Settings: MOSTLY DEFAULT.

  2. WHEN TAGGING: Make sure you have an activation tag, and despite it being "PHOTOGRAPHY" there's a strong leaning towards using WD/Waifu (Danbooru spaghetti as ElanElan calls it, sorry that's just too cute) - While you CAN USE CLIP/Blip, it will work BETTER when you're doing it the "Works on everything". (I haven't even tried blip captioning for SL except in the instance of landscapes)

  3. Repeats 5-10 depending on the size of the data, AND depending on if you're using bmaltais, a colab notebook, Kohya-ss scripts locally OR CivitAI trainer.

  4. Epochs - fudge this around where possible for SD 1.5, Holo's notebook won't go above a certain number of steps, so you need to check your repeats and Epochs.

  5. STEPS: SDXL - good god, I'm still trying to figure this one out - SD 1.5? 2000 MINIMUM, or it will be WEAK AF. (In fact about 3k-5k is about the good - you WILL have 3d overfit, but for this style it's better to have overfit and dial it down for realism or otherwise than not at all)

  6. Size/DIM: We've defaulted back to 32/16 cause of Civit trainer and lazy. 128dim CAN be better but it's not always. This is entirely a fight level preference, XD just do what you need to do and don't argue :P (PSST: SDXL? don't go below 32dim on this one, UNLESS you know it'll work)

  7. LORA/LYCORIS? - Preference.

  8. CLIP SKIP? -- 2, i'm sorry but even for realism 2. xD

  9. TRAIN IT ON SD 1.5 - if you train it on NAI or AnyLora it'll retain MOST OF ITS 3d features and might break. YOU CAN train it on whatever you want -- i'm just saying if you want it to work on EVERYTHING train it on 1.5. (If you're doing SDXL, train it on SDXL base)

If you CAN'T GET on SL AT ALL:

  1. Flickr.

  2. We can't do it for "FREE" because time + cost of making an avatar, but we are available to make data FOR you.

  3. UH, find a different game? :P

  4. Google Images!

  5. ASK A FRIEND!

EXAMPLES:

Dataset:

Output:


Conclusion:

"DUSK WHY DIDNT TORI GET PORTED TO SDXL SHE SMEXY" - Her dataset was severely lacking. Keep in mind we've been training on our alters, and if they don't have an updated dataset since we've been on SL for like 10+ years (oh god 15 years... pls kill me now lol) - we won't have as large of a dataset.

We're NOT the be all and end all to LORA training, we're a "FUDGE IT SEE WHAT HAPPENS" - and we are extremely lazy sometimes.

When it comes to this, it is about preference.

"DUSK YOU FORGOT TO TELL ME WHERE I ATTACH MY HEAD" - ... snorts up ya CENSORED

If you want more indepth tutorials on how to work SL, we'll make a HUGE tutorial on our website in time, we didn't expect to write this before going in and making more data.

If you want a buddy to HELP you learn in world? Join our discord.

Btw, we're under Tokyo Enyo and Starlordx1983 in Second Life, and are happy to schedule in world hangouts!

ONE LAST THING!

THIS TUTORIAL CAN ALSO WORK ROUGHLY FOR IN GAME SHOTS FOR OTHER GAMES AND SOFTWARE.

We do this. with FFXIV a lot, though FFXIV is a PAIN IN THE ASS if you don't have reshade/gshade. (Aka: you REALLY don't have the same lighting control).

If you know how to use Unreal Engine, and have access to assets - this tutorial is a rough guide anyways.. i mean it's not complete - ther'es no "HEY TURN YOUR HEAD THIS WAY"

"DUSK YOU GONNA WRITE THIS BETTER LATER I CANT UNDERSTAND YOU"

Yes. but not only will we GPT translate it later on, but we'll rejig it so it works for other games and we'll ahve our website tutorial up for SL by then.

Questions? don't hesitate to ask in the comments.

EDIT:

Make SURE when you do this, to appreciate all the creators that you're buying from on SL. AI is a hot topic still, and these people just like RL fashion creators spend time creating things. While most are gonna cringe that you're using their products in AI, one of the things you can do to help this is - keep the economy going. If they have a Ko-Fi or patreon, join it, support the creators. Just like RL artists, don't leave em' high and dry just because we want toys to play with!

About & Sponsored

Want to see more? We're starting to release EXCLUSIVE Content via our patreon: https://patreon.com/earthndusk

KoFi: https://ko-fi.com/duskfallcrew/ https://ko-fi.com/earthnicity

"WE"? - We have Dissociative identity disorder, ADHD, Autism and CPTSD - "WE" as in we're a system of over 200 alters, and we're not ashamed about it. We believe that AI can break down barriers in some aspects of mental health, but we also believe that AI can hinder aspects of it.

Our website is open! Join the forums, post about your work!: https://www.end-media.org

Our photography is available for FREE via Unsplash and feel free to use it in a Lora or a Model: https://unsplash.com/@duskfallcrew

WE ARE PROUDLY SPONSORED BY: https://www.piratediffusion.com/

Join the discord! We'll be around to chat about the tutorial and other games :https://discord.gg/5t2kYxt7An

JOIN OUR DA GROUP: https://www.deviantart.com/diffusionai

JOIN OUR SUBREDDIT: https://www.reddit.com/r/earthndusk/

7