indexed has turned out to be my favorite model I've worked on so far.
I've found it balances strengths expected of an Illustrious model well - and handles high resolution generation with satisfying detail. It's fun to prompt and feels steerable.
I really hope you enjoy it - I have!
Inspect the showcase images to see prompts I've had fun with.
The recommended resolution is 1168x1712 (or other variations of ~1414^2).
Setting for all images in showcase:
All images in showcase are straight from the model - no post-processing.
Steps: 30
CFG: 6
Sampler: DPM++ 3M K or Euler A
Resolution: 1168x1712
Description
Further training to better handle highres and ability to produce "traditional media" textures
FAQ
Comments (25)
After some testing, it looks like tag "dark" makes picture darker on v1b than on v1.
That's right : )
I randomly found this through Shiiro0's collection while looking for new models, and I'm so confused? This is, as far as early testing goes, one of the best models I've ever tried? It's prompt adherence is genuinely so good. So many different style adjustment tags that straight up don't work -- or are way too over-tuned -- in other checkpoints I've tried work flawlessly here. Every LoRA that I've tried works near perfect. Clothing doesn't schizo randomize over different seeds; backgrounds stay relatively consistent even with incredibly light prompting -- it just works.
And most importantly, the negative actually has impact??? I can't tell you how many models I've tried where I'll get some weird piece of clothing, background noise, furniture, etc that I'll try and adjust through a negative just to fight with it for 20 minutes and give up.
I'm probably going to stress test silly mismatched tags and bad prompts later just because I haven't had a bad output yet. This model makes me feel like I've stumbled upon some secret, dark magic that only a select few know about. Some shit that my ancestors sealed away. I can't believe it doesn't have more traction.
I'll come back in a couple of weeks and edit this for further thoughts / feelings. But as far as right now goes?
I love it. Thanks.
Such high praise! Very appreciated, thank you!
I hope to improve it even further for the eventual v2!
Happy that you are enjoying the model!
I can tell from my tests it has the potential to be the best illust model ever created... there is witchcraft going on :) if you continue improving it oh boy...
Thank you for those kind words! Really glad you like it
You can check out this map: https://staticfile-e7ff9.wasmer.app/?source=https%3A%2F%2Ffiles.catbox.moe%2Fnawcqn.json
It shows indexed v1b generated images with the prompt <artist> for the top 4096 artists on danbooru.
Loading the page can take a while so you know - catbox is slow with big files. After first load it's cached though so you can revisit it without reloading.
Here's the a smaller (top 2048) version: https://staticfile-e7ff9.wasmer.app/?source=https%3A%2F%2Ffiles.catbox.moe%2Fgum73h.json
thank you!
This is such an easy and creative cp. I tried various now to create training data for a Lora, and this one is by far the most consistent and interesting. Brilliant. Gotta try the new version now.
I actually use it at 4.5 cfg. Clean details even at lower res.
Good model
Finally getting to use this feels like I've been invited to a secret cabal.
Blood for the blood god
@purplelady
Imperial Fist intensifies
I've never been so excited to click on a link
Finally a good notification!
What is DPM++ 3M K? I know DPM++ 3M SDE but Ive never seen K and my Swarm UI doesnt have it either.. getting pretty meh results with other samplers
Just use Euler A would be my recommendation
Euler A, 30 steps, 5.0 cfg.
(K is for Kerras btw)
@epiTune thanks will try!!
@luckmiriwinds
Check this comparison https://civitai.com/images/111977048
THE KING IS BACK!!!
yup. found my fav checkpoint
I never knew one could use dpmpp 3m on an illustrious checkpoint. Perhaps that got changed at some point?
Check this comparison https://civitai.com/images/111977048
aaa v2 is so nice!!