Sudachi features a clean anime style with thin to thick outlines.
My Discord for everything related to anime models and art.
If you'd like to support my work, you can do so at: SubscribeStar, every bit helps and is truly appreciated!
____________________________________________________________
Supporters:
Thanks to my supporters tamashiicolle, pttcot, SETI and Kodokuna
____________________________________________________________
Using the model:
The model was trained primarily on images with minimal disruptive elements, such as floating text, logos, speech bubbles, or signatures. If any of these elements appear in an image, please include "signature" in your negative prompt.
Danbooru Metadata and franchise tags are not used in training, so please avoid using them in prompts. Metadata such as "highres" and franchise tags like "re:zero kara hajimeru isekai seikatsu" are not used in training.
A CFG scale between 3 and 7 is recommended. For best results, use CFG 6 with the Euler a sampler.
Refer to the example images, to understand, how the model should be used.
The model is designed to be simple and straightforward to prompt. Use the following tag order:
Number of characters
Character Tags
All remaining tags
Example: 1girl, rem \(re:zero\), standing, masterpiece, upper body
Here are some resolution options for SDXL:
1536 x 640
1344 x 768
1216 x 832
1152 x 896
1024 x 1024
896 x 1152
832 x 1216 (most recommended)
768 x 1344
640 x 1536
Lora use:
NoobAIs Loras working for the most part better in compare to Illustrious loras
License info:
Model use the Fair AI Public License 1.0-SD license.
Description
A update to my sudachi model with more data out of old sudachi SD1.5.
It is less flat in style in compare to V1 and resembles much more old sudachi.
It have some problems with hands unfortunately.
FAQ
Comments (7)
"Metadata and franchise tags are excluded—please do not use them. Tags such as "highres" and franchise-related tags like "re:zero kara hajimeru isekai seikatsu" are not used in training
(...)
Example: 1girl, rem \(re:zero\), standing, masterpiece, upper body"
Hmmm... So... Which is it? Metatags or no metatags?
masterpiece is a extra added quality tag and not a danbooru metatag
This is becoming borderline absurd. You are literally gaslighting your users at this point.
The Metadata Lie: A 'quality tag' like 'masterpiece' is, by definition, a metatag. It describes the subjective quality of the data, not the actual content of the image. If I remove 1girl, the girl disappears. If I remove masterpiece, she’s still there. That is metadata. Claiming it’s an 'extra added quality tag' doesn't change its function; it’s just semantic gymnastics to cover up a poorly written description.
The Franchise Contradiction: This is the most glaring part. You explicitly state: 'franchise-related tags like re:zero kara hajimeru isekai seikatsu are not used in training'. Then, two lines down, your own example is: rem \(re:zero\).
How can it be 'excluded' and 'not used' while simultaneously being the primary example of how to use the model?
If it’s not in the training data, why include the franchise tag in your example at all?
Your description is a mess of contradictions. You are telling people not to use the exact tags you are using to showcase the model. Instead of making up new definitions for what a metatag is, you should probably just fix your description to reflect what’s actually in your dataset. Or is it because you actually do not know how your own model functions?
@Phatcat
All of this is written in the context of Danbooru tags:
Tag "masterpiece" is not an official Danbooru metatag and is based on the image score.
Metatags (excluded in training):
https://danbooru.donmai.us/wiki_pages/tag_group%3Ametatags
Character Tag (used in training):
rem (re:zero) refers to the character tag, not the franchise tag
https://danbooru.donmai.us/wiki_pages/rem_(re:zero)?z=1
Franchise Tag (excluded in training): https://danbooru.donmai.us/wiki_pages/re:zero_kara_hajimeru_isekai_seikatsu?z=1
I chose to exclude tags, because of the SDXL token limit (75 tokens, 77 total). Franchise tags are often relatively long and are not much used, so removing them helps keep prompts within the token limit in training. It’s simply to free up space for more important tags. The same goes for metatags, training on something like "tagme" would be pointless.
In short: Danbooru metatags and franchise tags are excluded from the training data. I added "Danbooru" to the description to make make it a bit clearer. No offense, but you might need a Snickers. I’m not sure why you’re reacting so strongly.
The fact that you’re hiding behind a wiki’s specific categorization to explain why a character tag containing a franchise name isn't a franchise tag is exactly the 'semantic gymnastics' I was talking about. You updated your description because of my 'strong reaction,' so clearly the disconnect was real. You enjoy the Snickers; I'll stick to models with coherent documentation.
@Phatcat I’m always open to making things clearer or better. I don’t control how the character tags are created, "(Re:Zero)" is there to make the tag unique. Not some kind of 'semantic gymnastics' (are you alright?). Thats said, the way you're responding is unnecessarily confrontational, with ChatGPT like responces, even though I already explained why it was set up the way it is.
GREAT MODELS !!!!!!!!!!!!!!!!!!!!!!!!!!!