VAE is built into the model and does not need to be loaded
Prompt word reference example diagram and backmap area.
Recommended basic prompt words: score_9,score_8_up,score_7_up,
Recommended HD redraw rate: 0.39-0.47
Recommended starting resolution: 800X1280
Arbitrary sampler. CFG: 5-12 Recommended
It should be noted that the prompt word weight should not be set too high,
Too high is prone to flower pattern.
Description
This model is the FB8 version, which is mainly aimed at painting styles
This version is experimental in nature, the amount of data is small, and there are many problems!
You can download and play if you like.
Mapping workflow Download: https://www.liblib.art/modelinfo/352d777c1c284d4b9baf1ed0252d1c23?from=personal_page&versionUuid=19570be720e54cedbe739ba4d6d35c10
这个模型是FB8版本，主要是针对绘画风格
这个版本是实验性质数据量较小，问题也挺多！
喜欢的可以下载玩玩。
FAQ
Comments (20)
沙发！我抢到F1首发了，第一个
哈哈。你还没睡呀，其实这个里布我发好久了。今天才更新C站
@XUERYCJ 呃...我好久没登陆liblib
@METAFILM_Ai 你最近忙啥呢
@XUERYCJ 做模型呀！只不过，不混圈子了
@METAFILM_Ai 哦这样
I think this might be my favorite model on here.
Thank you for your love
4x_NMKD-Siax 我没有这个放大算法。请问我可以去哪里弄到？
这个放大不是必备的。可以用任意你喜欢的放大算法。你想下载可以去B站收一下或者哩布这类AI网站找找看
https://huggingface.co/ 直接搜就能找到，
或者直接在网站内搜索也能找到不少
I've used this model extensively, and while I don't think I've reached its ceiling yet, I do have a more experienced opinion on how it stands. Also, I have been using it for a while and I love it a lot, strange not to talk about it more.
Red-Blue is an amazing model, to me this model is even better than autism-pony, it has better quality (imho), it has really good prompt adherence and flexibility that rivals base pony.
Relevant to note that I am biased towards a good Lora compatibility and improvement, more so in artists styles and what not.
Pros:
- Really good Lora compatibility, one of the best models in this aspects, in my opinion even better than options like Autism.
- Base quality is extremely good, can even improve some Lora's quality by itself, which is great.
- Can use natural language without breaking too much, but keep in mind, if you don't use realistic on negatives, "anime" tag, or a lora, it will give you a very western art piece (going for a very realistic approach).
- Tame, but can be very exaggerated if needed.
Cons:
- Realistic in the negatives can break your images very easily. It can reduce the quality of the image, and also impacts the anatomy. If you want your images to be more on the anime side, use [anime] before scores and don't use natural language.
- Very little support for western/cartoon/toon. And this is totally understandable, as this is more of an "anime" specific checkpoint, BUT it does support cartoon Loras very well, even better than other options. (A cheat around this is to use tags like: minimalistic, flat colours, thick outlines/lines, toon/cartoon, and use realistic in negatives. Keep in mind that using realistic in negatives will lower the quality of the image, and in most cases will introduce anatomical errors).
- Character recognition is not the best, but I tend to work more on OCs, so I don't mind it too much. Same with artist recognition, but again, a Lora is always better.
Thanks for that detailed review. This model also mostly works well for me, except it seems to have trouble with the "1boy" tag which often results in the man not appearing at all. Have you encountered this and any idea how to fix it?
@heavyiron7914 I haven't had any issues in that aspect, some of my posts have male characters. I can try to help you out if I can, you can DM me if you want.
感谢绪儿分享，其他都好，就是C站下载速度很慢了。
Anyone has problems with the "1boy" tag with this model? For all pony models out there, 1boy is that one thing that consistently work, but with this one it often results in nothing. Just no man showing up.
Any idea how to make it work? Or do I need to use a different tag?
你可以试试负面提示词中削减它出现的几率
不好用，很不稳定，“signature,watermarks,”这两个负面提示词对画面风格影响非常大
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.