If you enjoy my work, please give me a five-star rating; it means a lot to me! Check out other creations here.
如果你喜欢我的作品，请给我一个五星好评，这对我非常重要！这里看其他作品。
I don't know English, the following translation is from chatGPT.
我不会英文，以下译文来自chatGPT
RimLight
"RimLight," as the name suggests, is designed to pursue better rim lighting effects, serving as a complement to "DarkLight."
《RimLight》顾名思义，就是为了追求更好的轮廓光的效果，作为《DarkLight》的一种补充。
DarkLight
Bring more realistic and artistic lighting / shadowing to your images,with better contrast backlighting without losing any detail
带来更真实且充满艺术感的光照/阴影，有着更高对比度的背光而不丢失任何细节
Advantages of DarkLight
1.Higher threshold shading: better shadows, especially when backlit (such as adding silhouette prompt words)
阈值更高的明暗度：更好的阴影，尤其是在逆光时（如加入silhouette提示词）
2.Flexible and controllable: increase or decrease the weight to adjust the contrast
灵活可控：增减权重即可对对比度进行调节
3.Lower weight: In a large number of tests, the weight is often set to 0.3 to be effective.
更低权重：在大量测试中，权重往往开到0.3即可发挥作用
4.Retention of dark details: especially prominent on realistic characters (others have yet to be tested)
暗部细节的保留：在写实人物上尤为突出（其他尚待测试）
5.Better black and white images: It’s not difficult to understand, after all, the contrast levels are richer
更好的黑白图像：不难理解，毕竟层次对比都更丰富了
6.Brighter colors: It’s a bit surprising, but the contrast does not only affect black, white and gray
更亮的颜色：这有点令人惊讶，但对比度不仅影响黑色、白色和灰色。
about prompt
Most of the tags in the data set include (1girl, spotlight, dark theme), so putting them at the head of the prompt word will have better results.
1girl can be changed into any character, such as 1man, 1cat……
You can add a color word in front of spotlight, such as red spotlight, then the light source will turn red. It may take a few more tries, but the effect is already very good without adding it.
Dark theme can also modify the color, such as red theme, then the tone of the dark part will be more red. In addition, if you use green theme, there is a certain probability that a forest will appear in the background. This depends on the large model. I will update it in the next version. improvements in
在数据集中大部分的标签都包含（1girl,spotlight,dark theme）所以把它们放在提示词头部，效果更好
1girl可以换成任意角色，如1man,1cat
spotlight前面可加颜色词，如red spotlight，那么光源将变成红色，可能要多试几次，但不加其实效果也已经很好了
dark theme同样可以修改颜色，如red theme，那么暗部的色调将更多呈现红色，另外，如果你使用green theme，背景则有一定概率会出现森林，这个取决于大模型，我将会在下一个版本中进行改善（也许maybe）
prompt:
1girl,spotlight,silhouette,portrait,parted lips,closed eyes,beautiful detailed skin,one hair bun,turtleneck_sweater,dark background,<lora:spotlight:0.84>,
Negative prompt:
nsfw,logo,text,bad anatomy,verybadimagenegative_v1.3,extra limb,missing limb,disconnected limbs,malformed hands,ugly,
Steps: 60,
Sampler: DPM++ 2M Karras,
CFG scale: 5,
Seed: 1003785097,
Size: 640x960,
Model: Realistic Vision V5.1
If you'd like to support me and do more : ☕ https://ko-fi.com/vjleoliu ☕commissions open
checkpoints
Each lora is trained through hundreds of high-definition pictures, trained on 1.5 checkpoints, and verified on 10 famous checkpoints, almost including all the realistic styles and 2.5D styles on the leaderboard, the list of checkpoints is as follows:
推荐模型
AbsoluteReality
chilloutmix_NiPrunedFp32Fix
CyberRealistic
deliberate
DreamShaper
Edge Of Realism_EOR
Ether Real Mix
lyriel
Realistic Vision
ReV Animated
Since you've reached the end, please give me a five-star review;
it's very, very important for my ongoing creativity! Thank you!
都看到这了，不给个五星好评么？
I hope you equally enjoy my other works.
Description
Respect copyright and the fruits of personal labor. If you use this model elsewhere, please indicate the author. This model is for learning and communication purposes only. Please do not use it for any commercial activities.
FAQ
Comments (31)
Any chance for SDXL version?
I hate hard lights, directional lights and especially spotlights but this one is looking very stylish and has an aesthetical approach so It is quite nice. I will do some tryouts myself and post if it's any good. 🔦✨
silhouette: shape can be distinguished by the light more
dark theme: makes it real dark, best if used with brackets to avoid pitch black
"spotlight" is the caller, lora weight could be used to increase general brightness (that is what I did for this)
@Twnsth You have discovered all the keys to this lora
"Failed to match keys" ..... what causes this issue ?
lora does not contain any match keys. This problem may be related to webui.
I love this lora, please use your magic to make more like this one! Something related to photography would be my wish.
I've been working hard, thank you for your support, it means a lot to me
Great model. Thank you !
With the right word prompt, it can create amazing images with a professional photographic style.
Thank you for liking it
I like this and want to use it for other things. I have auto lora trigger words setup and the "1girl" is obviously not helping. Can you remove it?
1girl is not the main trigger word. You can change it to any subject you want, even animals.
@vjleoliu I don't think a Lora should have anything close to a subject. It really should just be enhancing whatever subject the user chooses. That's like a Lego Lora having "house" built in and being forced to change it to car?
@opensourcefan I agree with your perspective, but light and shadow do require a subject. Removing the subject from them would render light and shadow meaningless. I will consider enhancing the LoRa's generality in future versions so that it can accommodate a wider range of subjects. However, this would imply a significant increase in the difficulty and workload of collecting datasets. I hope there's a better way to address this issue, and thank you for your input.
SOLVED: See Post.
Seems to have an issue keeping the subjects eyes open. Not entirely sure what's going on here. Tried adding eyes open to the positive prompt, then eyes closed to the negative, then both at the same time.
AnythingV4, and AnythingV4.5. Going to try a few more models and see whats going on.
Reducing the weight causes the subjects eyes to open slightly. Fully at 0, closed at 1.
Odd.
EDIT: Did it with DarkSun too.
Adding: closed eyes, closed_eyes, closed_eye,
To the negative embedding reduces the closed eye effect enough that you can use the LORA to .75 weight. XY plot time.
EDIT2: Using an XY plot I managed to figure it out. It's the clip strength. The higher the clip strength, the more the eye close.
交流合作，搜索QQ群482570609，请注明来意
what is this trained on? it's great lora but always makes humans/faces. any chance of a more generic one or sharing the type of data so we can recreate?
This is the coolest lora on civit!
Will you be making a pony version??
sorry,what's pony version?
pony xl is all the rage now! it's a powerful redo of sdxl checkpoint, but I also ignored it until now because it had a weird name lol
I see, thanks for letting me know. sdxl is a much more expensive version, and I don't think I'll consider making a version of sdxl for a while if there's no special reason for it
@vjleoliu bummer, but I understand!
@vjleoliu That's too bad. I only use SDXL now and this Lora is one of the few things I miss from SD1.5.
@THX1139 Thanks for the like, as I said before, training lora is a public service, and my current investment can only support me in training the sd1.5 version of lora, so I hope you'll understand.
@vjleoliu yeah I understand, thank you for what you've contributed!
I use it on Pony seems to work.
Occasionally get this error Networks with errors: spotlight (73)
I'm on A1111 and using a Pony checkpoint if that matters. Is it referring to the LORA or the token "spotlight"
this is lora is working for sd1.5,It does not support other versions