- Nov 5, 2018
- 1,319
- 3,218
- Sep 13, 2020
- 471
- 424
- Sep 13, 2020
- 471
- 424
so where's the pic where you drip the candle wax on her face? or are gonna singe her nipples you mean, mean man?! poor Mila! please don't hurt my darling !
You don't have permission to view the spoiler content.
Log in or register now.
- May 6, 2020
- 60
- 447
![Mila AI [1].jpg](https://attachments.f95zone.to/2025/01/4490791_Mila_AI_1.jpg "Mila AI [1].jpg")
Last edited:
- May 6, 2020
- 60
- 447
![Mila AI [2].jpg](https://attachments.f95zone.to/2025/01/4490834_Mila_AI_2.jpg "Mila AI [2].jpg")
![Mila AI [3].jpg](https://attachments.f95zone.to/2025/01/4490835_Mila_AI_3.jpg "Mila AI [3].jpg")
![Mila AI [4].jpg](https://attachments.f95zone.to/2025/01/4490836_Mila_AI_4.jpg "Mila AI [4].jpg")
- Sep 13, 2020
- 471
- 424
From our entire community, we would like to congratulate Mila on her past birthday!
View attachment 4464780
Can you also try the one in the game thread? It's working for me, and I am not sure what is the problem
Also didn't work, but it was actually a misleading error message from Discord. Got that sorted. Thanks for the prompt reply and apologies for not checking that more carefully before posting here. My bad.Can you also try the one in the game thread? It's working for me, and I am not sure what is the problem
Now that I have access to the discord and have spent a considerable amount of time searching through it ... I'm afraid I still cannot locate the basic prompt for Mila advertised in the OP. I must be a little ... err ... slow. Can anyone point me in the right direction? Thanks.
Do you see #general channel? If not check out the rules channel and hit the reaction (any kind). It will give you a fan role. Then you'll see the rest of the channels.
After that go to the #general channel and check pinned messages (pin button in the upper right corner). The little guide is there. There is a bigger one, but it's in the #guides channel that is only for SS subs, but there is more advanced stuff)
That did the trick. I obviously figured out the part where you had to accept the rules before you got to see anything, else I couldn't have browsed the channels at all - as you say. I did miss how to toggle the pinned theads, though, thanks for that.
In case it isn't obvious, I haven't used Reddit for a long time (or much at all) - in fact, that was the cause for my earlier problem, I got the message that the invite had expired. When I switched to the app everything appeared to be in order. But in the webapp, it wanted me to authenticate via phone.
Anyway, thanks for the heads up. I set up Stable Diffusion, maybe I'll be able to get this stuff running.
Any tips on getting the face right? I've been trying a ton of images now, and while - if I closely stick to the prompt that was proviced - the pictures come out with mostly the "right" style, the face always looks entirely different.
Meanwhile, here's a image I did that avoided the issue. Nothing to write home about, I'm sure, but keep in mind I just started with this.
Meanwhile, here's a image I did that avoided the issue. Nothing to write home about, I'm sure, but keep in mind I just started with this.
- May 6, 2020
- 60
- 447
For faces, use ADetailer, and the rest will come with experience.Any tips on getting the face right? I've been trying a ton of images now, and while - if I closely stick to the prompt that was proviced - the pictures come out with mostly the "right" style, the face always looks entirely different.
Meanwhile, here's a image I did that avoided the issue. Nothing to write home about, I'm sure, but keep in mind I just started with this.
Also you will need my Mila Lora from the discord channel, and you will need to put in the Lora folder to make it work) Oh and I use refiner model for face and lately for base image too) Though I don't like thin neck it makes, so I fix it with photoshop and low den str later if needed.Any tips on getting the face right? I've been trying a ton of images now, and while - if I closely stick to the prompt that was proviced - the pictures come out with mostly the "right" style, the face always looks entirely different.
Meanwhile, here's a image I did that avoided the issue. Nothing to write home about, I'm sure, but keep in mind I just started with this.
I know that it's my fault since I started here asking a stupid question, but believe it or not, I was aware I had to use the Mila LORA.
I'm using the prompt posted on the Discord and the Mila Lora posted there, too. Everything looks ... kinda sorta okay, but the face looks nothing like Mila. Sometimes, it's totally distorted, sometimes it's a really nice looking face, but completely different from Mila as seen in the game. It's almost as if I'm missing another Lora or something.
I'll post some examples tomorrow, ran out of time to fiddle around with AI for today.
Thanks.
Somewhere I got this image: "126_walk_date_cafe_smirk.png", I don't even remember where from.
But throwing it in PNGInfo reveals all the used prompts.
I tried generating that image, and what I got was way off. First, all the Mila images tended to be naked, but I presume that's why in the actual prompt, there's a seperate _vanilla Lora being used (I don't think this was ever shared, was it?).
I got around this by adjusting the negative prompt.
Next I had the issue that SD tended to distort Mila or put multiple instances of Mila into the image, in my experimentation so far I concluded that this often happens if SD has "too much space". I failed to address this reliably with the prompt, my solution was to either reduce the image size (which can technically then be upscaled) or using Regional Prompting. But I believe regional prompting is not baked into the PNGs.
Then I got halfway decent images, but this time they were nothing like the original style.
In the output window, I noticed this Lora being reported as missing: "Smooth Style 2 SDXL_LoRA_Pony Diffusion V6 XL"
I'm not able to locate a download for this specific Lora, either, I was under the impression Smooth Style 2 was like a trigger word being filled into the main model/checkpoint being used, but apparently this is not the case?
Can anyone provide a download link to this Lora?
Thanks.
But throwing it in PNGInfo reveals all the used prompts.
I tried generating that image, and what I got was way off. First, all the Mila images tended to be naked, but I presume that's why in the actual prompt, there's a seperate _vanilla Lora being used (I don't think this was ever shared, was it?).
I got around this by adjusting the negative prompt.
Next I had the issue that SD tended to distort Mila or put multiple instances of Mila into the image, in my experimentation so far I concluded that this often happens if SD has "too much space". I failed to address this reliably with the prompt, my solution was to either reduce the image size (which can technically then be upscaled) or using Regional Prompting. But I believe regional prompting is not baked into the PNGs.
Then I got halfway decent images, but this time they were nothing like the original style.
In the output window, I noticed this Lora being reported as missing: "Smooth Style 2 SDXL_LoRA_Pony Diffusion V6 XL"
I'm not able to locate a download for this specific Lora, either, I was under the impression Smooth Style 2 was like a trigger word being filled into the main model/checkpoint being used, but apparently this is not the case?
Can anyone provide a download link to this Lora?
Thanks.
There is no vanila Lora, and yeah Mila likes to go topless, or full nude. That's because I was focused in maintaining her body parts as stable as could be in training, so there are a lor of nude pics in dataset.
To make pictures without the flashing privates part add "rating nsfw" and "rating questionable" in negatives, should do the trick.
My workflow consists in lots of inpaintings, so the image data will have only the last generation seed and params, so don't expect to make something similar from those data. Not even one image from the game can be generated without the adjusting intermediate results - it's a process of making a base, change it in photoshop, regenerate in img2img, fix it in photoshop, regenerate in inpainting, fix it in photoshop - etc.
In generating in txt2img (which I almost don't use now, beacause it's easier to control the composition and colors if you make/draw the base picture, then trying your luck in SD gacha) you may get multiple Mila's or multiple men. You will get less of that if you stick to 768x1152 and 1024x1024 resolution for base picture. But even then - you still will get that from time to time. It's anavoidable. Just keep the gacha rolling or try to change the prompt. Also the model may not know stuff that you want to make. Use/train Loras for that)
Regional prompting and controlNet are cool but they affect the style and lessen your control in my opinion. That's why I don't use them since 1.2.4.1 version.
It's smooth anime style you can
You must be registered to see the links