I realized you can redraw parts of the AI generated image but for some reason the dumb bot decided to add that black line to the masterpiece and I just noticed it after uploading.
I can retry in 3 days if you delete this mods, one leg has a relict of some swimwear too. I'm on a mission to perfection. Just leave me some uploading capability lol
Yes, the positive prompts include "drinking vodka". It does not appear because the Lora does not recognize it but... it seems that for some reasons it helps
It's hard to find the model used for images that are uploaded from Twitter/Pixiv unless the artist has specified it in their page/it's a self upload, or if the metadata is included.
The metadata here does exist but I can't find anything about a model called "Chiruchiru7" so it might as well just be a private model. But anyway, keep in mind that the uploader who uploaded a Pixiv/Twitter image most likely won't have an idea of the model used unless they already added it.
AI must be really trolling me: each time there was an issue with the image and I had to settle with this one, with physics of the... substance a bit off on her breasts. It was either that or eyes being off. About 4 hours of generation which was overkill but I hope you like this one
If anyone's asking, all of my uploads are generated by myself. I just figured out where to add the "artist's" name when posting and will do so from now on
If anyone's asking, all of my uploads are generated by myself. I just figured out where to add the "artist's" name when posting and will do so from now on
It does look really nice, the only issue would be the fingers on one hand with the little finger being too big. The image size perhaps could be bigger but I don't think that's decisive.
Same btw, I'm also in "pending" for a lot of uploads
I have seen a few AI images from Dall-E 3, it seems it can generate text more consistent and readable. Still has some problems but some images are impressive.
Reminds me of the film 'The place promised in our early days' one of several anime films that reintroduced me back to anime and manga some 12 years ago. Always was fascinated by the tower.
For some reason training this lora on the model with noise offset mixed in just gives overall better results... More detailed BGs, lower amount of problems with hair and eyes, etc. Will do some other LoRAs to check this further.
Honestly I think it looks great! I’d love to see more like this
Thank you! I've been trying out different styles. The model version for this does it nearly every time with the settings I use. I'll have to play around with it some more and get something uploaded today.
Antlers Anon, I just wanna say that you're my favorite artist on here. Everytime i think, "damn thats so good!" It ends up being one of yours. Thank you for your work!
What do lbw=char and lbw=paint mean? Do I need an extension for that?
These are written to use the LoRA Block Weight (https://github.com/hako-mikan/sd-webui-lora-block-weight), which allows for hierarchy-specific specifications when applying LoRA. Originally this notation was only used in LyCORIS, but now LoRA is also written in this way. As this is an extension, it must be installed first.
In this case, I set up "paint" and "char" as presets first and then use them.
Yo, How Did You Prompt Thesse Images? They Look So Good.
For this type of animation, I was using the loopbackwave script. For the prompt of the images, I took particular attention the face expression which changes every 7 to 15 frames
I used Based65 downloaded from CivitAI. I think perhaps there could be other versions of the textual inversions and the one LORA I used that could also be an issue? Let me know if using Based65 doesn't work and I'll try to get the versions of the textual inversions and add detail LORA I used.
One other important note: I don't have an Nvidia GPU so I can't use xformers, so try turning off xformers if you have it on.
Right, the Based65 from CivitAI is slightly different from the Hugging Face model but its probably this problem. And checking the LORAs I dicovered that I havent the "v5" version of "badhandsv5-neg" but It also doesnt look to be the problem Also, xformers was off since beggining.
One one hand, NovelAI has been extremely helpful is letting me design a cute character I can make a new OC (like this one), but on the other... about 95% of the results aren't what I consider good enough to be worth posting here.
I understand one of the hands looks a little bad, but I wanted to excuse it a bit because this is supposed to look (sort of) like a sketched image.
I understand that there are a lot of mistakes here, but this is literally the best thing that came out https://i.ibb.co/BBVr8D0/grid-0000.png Transparent cube was recognized as strangely as possible, but after I tried to change it to an aquarium, it only got worse, but here at least it looks beautiful...
Congratulations for your art! Im trying replicate it for study proposes but I cant it. The entire prompt is the same, the seed and the Sampling are the same, the original resolution is 512x768 and the Lora is the same (from Civitai). Excuse me, but what Based Mix version have you used?
I used Based65 downloaded from CivitAI. I think perhaps there could be other versions of the textual inversions and the one LORA I used that could also be an issue? Let me know if using Based65 doesn't work and I'll try to get the versions of the textual inversions and add detail LORA I used.
One other important note: I don't have an Nvidia GPU so I can't use xformers, so try turning off xformers if you have it on.
It's not rude to upload an image that's publicly available on the internet to other sites. Well... maybe it is, but it should be expected because that's how the internet works. Also, as I understand it, AI generated works cannot be copyrighted, therefore the person who generated this image technically cannot claim ownership over it.
If they wish to file a DMCA takedown request, it likely will be honored (as other requests have been in the past) but it wouldn't have to be, because of the whole "no copyright/ownership" thing. It would be a mere formality.
Congratulations for your art! Im trying replicate it for study proposes but I cant it. The entire prompt is the same, the seed and the Sampling are the same, the original resolution is 512x768 and the Lora is the same (from Civitai). Excuse me, but what Based Mix version have you used?
Oops, just noticed the repost is prohibited commentary, I'll add the artist tags and leave it up to the mods to decide if they want to keep it up or not! sorry q.q
You're more than fine. If the creator wants the image down, they can send a takedown request to the admin.
Oops, just noticed the repost is prohibited commentary, I'll add the artist tags and leave it up to the mods to decide if they want to keep it up or not! sorry q.q
i was wondering how you managed to get such a high quality generation with the sonia model until i realised you had another model for her, where did you get this sonia model from?
She's so cute!!! Have you thought about making an antler lora?
As in lora making antlers specifically? There are soma like that but I never felt the need to use them. Most models understand what having antlers on one's head looks like, more or less at least.
Can someone explain to me why this image is downvoted so much? It's not amazing but we had much much worse that are simply stuck at 0 votes. I'm just curious.
Can someone explain to me why this image is downvoted so much? It's not amazing but we had much much worse that are simply stuck at 0 votes. I'm just curious.
I guess you are right, but this has a more nostalgic feeling, which I like a lot.
The juice is worth the squeeze. It's easy to get discouraged by sameness/poor results when starting, but once you find just the right prompt and follow it, it becomes well worth it.
Finally what I was looking for when AI started: porn of Mea, by far the best but sadly underused girl in the series. Hope this continious. I am looking forward.
really like your works, which settings do u use in ultimate sd upscaler?
My workflow as of now is upscaling an image 2x with anime sharp separately first in Extras tab because the image are processed through img2img multiple times with varying denoise. With a tile size of around 896x896, the first upscaled image uses 0.5, 0.45, 0.4 denoise usually. Afterwards I take all those three images and pick out the details that are fine in each denoise and mask them in photoshop. After some fixes the finished product is upscaled once again 2x and put through img2img. With each 2x upscale, the ultimate sd upscaler will have a hard time processing each tiles since the image is too big for each tile to have context. Because of that I decrease my denoise levels to a certain degree where weird details won't pop up. In general, less than 1024x1024 size images can get away with 0.5,0.45, and 0.4 denoise, 2048x2048 can get away with 0.48, 0.45, 0.4, Larger than 4096x4096 sometimes can get away with 0.4, I tend to use 0.38, 0.34, and 0.3 denoise on the final upscale.
Hello, I am using the loopback wave script to make those. Parameters vary widely depending on the prompts and the amount of change needed throughout the animation. For more info about it, hit me up on discord at : kairos4463
hi there! ive send you an friend request, im irish man aka muppet
hi, i just wondering how do you do this? i would like to learn to make those myself
Hello, I am using the loopback wave script to make those. Parameters vary widely depending on the prompts and the amount of change needed throughout the animation. For more info about it, hit me up on discord at : kairos4463
