hello! viewer from the otherside of the ocean. Good video as usual! Let me share some findings I got. So, the prompts: Masterpiece, best quality are mostly for "anime" models, since the original model had those tags in it but they mostly wont have a lot of weight on "realistic" models. Realistic models benefits more from "photography" prompts, such as "volumetric light", "hyperrealism" etc. Both, anime and realistic benefist from negative prompts like you showed! Can I make a request for a video? there's two extensions that I think can be pretty useful but needs some research and testing: 1.- multidiffusion upscaler and 2.- Local Latent Upscaler Thank you for your hard work!
@@in-room I admit that the science of the underscore remains a mystery to me, sometimes it gives a completely different result if I don't put them, I sometimes play with double quotes etc ... but yes I admit I tested on the prompts negative and it makes me "just" modify the broadcast cloud.
After looking at the image, I'm convinced that the underscore makes noise. What we need to think about here is not whether or not to use underscore. The reason why I started this "Prompt study" is to find out that some prompts, for example, underscores, do not accept other prompts. Bottom line I don't think it matters if the picture I want comes out and whether it's an underscore or not! I don't know if it'll be delivered well because I used a translator...
@@in-room yes yes, I admit that I was irrelevant to the video (the power of negativ prompt) , and me too, I go on google translate (I am from France) sorry ;)
You've created an excellent content again! Thank you so much from Ukraine!
Thank you for watching!!
무지성으로 다 때려넣기보단
하나씩 넣고 적용해며 차이점이 있으면 추가하고
효과과 미비하거나 없으면 않넣고 그런식으로 하는게 좋을거같네요
네 맞아용ㅎㅎ 시청 감사합니다!
퀄리티 프롬프트는 별 의미 없이 그냥 넣었는데 영향력이 크네요
맞아요 ㅎㅎ 저도 얼마전까진 복붙해다 썼는데 이제 좀 파보려구요~ 시청 감사합니다!
nice video! attention heatmap might be a good extension to check and improve the prompt efficiency,thx alot.
Thank you for watching!!
hello! viewer from the otherside of the ocean. Good video as usual! Let me share some findings I got.
So, the prompts: Masterpiece, best quality are mostly for "anime" models, since the original model had those tags in it but they mostly wont have a lot of weight on "realistic" models. Realistic models benefits more from "photography" prompts, such as "volumetric light", "hyperrealism" etc. Both, anime and realistic benefist from negative prompts like you showed!
Can I make a request for a video? there's two extensions that I think can be pretty useful but needs some research and testing: 1.- multidiffusion upscaler and 2.- Local Latent Upscaler
Thank you for your hard work!
Thank you for the good information and watching!
As for the requests, there are things I'm planning, so I'll try to deal with them next time.
And both extensions are installed, but they're difficult and I haven't been able to use them yet either.ㅠ
verry good idea ! do you have test with and without underscore? (low_quality,worts_quality,normal_quality:1.5)
Yes, I'll test it
But I heard that underscores are perceived as noise?
@@in-room I admit that the science of the underscore remains a mystery to me, sometimes it gives a completely different result if I don't put them, I sometimes play with double quotes etc ... but yes I admit I tested on the prompts negative and it makes me "just" modify the broadcast cloud.
@@in-room
I have try some test, to know if it's important to get some link with the words : zopiak.com/sd/bat_woman_mobile.jpg
After looking at the image, I'm convinced that the underscore makes noise. What we need to think about here is not whether or not to use underscore.
The reason why I started this "Prompt study" is to find out that some prompts, for example, underscores, do not accept other prompts. Bottom line I don't think it matters if the picture I want comes out and whether it's an underscore or not!
I don't know if it'll be delivered well because I used a translator...
@@in-room yes yes, I admit that I was irrelevant to the video (the power of negativ prompt) , and me too, I go on google translate (I am from France) sorry ;)