【廣東話、附教學】一鍵動漫變真人|免費、免安裝、新手適用

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
Always seeing people making anime characters to real-life? Sounds powerful Can I do it myself? Want to improve your image generation skills learn about prompting? Anime characters to real-life Let's start! Hello everyone Welcome to "An IT-a" I'm Anita A few days ago, I realised the 3 animes that I'm watching suddenly ended together All of a sudden I feel so lost and down In order to cheer myself up I decided to turn them all into real-life people In this process I found that turning anime characters to real-life compared with turning real-life photos to anime is more difficult Last time even if I leave out the prompts it still works But if I leave out the prompts this time the result is quite disappointing Therefore this time besides sharing my Colab Project I'll also share the prompting skills to let everyone be able to do this by your own The prompting skills in this video not only applicable on this Colab project but all image generations Remember to watch till the end Same as before using Colab, free, no installation only need a Google Account Let's start now! First open the first link in the description Just like last time you decide by yourself whether to copy to you own Google Drive then "Change runtime type" choose "Python 3", "T4 GPU" Yes, what you see here is different from yours As I always need to use Colab I paid Run step 1 and 2 Choose style Here's 3 models good for turning into real-life First - Photon the result is more likely to be foreigner Second - Fantastic Mix more likely to be Asian depends on what nationality suits your character Third - REV Animated This is half-real-life I'll call it 2.5D But I see some YouTuber thinks this is already real-life style So I put it here for you If you have high standard try tuning "Strength" and "Control Net Weight" Just some parameters to control how much to follow your image If you have model that you like I still have "Custom model" section If you wanna know the details of this section watch my last video After choosing all the stuff Press Run Next, open the File System by clicking this icon Drag your image in Click OK Then fill in the filename, additional prompt, and negative prompt I'll talk about how to build the prompt later Dun worry This "Use Embedding" is If you found that some characteristics cannot be generated it may be caused by the embeddings you can disable it here Click Run Right Click to Save OK As the result depends a lot on a good prompt Let's try together on how to build the prompt for these 6 characters Start from my beloved Himmel Let's see the result without a prompt Oh... Turns into a girl... Not ok First we need to know more about this character it will be easier for us to describe Himmel he died in the first episode But you will be able to see him influencing everyone throughout the anime Although his name means the sky But I think he's more like a sun keeping everyone nice and warm Despite narcissism he got a bob hairstyle First we'll define him as "young man" Then describe his face In my mind he's handsome always giving a big warm smile warm smile Hairstyle sky blue hair, blunt bangs sky blue hair, blunt bangs If you can't think of the English of your description E.g. bang It's normal English is not our mother language We can search in Google "bang eng" My English is not that bad But in the process of image generation I learn a lot in English Clothing white cape, blue clothes Image composition close up Last is the background forest background Let's try Not handsome enough eyebrows are too thick There's no teeth in the original image I'll add in the negative prompt ugly face thick eyebrows teeth Next is Himmel's beloved Furīren Let's see how it looks without prompts Just a normal girl Descibing Furīren needs some skills Let's know more about her Furīren is an elf Her life is far more longer than human beings So she sees "Time" way different from us In the settings elf has way much less emotions than human So her face is always cool you can see her smile but seldom see her laugh And as she's quite old she has powerful magical skills making her so confident She even used "scary" to describe herself First, her elf ears is her biggest characteristic we need to keep it So the first word this time will not be "1girl" but "elf" Next is her face emphasize her pointy elf ears pointy ears hairstyle long white twintails clothing - white robe holding a scepter holding a scepter background forest background oops! add sexy, breast, nude to negative prompt To avoid this situation I'll add this to all female characters also add "ugly face" not very good finetune again Her skin seems to be pale pale skin emphasize more on the ears add the weight to 1.9 emerald eyes emerald eyes More description on feelings confident, cool, wizard Can't get the elf ears maybe because the embedding thinks that ears should not be like that Let's remove the embeddings Success! I also tried some close-up for her These 2 can still use embedding Not to waste time Let you guys see the results directly Next Pharmacist - Maomao Maomao learnt a lot of herbal knowledge from his father Another child detective She was born in an ear that woman had no place As she lived in dangerous area To avoid getting into trouble She'll draw feckles on her face every day to make herself look ugly In fact, her mother is the top prostitute so she got a very beautiful appearance The first thing we need to do is add "1girl" then her face blue eyes freckle on nose hairstyle green blunt bangs, braid confident smile background lantern at background First thing Clothes are wrong wearing a t-shirt too childish Dunno why her mouth so big the eyebrows are too thick First her clothes Dunno how to describe I tried "traditional clothes" but failed Let's assume it's kimono green kimono Add to negative prompt child, big mouth, thick eyebrows Done Next is Jinshi Dunno how to pronounce it Jinshi has a high position in the castle He's even one of the royal family As he's even more beautiful than woman In order not to cause trouble he kept pretending to be a eunuch As he is pretty enough to act as eunuch we should avoid all make characteristics For example, laryngeal core First "young man" girl-like, sexual neutral Hairstyle Hairstyle Hairstyle purple hair, high bun, half-up Clothing actually yukata will be better than kimono blue cape, yukata Add handsome I put "handsome" at the end is because it has the meaning of male but this time we should avoid this kind of meaning So I'll put it at the end plain color background to avoid a strange background I tried adding "laryngeal nucleus" in negative prompt But still can't prevent it from coming out Ended up adding a weight on "girl-like" Ta-da~! But... The result of 2.5D is still so men Let's do it again Done Ended up from a very feminine man to a very manly woman If you are interested in the full prompt click into the samle link main differences are related to gender-related keywords Hmm... Jinshi is quite challenging need to balance between male and female done a several times and not all models fit Need 2 sets of prompts Not many people know this person But it's an anime that I recommend you to watch But dun ask me to read it's name Rishe The story tells that she is a wealthy lady but already reincarnated seven times every time she died because of a war As she has chosen a different life each time She had a lot of experiences She is no longer an ordinary rich lady that young and naive And this time She became the fiancée of the key of the war - Arnold Rishe's special is that she has shiny pink hair wearing fancy evening clothes Rishe should be easy to describe Hair, clothes and expression should be enough 1girl pink shiny long hair blunt bangs emerald eyes confident smile red evening gown To be safe add "thick eyebrows" to negative prompt Super pretty Done at 1st trial Last but not least, Rishe's husband prince Arnold A mysterious character The eldest prince good at both civil, military, sword and strategy In the hearts of most people, he is a very cold person In the past few times he even killed his father and started the war which killed Rishe for 6 times But this time Rishe found the gentle side of him First "young man" Hairstyle blue shiny hair Next is Rishe's favourite his eyes blue eyes action - one hand holding chin add some "cool" black gloves firm eyes confident tuxedo in negative prompt ugly face thick eyebrows smile I found that I was kidding myself a bit by choosing this photo Since it's troublesome for stable diffusion to generate hands In other situations I'll use control net's open pose to handle it But this time, I've already used control net's depth model If I add a second control net Dunno if free version of Colab can run or not Ended up to increase the weight to 1.9 still like lucky draw bad hands or bad face tried a lot of times But this is fine In fact, there's still a lot of room for improvement E.g. add "Cinematic movie scene" to enhance the cinematic feel But if I want to finish talking all the techniques This video will become few hours long So I stop here My prompt this time can be considered entry-level Is the level of controlling what's in the image If theme matches let's see if we can have another video sharing more advanced prompting skills Remember, there's no absolute answer in prompting many things to consider E.g. model, the effect in your mind, etc Even for the same model there are many ways to achieve the same result So be patient try more changing the prompt step by step You can get the result that you want Let's summarize the techniques we used today First draft your first prompt try it out add or remove keywords accordingly if that doesn't help try changing the weights or the order of keywords the result will be different If you still cannot get your result try removing the embeddings see if it helps Or describe it from another angle like the case of Jinshi Alright! This is the end of this video If you think this video helps Just click the "Like" below If you have anything to tell me Leave me a comment Subscribe now if you haven't If you want to watch my other AI tutorials Click here now See you next time Bye Bye!
Info
Channel: An IT-a by Anita Wong
Views: 162
Rating: undefined out of 5
Keywords: ai, stable diffusion, ai art, colab
Id: ZQHei9MUCnc
Channel Id: undefined
Length: 13min 15sec (795 seconds)
Published: Thu Apr 04 2024
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.