REALLY COOL!!!
❤️❤️❤️💐☺️
Thanks! You made this possible by uploading OLDFIsh and RMHF to your models! 😍 😄
Hey digiplay, we haven't talked in a while!
I was trying to have my fun with other OldFish models, but they're throwing errors at me. I noticed your version that works fine is called "diffusers_fix", and seems to add stuff the originals had missing. How did you make it? My attempts to make a fixed model produces ones that work fine too, but they produce different images than the original so I was wondering what's the way to properly make a "diffusers_fix" version.
❤️Hi Yntec❤️ I miss you haha
There is no magic in diffusers_fix, because I do not Understand many codes or program , so I just mix a normal not broken model with it, fish model is 0.999 , another is 0.001.. (maybe, I fogot 😆)
I am not smart but I really like fish models, and I fond many models from him has this problem , but he has no contact ways (few month ago ,he didn't reply any questions on civitai.com , and I either can't find his Chinese webpage ) in his civitai.com page says that model has merge permission, so I try to merge this model with another model.. this fix is my FIRST merge , maybe I use wrong way 😆 or maybe I have another bug so generated images is not the same. (because it is not an Original model, actually)
After another merge experiment I think this fix model is pretty cool, so I upload to here, and I meet you.
I really don't know much, I see your new models are really cool and cute🥰,
looks very complex to merge.
Fish "hello world " model is normal in diffusers convert. if you want to fix another fish model model, I suggest maybe you can use this model and merge with problem model? I guest it will work? or ask Mr. fish?
( I check Fish's civitai.com page just now, he seems to start to reply another users now, in his new models.
recently I am still not in good health haha,cannot test and play models now.
but I hope I'll recover ❤️ share something and talk with you again . 🐱🐇🐣🦁😄
Thanks for replying digiplay! I also missed you, I'm your biggest fan! I was even worried you had quit because you got bored, I'm glad to know you'll be back when you're healthy!
OldFish 1.1 is my favorite anime model, it doesn't even look AI generated, I've been trying to fix it in the way you did, and even managed to make merges that are 1.0 OldFish and 0.0 the other model (then the other model doesn't matter and I always get the same outcome), but after everything is said and done, it never produces the same images! I suspect that what timevisitor did was bake in a VAE, in a way where the model is able to produce those sharp and colorful results without one, and then REMOVED it, to save space, because, the models draws as if it had it, you don't need a VAE at all. But then trying to fix it by adding a VAE... just, let me show you:
Top left is OldFish 1.1. Aah, it draws the dream, but one can't make diffusers out of it. The other 3 are my attempts to fix it, I mean, they're all fixed versions that can be used for diffusers. But they don't draw the same thing! Extra girls appear on the picture in different styles. BlessedVAE is the bottom right and it loses the thread and makes something completely different! 😄 But ColorVAE makes that composition too, so it looks it was based on Blessed2.
I'll continue researching this, hope one can day I can fix it while retaining the generated content. Nobody does what OldFish 1.1 does, not even other OldFish versions! There must be a way, and one day I'll find it, I hope! 😆
Wow are you online?
I just merge 1 special version and tested,
found many hands problem is also in original v1.1 ,
maybe use "just two hands" prompt can reduce these bug? haha
I am going to rest now :D
dizziness 🥹🥹
https://huggingface.co/digiplay/OldFish_v1.1/blob/main/OldFish_v1.1_Sp1.safetensors
btw, I still think your GOLDFish is better ❤️ my no.1❤️
Oh yes! I was there right when you uploaded it, and saw you convert it to diffusers! I hearted it as soon as I could, and checking your page was the first thing I checked when waking up, I'm reaaally close of going over the creeper line! 😆😆😆
I like GOLDFish a lot as a 2.5D model, but for pure anime nothing comes close to OldFish, those flat shadows are perfect, I intend to make more mixes with it, I still don't know how to thank you for introducing me to the model ❤️❤️❤️
hi, I don't know how and why, when I is slpeepy, suddendly model merge OK 😳
you can try :
https://huggingface.co/digiplay/OldFish_v1.1_diffusers_recover
and tell me generated images are the same or not,
I just test 2 pictures, it looks like really 100% the same but I don't know why😆
oh no, I test 3rd picture , but it not the same 😱🤣😭
pls wait for me...
Wow! I just got here and your model had already disappeared!
I am so sorry, let me test for a while, its very strange 🥹
when I test 3rd picture , the result is not the same 😭😭😭
pls wait for me...
No problem! I just hope you still save the model you deleted and store it on your OldFish 1.1 repo, even if it didn't work, it's very cool to have alternate versions of the model! Because, what if it's not the same, but it is better!?
Here's my repo for Hassanim:
https://huggingface.co/Yntec/Hassanim/tree/main
Only 1 model worked, but I saved all the others anyway :D - thanks for all your work!
Hi Yntec 💐
you can try model now: (the same with Original output)
https://huggingface.co/digiplay/OldFish_v1.1_diffusers_recover
and dream-mix version I put to here 😆😆):
https://huggingface.co/digiplay/OldFish_v1.1_personal_HDmix
💐wait for your feedback 💐
Awesome! Thanks fr the models! I'll try them now! ❤️❤️❤️
And... oh! I forgot to mention something very important about OldFish 11! Please check it you:
Load Old Fish in a fresh instance and generate a picture
Now, load a different model and generate a picture with the same settings
Now, reload Old Fish and generate the first one again.
It'll generate a different picture! Same prompt, same settings, same seed, same everything, and a completely different picture! What Old Fish generates depends on the previous model loaded and the previous generated picture! That's why it can be hard to fix it, you think you have it, but next time it doesn't match, because it'll generate something different with same settings! The only way to try it is to make all the ones beforehand from a fresh UI and never generate them with it after loading a different model!
Gonna check your new versions now! New toys! 😄😍
Haha! What I love about all this is that we're getting some many alternate versions of OldFish! You may have improved the faces it draws! 😍😍😍
I just noticed you uploaded other ones, I'll check them too! ❤️❤️❤️
Hi Yntec , I upload a video you may compare these two images, (note the tab)
I use different online server to run these different model(Original/I convert), and can get the same result. as you said , sometimes Original OldFish_v1.1 generated different images , but a lot of times, they can generated the same image, I don't know why ... 😆 maybe its an alive AI model? 🤣🤣
I think you are really expert.. maybe can find real problem and solution,
I've already tried my best 🥹😆🥲 going to rest now haha , see you next time 💐
tmpdm_s1037.png generated by 7be100b0f3, Original OldFish_v1.1 ,
AUTOMATIC1111 VERSION v1.6.0
1girl,19y.o,hold a blue book,mountain ,lake,close-up ,sakura trees ,golden hair,green eyes,tiny smile ,
Negative prompt: (bad and mutated hands:1.5), (worst quality:1.2), (low quality:1.2),
Steps: 15, Sampler: Euler, CFG scale: 3, Seed: 3930771012, Size: 680x512, Model hash: 7be100b0f3, Clip skip: 2, Version: v1.6.0
tmpfz5is6x5.png generated by 79225b2a0f, OldFish_v1.1diff.safetensors (I convert),
AUTOMATIC1111 VERSION v1.5.1
1girl,19y.o,hold a blue book,mountain ,lake,close-up ,sakura trees ,golden hair,green eyes,tiny smile ,
Negative prompt: (bad and mutated hands:1.5), (worst quality:1.2), (low quality:1.2),
Steps: 15, Sampler: Euler, CFG scale: 3, Seed: 3930771012, Size: 680x512, Model hash: 79225b2a0f, Clip skip: 2, Version: v1.5.1
(you may download these two pictures, they are including generated information )
Thanks a lot digiplay!
I have created this space so you can check what I'm talking about:
https://huggingface.co/spaces/Yntec/Oldfish11-Webui-CPU
On there you can test it, if you generate images with OldFish 1.1, you'd get the "original outputs", the ones we're after! If we generate a single one with your 1.1diff version, then it'll get "absorbed", and the pics will match! And we won't ever be able to generate the "original outputs" anymore, not even with OldFish 1.1! That's crazy, you'll get images with the exact same metadata but different!
What's insane is the "original outputs" become impossible to generate, merged models will not reproduce them either (understandable, since at that point in time not even OldFish 1.1 makes them anymore), and the only way to get them again is to restart the UI from scratch!
I hope you can test it before someone else loads 1.1diff because it's a public space, but you can clone the space and see what I'm talking about, if you clone it and build it from scratch by restarting the space that's a sure way to force OldFish 1.1's "original outputs"! This has brought us all your OldFish variants so I'm happy this situation persists! 😄😄😄