https://huggingface.co/FlibusteCentrifuge/WanAlternateKing/tree/main for those who want to download it on Hugging Face (more stable for long download if your internet is not fast) I nammed it WanAlternateKing but it is the same model.
I decided to remove version 2.0. There was a tendency by Wanking V2.0 at creating too much blurry video (I was hoping it was just a bad seed luck in my few tests, but it consistently continued to produce these every now and then, probably too much loras added to the merge created this instability, I'll continue to test on my side and see if I can make a better version, by reducing loras and/or or lowering the strenght of the merged loras, but I can't give you any promises that I will put another version back, anyway, thanks for those who tried V2, sorry for the waste of time, and have fun.
V2 was quite good, borderline amazing in some cases when NO LORAS was added, but there is no point in having a model that can't handle loras, in the end it's just a waste of time so I removed it for this reason.
What's better than Wan2.1 720p I2V ? Not much you may say... But WanKing is here to (maybe) change your mind about that!
this model is the result of a very raw, Wan2.1 720p I2V + 6 or 7 loras merged at 50% in it, so it's basically a variant of Wan2.1 720p I2V fp8 model, non GGUF.
And WanKing users are called Wanke.... oh you probably guessed yourself...
This is a test merge, I didn't really put it to the test, I did make few videos and it was superior output in most case than Wan2.1 720p I2V fp8 model, but it's not night and day, just test for yourself, maybe create a GGUF version if you have the know-how, or just don't even bother downloading it, I can't say there is ANY use for this model (and it's then probably a waste of 16GB of space) but it don't seems to be of less quality than base model so why not give it a try ?
Lora Merged (max strenght for each Lora was 0.5 and merged at 50% so I guess that 0.25 for actual max strenght in the model ? but it's probably more complicated that that)
WtSSTaDaMiT - Translucent Dress
https://civarchive.com/models/1378657/wtsstadamit-translucent-dress?modelVersionId=1557740
by https://civarchive.com/user/switch_ai
The Walking Back - Wan I2V 14B
https://civarchive.com/models/1363473?modelVersionId=1544772
by https://civarchive.com/user/ivever
Bouncy Walk - Wan I2V 14B
https://civarchive.com/models/1361346/bouncy-walk-wan-i2v-14b?modelVersionId=1537915
by https://civarchive.com/user/ai_build_art
Wan 2.1 Deepthroat, Blowjob, Fellatio, Oral, Lora, T2V, I2V
by https://civarchive.com/user/aipinups69
Bouncing Boobs - Wan I2V 14B
https://civarchive.com/models/1343431/bouncing-boobs-wan-i2v-14b?modelVersionId=1517164
by https://civarchive.com/user/ai_build_art
Wan 2.1 Big Natural Breasts - Saggy Tits, Huge Breasts, Lora, T2V, I2V
by https://civarchive.com/user/aipinups69
Boob grab enhancer (Pine Enhancement suite)
Description
Nothing to say, except it's 0.5 GB smaller than base one, no idea why, but heh, it's already that less to load in Vram
FAQ
Comments (11)
Nice, I was wondering about trying something like this since full fine-tuning would be insanely expensive. Could you share which loras you merged in so we don't end up using those on top and ending up with an effective weight >1? Thanks malaka.
neat , going to try it out. question, how does one merge wan loras? appreciate any pointers. Thanks!
Is there a comfyui workflow for merging WAN LORAs into models or did you use something else to do it?
I dunno, maybe Wankinian? Wankinese? Wankingite?
I mean, you wouldn't call a resident of london a loner, would you.
So I can use it for 480P I2V?
Chance of getting GGUF for us VRAM poors?
Is this just about eating phallic snacks?
It's looking good. Can you list the new loras you put into v2? Was Wan_Cumshot one of them? Detailz-Wan? I just want to make sure I don't add unnecessary loras into VRAM. Thanks!
Hi @NoArtifact
Please upload both fp16 and fp8 on your huggingface I will convert for you into GGUF. fp16 is better since ít quality higher than fp8 and I can convert into Q8, Q6 or Q4 to save Vram no need to convert fp8
How do I use this? I placed it in the checkpoints folder but it does not show up with the other checkpoint selections. Only the official Wan models are showing up. Loras folders are working fine.
For those that want to create a better merge, there is now a merge Wan2.1 node in comfyUI latest version (maybe it was already there when I did mine, but I didn't see it, I used a simple merge model node) that allow apparently much more detailed merge with various block and weight per block, I have no idea how it work and I didn't tested, but I'm pretty sure this could yeld a much better result than the very basic merge I made, just for those interested.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.