tawsLoRa 2.1 char loras works on 2.2. btw if I blur faces of some poses like these ones will this lora won't affect other lora's face? check this lora "Character Friendly Body Types - FLUX - Candice-v1-STRONG | Flux LoRA | Civitai".

idk how he prepare dataset and caption it.

@tawsLoRa Training Wan 2.2 character loras is easy and fun https://civitai.com/articles/17740

I downloaded and tried dragging and dropping your pinned sample clips into comfyui and didn't work. No metadata in the video.

@tawsLoRa it looks like there is no workflow metadata in those samples. Either it was never there or maybe civitai is removing it now

@tawsLoRa Much appreciated, thank you

tawsLoRa I see ill try that, thanks

If it says T2V then it wont work in I2V

@fox23vang226 wrong. It does.

@m0no1 I'll try and report back

works great with I2V!

I try with both diffusion and GGUF models (NSFW versions), combined with WanImage or PainterI2Vadvanced and basically it works even if i didn't work with all images and it seems it likes a lot pictures with the character seated or half-lenght framed. With other poses i got mixed results (something ok, something horrible). I added the LORA "Fingering" and i was able to get sometimes funny outputs. Having said that, i was unable to get video with the character wrapped by tentacles. And prompting to produce more than 2 tentacles it became very hard to manage via prompt and reach consisten and repeatable results.

@tawsLoRa But they have the exact same hash so they are quite literally the same file no? Meaning if you just loaded the "High" lora twice it would be the same as downloading both high and low, saving space

@tawsLoRa To be clear, you may have made a mistake. If not, then you uploaded the exact same file for high and low Wan2.2-T2V_Tentacled_v1_*_14B.safetensors. They have the same md5 chucksums, meaning they are 100% the same file with different names.

The high and low for Wan2.2-I2V-Tentacled-v2.1-*-tawsLora.safetensors have difference checksums, meaning they are indeed different files.

@tawsLoRa I just realize that I was checking the high lo for v1, not 1.2. For 1.2 T2V high and low have different checksums.

@tawsLoRa Mmm, maybe something happened during the merge, or you grabbed the same file twice then? The hash of the high and low versions are the same, which means they are completely byte-for-byte identical.

I calculated the hash for both high and low files on my own machine as well, just to see if the identical hashes listed here on Civit was some sort of weird bug, but both files are identical on my machine too (and the sha256sum from my machine agrees with the CivitAI value.)