benini much better than wan2.2
In my tests, benini is much better than wan2.2. R2V is a lot more convenient, and benini can use wan2.2 loras as well , even NSFW. I wonder why you fall back to play wan2.2 a lot
Well, the real reason is storage. we all have 8TB of free storage here and publishing models here allows me to delete the local files and do something else, or restore them when I need them again. so it is not fall back but deleting local wan2.2 files :)
When it comes to NSFW, this one and the LightX2V loras here (https://huggingface.co/rzgar/Wan2.2_LightX2V_4Step_Uncensored) are better than the Bernini Enhancers (but not tuned for Bernini-R).
This one exists because I did a lot of experiments to find a way to achieve it, and I think I’ve found it (or am close to it): one LoRA for both models without side effects and with over 80% accuracy.
that lightx2v lora is not only 4step lora, but nsfw I2V and T2V loras in one. Next time I only download these models and I have everything that one has to offer.
thanks for your reply. i will try your lightx
LightX2V_4Step_Uncensored too blur in bernini, both weight 1.
LightX2V_4Step_Uncensored too blur in bernini, both weight 1.
Those loras won't work with Bernini :)
Ok , i tried other nsfw loras from civitai , works good with your Bernini 4 step lora. Thanks for your Bernini 4 step lora, really helps. we can only use wan2.1 lora with 3+1.5 before
Ok , i tried other nsfw loras from civitai , works good with your Bernini 4 step lora. Thanks for your Bernini 4 step lora, really helps. we can only use wan2.1 lora with 3+1.5 before
You’re welcome. ya I had the same problem. Wan2.1 Lightx2v just won’t work nicely with Wan2.2. It works only for portrait, walking, or static shots, beyond that, it fails. It also trashes the high-noise motion noises because they aren’t trained together.
this 260412 version wan2.2 lightx2v has really good and best details than others.It is trained by 720p videos. Wan_2_2_I2V_A14B_LOW_lightx2v_4step_lora_260412_rank_64_fp16.safetensors
https://huggingface.co/Kijai/WanVideo_comfy/tree/main/LoRAs/Wan22_Lightx2v
this 260412 version wan2.2 lightx2v has really good and best details than others.It is trained by 720p videos. Wan_2_2_I2V_A14B_LOW_lightx2v_4step_lora_260412_rank_64_fp16.safetensors
https://huggingface.co/Kijai/WanVideo_comfy/tree/main/LoRAs/Wan22_Lightx2v
yes, those are extracted from the full model https://huggingface.co/lightx2v/Wan2.2-Distill-Models/tree/main , I've tried them much better than seko versions