(if you make things with this. please post them below id like to see how other people are using things its taken me a lot of work to get where i feel this is user friendly and optimized. all i want is to see your creations for the time :D happy generating)I think i've finally done it. gotten the best looking generations while being reasonably quick..this version i did test the lightx2v base models and ultimately decided that using the lightx2v loras with wan base models was the better result.this version utilizes the Triple Ksampler node (which is amazing)it also has OPTIONAL upscaling.if you want to upscale just unbypass the entire upscale group and push run.if you dont.. bypass the entire groupThis version also has THREE power lora loaders because i find that not running loras on your first pass can massively effect video generation so you have have to load your high loras into the wf twice! ( DO NOT ADD DISTILLATION LORAS INTO THE POWER LOADERS)despite all this i am generating VERY good looking videos at 512, 768, and 1024 resolution.512 is obviously the fastest. i reccommend sticking to 512 if you are upscaling.. and going no higher than 768x768 if you arent. i also designed this to produce square videos which pad your input image to not crop out parts of it. (the image resize node by input image should match the wan video size)there are notes included in the workflow to help you.======================================================after testing for a while with v3 i could never get the first few frame blur out. so i added an actual upscale model and ultimate sd upscale (4xNomos8kDAT)this produces way better results but the generation times are significantly longer.you can slightly reduce this by doing upscale of 1.5tile of 768x768instead of upscale of 2.and lowering steps to 6 from 4 on upscale modelworking now on just a triple sampler refinement stage to see results here==========================================================Version 3(show me what you make please)added a second pass for upscaling into a 3rd ksampler this creates much more detailed final outputs. it does add to generation time so if you just want to bang out some videos keep using v1 or v2.swapped to these distillation loras: Wan_2_2_I2V_A14B_HIGH_lightx2v_4step_lora_v1030_rank_64_bf16 on the high noiselightx2v_I2V_14B_480p_cfg_step_distill_rank256_bf16 on the low noiseon the 3rd sampler (the klownshark one) you might consider lowering cfg to around 1.5 and denoise somewhere between 0.25 and 0.4. i think it will solve the first few frames of renders being blurry and having sometimes added artifacts. doing some test runs as we speak for optimal settings here========================================================Version 2 (PLEASE POST WHAT YOU MAKE)added a resize image node to scale images to size of video to prevent unwanted croppingtested different settings to find optimal steps for fast generation / good motion(22 total steps 12 high / 10 low seems to be the sweet spot) i have also tested the MoE Ksampler but ultimately decided that 2 samplers is the better route despite MoE producing decent results it does add 2 to 3 minutes to generation time.removed all of my loras and left only Dr3amly pre-loaded so you wont get so many errors.. clear labels as to where to put your own lorasnotes for suggested settings and defaults near settings you can change / tweak.=================================================Wan 2.2 I2V with loras for NSFW loaded First off I have to give credit to The guy who made the Wan 2.2 I2V 60FPS workflow with a bunch of upscaling. This is just a modified version of that with some NSFW loras (All Downloaded from Civtai Website) Preloaded into it. and i chained in a couple power loaders. I chained the Lightning 4 Step Loras with a couple of power lora loaders for your custom loras to be added in a simple place. The ones already loaded that you do not like / do not have / cant get.. just delete them. It is also clearly labeled where you should load each high / low noise if you wish to add and experiment with your own loras. I am still Fairly new to this so please if you see ways to improve this do not hesitate to let me know. The idea was basically fast generation without all the upscaling and whatnot. I find it produces some pretty great results and i like the fact that i can load in multiple loras and toggle them on and off at will with the power loaders. My system is 32gb of ram running an RTX 4090 and i generate videos rather quickly with this workflow

Simple Wan 2.2 I2V NSFW WorkFlow (updated to V4.0FINAL)

I think i've finally done it. gotten the best looking generations while being reasonably quick.. this version i did test the lightx2v base models and ultimately decided that using the lightx2v loras with wan base models was the better result. this version utilizes the Triple Ksampler node (which is amazing) it also has OPTIONAL upscaling. if you want to upscale just unbypass the entire upscale group and push run. if you dont.. bypass the entire groupThis version also has THREE power lora loaders because i find that not running loras on your first pass can massively effect video generation so you have have to load your high loras into the wf twice! despite all this i am generating VERY good looking videos at 512, 768, and 1024 resolution. 512 is obviously the fastest. i reccommend sticking to 512 if you are upscaling.. and going no higher than 768x768 if you arent. i also designed this to produce square videos which pad your input image to not crop out parts of it. (the image resize node by input image should match the wan video size) there are notes included in the workflow to help you.

Files with SHA256: 3715fe1575f26f1be5b79a1428b644807173b904366dc1b28cb306526ee49a33