READ EVERYTHING BEFORE USING!!!!If you even see this it's a miracle as my intro video is a test video to show you what not to do with this. Read it's prompt for further explanation.Going on...So this is a HEAVILY modified version of <a target="_blank" rel="ugc" href="https://www.patreon.com/aitrepreneur">https://www.patreon.com/aitrepreneur</a> Ultimate Wan 2.2 workflow that he offers for free on his patreon. If you want this in a basic format go to his patreon.So what is the difference?<ul><li>Sage Attention Nodes (A simple installer video to install from scratch is shown below.)</li></ul><ul><li>Florence (Auto prompt generation/enhancement)</li></ul><ul><li>New Video to End Frame I created myself (Ever generate a video then want to extend it but with a specific end frame? Well, you can with this. it takes your videos last frame and treats it like a first frame, then with you inserting a last frame it does the standard first last frame generation and stitches them together in the final product. But don't do it more than 3 or 4 times.)</li></ul><ul><li>Merge of Aitrepreneurs looping workflow and Ultimate workflow.</li></ul><ul><li>Color matching from I learned from GFrost in Start Frame to End Frame video generation (Honestly I don't see much difference with or without it but it's there for those who want it.)</li></ul><ul><li>Wan 2.2 5B Upscaling like what <a target="_blank" rel="ugc" href="https://www.patreon.com/c/aifuturetech">https://www.patreon.com/c/aifuturetech</a> had (Best video upscaler though it requires massive amounts of vram to do anything beyond 5 second videos. I recommend this section only if you are going for ultra realism and you have a minimum of 12 gigs of VRAM.)</li></ul><ul><li>Added sizing nodes for quick changes in loops between 720 and 480 resolutions. (I do allot of 480 instead of 720 and found it tedious how in Aitrepreneur had his looping section set to 720 and I had to resize each of the five nodes over and over. now you just change it in the beginning.)</li></ul><ul><li>Spacing for Loras. (Sounds silly but the original workflow was only spaced out enough for someone to use two or three loras? I spread it out so you can load it up with your collection. You know who you are. ;) )</li></ul>Basically, a whole lot of tedious little upgrades.Is it a mess? Yes. Does it work? Yes. Does it do more than most? Yes!The image section is nice, but I really didn't do much with it because Qwen is so much better. It's better to get a Qwen workflow running and generate your images on it to then start your videos off with or a good sdxl if you like illustrious images like the smooth series. Qwen Edit is the king for making consistent characters out of a single image to create your own Loras. Too bad Qwen takes so long to generate.INSTALLATION!!!aitrepreneurs auto installer is nice and all for the basic version he offers on his paywall patreon but it does not have Sage attention in it and when I tried installing Sage attention in his ComfyUI I never could get it to work. So good news for you....<div data-youtube-video><iframe width="640" height="480" allowfullscreen="true" autoplay="false" disablekbcontrols="false" enableiframeapi="false" endtime="0" ivloadpolicy="0" loop="false" modestbranding="false" origin playlist src="https://www.youtube.com/embed/CgLL5aoEX-s?rel=1" start="0"></iframe></div><a target="_blank" rel="ugc" href="https://www.youtube.com/@pixaroma">Pixaromas</a> auto installer is awesome and it's free! Yes! Just follow along in his video and create a new ComfyUI that DOESN'T require redownloading your models. It has an auto patcher to have multiple ComfyUI installations use the same models. Watch the whole video, it's awesome.So this is ALMOST a complete suite. The only thing that this does not do is lip syncing. BUT that is now available using Wan 2.1. Because it uses Wan 2.1 and not Wan 2.2 I did not include it in this one. But if you want lip syncing check out<div data-youtube-video><iframe width="640" height="480" allowfullscreen="true" autoplay="false" disablekbcontrols="false" enableiframeapi="false" endtime="0" ivloadpolicy="0" loop="false" modestbranding="false" origin playlist src="https://www.youtube.com/embed/CA-CQo_Q198?rel=1" start="0"></iframe></div>Tips:DO NOT USE UPSCALED VIDEOS FOR VIDEO TO ENDFRAME OR LONGER BASE VIDEO TO VIDEO!!!Use h265-mp4 with yuv420p10le or h254-mp4 with a yuv420p with a framerate of 24 for best results and best images for adding onto videosIf you use every section of this workflow you will have allot of previews playing at the same time. These eat up VRAM. I recommend you flip to a different window after hitting RUN or going to a section of the workflow where you don't have anything on the screen before hitting RUN to preserve that precious VRAM that without will slow down your generation time.If you join <a target="_blank" rel="ugc" href="https://www.patreon.com/aifuturetech">https://www.patreon.com/aifuturetech</a> for his lip sync workflow, use his workflow in a clean installation separate from this one. His VRAM purges and Ollama have issues with this workflow. I swapped them out on my own comfyui build but I don't post anything that someone else is charging money for that they create. That's why I don't mind posting this workflow, because Aitrepreneur offered his original one for free.Careful using the video to end frame too much. If you run allot of it the start of the video will slowly degrade. You can see that in the test video I uploaded where in the beggining the big buff guy has a reddish tint to his skin instead of brown skin. This is because every time you run it, it decompresses the images, then when it is finished it recompresses them. Doing it once or twice is fine, but constant will cause color changes and other degrading factors. If you are going to do allot of start to end frames, it's better to splice the pieces in a movie mixer of some kind.LOW VRAM USERS, the trick is getting the model that works for you. The smaller the VRAM, the smaller the model. I run an RTX 4070TI Super with 16 gigs of VRAM, 64 gigs of regular ram and run a Q5 K_M or Q5 K_S version of WAN 2.2. I also noticed I can do over 121 frames at a go at 480X854 but the moment I try 720X1280 I need to lower my frame count to around 81 frames. If you have lower VRAM you will have to juggle model size vs frame count vs resolution. The larger the model, the lower the frame count or resolution. Obviously the inverse is true as well. If you just want to make anime and cartoon like videos, 480X854 then upscale is just fine.I can't help, I showed you what I did and how I got it running. I don't know your system.Post whatever you want here for pics and vids. I noticed the more horrible videos are uploaded to something, the more people download it. So show me the worst or best you can do with this.I took this down earlier because after uploading it I saw a couple flaws in the first to last frame, then after running it, it looked like hell. So I took it down until I could figure out why. Turns out that my kids messed with the settings. Make sure your combined video is h265-mp4 with a yuv420p10le for best results and a frame rate of 24.Beyond that, butcher this thing like I have done. Make it bigger. Add more stuff on it. Make a bigger one so that this one is forgotten.

Files with SHA256: 45ace1e3b49fe6c1fb72e8ba68470670412424266da2bc17add341275a400afc