The idea with this workflow is to allow a smooth transition between characters with a static background. It's not perfect, but it's a good start, I think. The full generation for 48 frames took 17gb of VRAM and 50 minutes for my 3090. If you're trying to save VRAM, I'd just skip the hires section of the workflow.Full credit to Latent Vision for the base workflow:<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/jc65n-viEEU" start="0"></iframe></div>Additional things I'd like to try<ul><li>IC-light to increase lighting consistency (will this work if character/camera rotates?)</li><li>IPAdapter for faces</li></ul>One last note: the Limitless Vision model is fine, but I think this would probably work better with other models like Realistic Vision or Dreamshaper. The problem I ran into was that those models don't seem to "understand" the IPAdapter images I was feeding in. More experimentation is needed!If you are missing any models, see below for a partial list. You will also need the IPadapter/clipvision models &amp; openpose controlnet<ul><li>Rename to controlGIF_normal.ckpt in /models/controlnet: <a target="_blank" rel="ugc" href="https://huggingface.co/crishhh/animatediff_controlnet/resolve/main/controlnet_checkpoint.ckpt?download=true">https://huggingface.co/crishhh/animatediff_controlnet/resolve/main/controlnet_checkpoint.ckpt?download=true</a></li><li>Put in /models/animatediff_models: <a target="_blank" rel="ugc" href="https://huggingface.co/wangfuyun/AnimateLCM/resolve/main/AnimateLCM_sd15_t2v.ckpt">https://huggingface.co/wangfuyun/AnimateLCM/resolve/main/AnimateLCM_sd15_t2v.ckpt</a></li><li>Put in /models/loras: <a target="_blank" rel="ugc" href="https://huggingface.co/latent-consistency/lcm-lora-sdv1-5/tree/main">https://huggingface.co/latent-consistency/lcm-lora-sdv1-5/tree/main</a></li><li>Put in /models/loras: <a target="_blank" rel="ugc" href="https://huggingface.co/guoyww/animatediff/blob/main/v3_sd15_adapter.ckpt">https://huggingface.co/guoyww/animatediff/blob/main/v3_sd15_adapter.ckpt</a></li></ul>Tip jar: <a target="_blank" rel="ugc" href="https://www.patreon.com/blankage">https://www.patreon.com/blankage</a>

AnimateDiff Workflow: IPAdapter Transformations

Version 2 borrows heavily from <a rel="ugc" href="https://civitai.com/models/254494/animatediff-morphing-workflow-prompt-travel-controlnet-upscale-and-face-restore-reactor">Katsumend’s workflow</a>, so huge thanks to him for sharing that! The biggest change from version 1 is switching to LCM and using the noise replacement layer. Again, thanks to Katsumend who I stole that from. I also removed the background cutout part of the workflow for now so I can focus on the actual generation settings. I’m not completely happy with the way the colors are looking, but I decided to just release for now.Future things that need testing/optimizing<ul><li>Beta schedule</li><li>Step count</li><li>Not using LCM</li><li>A way to use if statements to make adjusting key frame count easier</li></ul>

Files with SHA256: 056987f81f84f9a39ba3c9787b88f9bf1d0868263b6a1bd2abd1124336dc1b44