<h3 id="comfyui-workflow">ComfyUI Workflow</h3><ul><li>Needed Custom Node:<ul><li>Git-Hub: <a target="_blank" rel="ugc" href="https://github.com/r-vage/ComfyUI-RvTools_v2">RvTools_v2 (Updated: 2.4.45)</a></li><li>Comfyui-Registry: <a target="_blank" rel="ugc" href="https://registry.comfy.org/nodes/comfyui-rvtools_v2">RvTools_v2</a>if you install the node with the manager choose nightly because i don't upload every change to comfyui registry. if you choose latest you'll install the latest version stored on comfy reg. nightly is the current version from git hub.</li></ul></li><li>if you get type mismatch errors, install the frontend version 1.26.0<ul><li>python_embeded\python.exe -m pip install comfyui-frontend-package==1.26.0</li></ul></li><li>if you want to use pulid or reactor you need to install numpy 1.26.4<ul><li>python_embeded\python.exe -m pip install numpy==1.26.4</li></ul></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/articles/15787/listing-links-resources">Resource Listing</a></li><li>How to install Triton &amp; Sage 2 and Flash Attention 2 Wheels (for e.g. Florence2)<ul><li>when using torch 2.7.x+cu128 use triton 3.3.1 (3.4 will cause issues)</li><li>python_embeded\python.exe -m pip install -U "triton-windows==3.3.1.post19"</li></ul></li></ul>Video to Video:<ul><li>If the control video group pose and depth are bypassed, the group control video aio is activated but the preprocessor in this group is also bypassed, the frames from the video are used unchanged... I'm just mentioning this. The prompt still allows you to make some changes.</li><li>it is possible to use the standard high/low models + vace modul or bypass model_2 in the model loader and generate video group to use the w2.1 fusionx vace model. in case you've bypassed model_2 you have to change the split step back to -1 ;)</li></ul>

Wan 2.1/2.2

<ul><li>image to video with a14b or 14b std</li></ul>

Files with SHA256: 3a2b1f987dc56e82e27a31508b98f78443a7d3b62e5d6ed9903d2443a3b090fb