PlagueKind - Minimax H3 Sparse Attention - Refmod - LTX Workflow - Eros | Sulphur Compatible | FaceID

⚠ 注意：チェーン全体の使用時にRTX VSRは避けるよう作者が注意（96GB級RAMでない場合）

### V12 Notes - UPDATE NODEPACK TO V1.6.1 AND COMFYUI TO 0.44.0 or newer, and pytorch 2.13cu130 or newer. - New media loader all images and videos top layer now. - New upscaler, issues now fixed. - Switched to my quant of pika singularity prism beta7. which is only 0.5% off of int8HQ, makes the same video for 5gb less. - Added Lokken Text encoder, same deal as my model, amazing quality. - Added lynnreal light vae, I didn't add sol vae because it was too soft. - Recommended 5-12 steps, with default settings. - Updated my H3 node to now support 3 videos with audio, and 3 extra audio slots as well. - You can trim videos, add extra audio refs, and create refmods in the subgraph. - Shift settings are inside the sampler subgraph incase you need to disable it for using V11 settings, or to adjust for other turbo situations. - This is by far the fastest WF with least resource consumption I've made.