ComfyUI Extension: I-Ching-Informed Adaptive ODE Scheduler

Authored by eddyhhlure1Eddy

Created

Updated

225 stars

The ode v2 version is a creative sampler characterized by low contrast, near-realistic quality, and artistic creation. It can only be used with high noise for 6 steps, low noise for 2 steps, and a middle split step of 4 steps. It requires a unique LoRA, which is currently in training and undergoing refinement, and is only in the testing phase. This version offers multiple options for you to choose from, and my recommendation is to go with robust-smooth or even lower. Stay tuned for the official release.

Custom Nodes (0)

    README

    👇ode+ 20251027

    This update's ODE+ is based on the SA ODE version with 8-step optimization parameter fine-tuning. It eliminates most of the native SA ODE drawbacks at the original 8 steps. No correction was applied in the corrector section, but a PC2 path was used to implement a similar correction strategy. Significant improvements were also made in LoRA training, resulting in a cleaner image output.

    SA ODE <img width="283" height="379" alt="image" src="https://github.com/user-attachments/assets/bedb095e-5680-4c5b-810e-1b4f2d4dc308" />

    ODE+PC2+FW3 <img width="300" height="375" alt="image" src="https://github.com/user-attachments/assets/e831c8d8-7ca8-47a1-a64b-d4ad8ded5a93" />

    👇odev2

    I-Ching-Informed Adaptive ODE Scheduler

    DOI License Stars

    Paper: IChingWuxingScheduler: A Novel Noise Scheduling Algorithm Based on Traditional Chinese Philosophy for Diffusion Models 2025 21/10 add iching schedulers new odev2 with lora:https://huggingface.co/eddy1111111/WAN22.XX_Palingenesis/tree/main

    output

    \The ode v2 version is a creative sampler characterized by low contrast, near-realistic quality, and artistic creation. It can only be used with high noise for 6 steps, low noise for 2 steps, and a middle split step of 4 steps. It requires a unique LoRA, which is currently in training and undergoing refinement, and is only in the testing phase. This version offers multiple options for you to choose from, and my recommendation is to go with robust-smooth or even lower. Stay tuned for the official release.

    Original source from kijai's ComfyUI-WanVideoWrapper https://github.com/kijai/ComfyUI-WanVideoWrapper/tree/main/wanvideo/schedulers

    Modified and optimized by eddy

    new change: humo_lcm `keep face

    👇humo

    <img width="852" height="415" alt="image" src="https://github.com/user-attachments/assets/4125f046-0041-4769-aaaf-5c2ad9478126" /> <img width="849" height="426" alt="image" src="https://github.com/user-attachments/assets/70d3f33c-6cd8-42d5-ac17-f1bf19d12604" />

    tigger:“女人一边用模特步走路，一边对着镜头说话，走在夜晚大街上” lv_0_20251021160044 mixbase:https://huggingface.co/eddy1111111/humo