Document PiFlow conditioning and reference-tail mask limits

README.md

@@ -40,6 +40,8 @@ hf download t8star/Kandinsky-Comfy --local-dir ComfyUI/models --include "diffusi
 
 The repository folders match ComfyUI's model folders. Keep `audio_vae/bigvgan_vocoder/config.json` beside `bigvgan_generator.pt`. Import a [workflow](https://github.com/T8mars/Comfyui-Kandinsky-T8/tree/main/example_workflows), select the model in **UNETLoader**, and use `weight_dtype=default`.
 
+PiFlow requires CFG=1, denoise=1 and one global, full-range conditioning; masks are limited to the Image workflow's clean reference tail.
+
 ## Requirements and validation
 
 Requires **ComfyUI 0.39.0+**, **comfy-kitchen 0.2.37+**, the T8 nodes and an NVIDIA CUDA GPU with INT8 Tensor Core support. Default output is **864×480, 121 frames, 24 fps**, with audio. Pro distill was tested using dynamic VRAM/offloading on **RTX 5090 Laptop 24 GB / 64 GB RAM**.

README.zh-CN.md

@@ -22,6 +22,8 @@ hf download t8star/Kandinsky-Comfy --local-dir ComfyUI/models --include "diffusi
 
 仓库目录与 ComfyUI 模型目录一致。BigVGAN 的 `config.json` 与 `bigvgan_generator.pt` 须放在同一目录。导入[工作流](https://github.com/T8mars/Comfyui-Kandinsky-T8/tree/main/example_workflows)，在 **UNETLoader** 中选择模型，保持 `weight_dtype=default`。
 
+PiFlow 使用 CFG=1、denoise=1 和一条覆盖全程的全局条件，遮罩仅支持 Image 工作流的干净参考尾帧。
+
 ## 环境与验证
 
 需要 **ComfyUI 0.39.0+**、**comfy-kitchen 0.2.37+**、T8 节点及支持 INT8 Tensor Core 的 NVIDIA CUDA 显卡。默认生成 **864×480、121 帧、24 fps** 带音轨视频。Pro 蒸馏版在 **RTX 5090 Laptop 24 GB / 64 GB RAM** 上通过动态显存与卸载运行。