diff --git a/README.md b/README.md index 626914e..feaf5f2 100644 --- a/README.md +++ b/README.md @@ -13,7 +13,7 @@ A ComfyUI custom node for loading and applying LoRA (Low-Rank Adaptation) to Nun **Currently under development and testing. Debug logs are being output extensively. This does not affect functionality.** -> Latest release: [v2.5.1 Release Notes](https://github.com/ussoewwin/ComfyUI-QwenImageLoraLoader/releases/tag/v2.5.1) +> Latest release: [v2.5.2 Release Notes](https://github.com/ussoewwin/ComfyUI-QwenImageLoraLoader/releases/tag/v2.5.2) > ## Source diff --git a/RELEASE_NOTES/changelog.md b/RELEASE_NOTES/changelog.md index 89fc83c..c244348 100644 --- a/RELEASE_NOTES/changelog.md +++ b/RELEASE_NOTES/changelog.md @@ -1,3 +1,10 @@ + + + + + +
EN中文
+ ### v2.5.2 (latest) - **Fixed**: Suppressed the cosmetic ComfyUI startup warning `WARNING: Potential Error in code: Torch already imported, torch should never be imported before this point.` ([Issue #53](https://github.com/ussoewwin/ComfyUI-QwenImageLoraLoader/issues/53)). The warning fires because the mandatory `apply_rotary_emb` prestartup shim must import `comfy.ldm` modules that import `torch` at module level (all CUDA env setup already ran, so it is harmless). A one-shot root-logger filter installed early in `prestartup_script.py` drops only that single message and lets every other log through. Opt out with `QWENIMAGE_SUPPRESS_TORCH_WARNING=0`. - **Technical Details**: See [v2.5.2 Release Notes](https://github.com/ussoewwin/ComfyUI-QwenImageLoraLoader/releases/tag/v2.5.2) for complete explanation diff --git a/RELEASE_NOTES/v2.5.0.md b/RELEASE_NOTES/v2.5.0.md index 9166e92..2d69ab0 100644 --- a/RELEASE_NOTES/v2.5.0.md +++ b/RELEASE_NOTES/v2.5.0.md @@ -1,3 +1,10 @@ + + + + + +
EN中文
+ # Complete Explanation of the ControlNet Issue in Nunchaku Qwen Image ## 1. Complete Explanation of the Issue diff --git a/zhmd/README.md b/zhmd/README.md index 02634f0..5fcdcd9 100644 --- a/zhmd/README.md +++ b/zhmd/README.md @@ -13,7 +13,7 @@ **目前正在开发和测试中。正在大量输出调试日志。这不影响功能。** -> 最新版本: [v2.5.1 发行说明](v2.5.1.md) +> 最新版本: [v2.5.2 发行说明](https://github.com/ussoewwin/ComfyUI-QwenImageLoraLoader/releases/tag/v2.5.2) > ## 来源 diff --git a/zhmd/changelogzh.md b/zhmd/changelogzh.md index da2cbaa..d64c164 100644 --- a/zhmd/changelogzh.md +++ b/zhmd/changelogzh.md @@ -1,4 +1,15 @@ -### v2.5.1 (最新) + + + + + +
EN中文
+ +### v2.5.2 (最新) +- **已修复**: 抑制了 ComfyUI 启动时的无害警告 `WARNING: Potential Error in code: Torch already imported, torch should never be imported before this point.`（[Issue #53](https://github.com/ussoewwin/ComfyUI-QwenImageLoraLoader/issues/53)）。该警告出现，是因为必需的 `apply_rotary_emb` prestartup 垫片必须导入在模块层级导入 `torch` 的 `comfy.ldm` 模块（此时所有 CUDA 环境设置均已完成，因此无害）。在 `prestartup_script.py` 早期安装的一次性根日志过滤器仅丢弃这一条消息，其余所有日志照常输出。可通过 `QWENIMAGE_SUPPRESS_TORCH_WARNING=0` 关闭该抑制。 +- **技术详情**: 请参阅 [v2.5.2 发行说明](https://github.com/ussoewwin/ComfyUI-QwenImageLoraLoader/releases/tag/v2.5.2) 获取完整说明 + +### v2.5.1 - **已添加**: 通过 `Krea2ControlNetLoraLoader` 以及 `NunchakuQI&ZITDiffsynthControlnet` 中的 Krea2 路由，新增 Krea2 depth ControlNet LoRA 支持。从 `controlnet` 文件夹加载 Krea2 depth controlnet-lora 文件（例如 `krea2-depth-control-lora.safetensors`），将其 `MODEL_PATCH` 输出连接到 controlnet 节点的 `model_patch` 输入，并在 Krea2 / SingleStreamDiT 模型上应用 depth 条件。 - **技术详情**: 请参阅 [v2.5.1 发行说明](v2.5.1.md) 获取完整说明