From 3e87ca260ec7123f80446ca51399693f190a63ec Mon Sep 17 00:00:00 2001 From: coderfromthenorth93 <213232275+coderfromthenorth93@users.noreply.github.com> Date: Tue, 4 Nov 2025 20:51:34 -0500 Subject: [PATCH 1/3] Add/update changelog/index.mdx --- changelog/index.mdx | 36 +++++++++++++++++++++++++++++++++++- 1 file changed, 35 insertions(+), 1 deletion(-) diff --git a/changelog/index.mdx b/changelog/index.mdx index d84b4d8d..e5ee11d5 100644 --- a/changelog/index.mdx +++ b/changelog/index.mdx @@ -4,6 +4,41 @@ description: "Track ComfyUI's latest features, improvements, and bug fixes" icon: "clock-rotate-left" --- + + +**Memory & Performance Optimizations** +- Introduced **Mixed Precision Quantization System** for optimized model loading and memory usage +- Significantly accelerated model offloading using pinned memory with automatic low-RAM hardware detection +- Added **RAM Pressure Cache Mode** for intelligent memory management under resource constraints +- Enhanced FP8 operations with reduced memory usage and fixed torch.compile performance regressions +- Improved async offload speeds and resolved race conditions that could cause workflow corruption + +**New Execution Features** +- **ScaleROPE Node**: Added support for rope scaling on WAN and Lumina models, enabling better resolution adaptation +- Enhanced subgraph execution allowing multiple runs within single workflows +- Improved caching system with proper handling of bytes data and None outputs + +**Portable & Driver Support** +- Updated stable portable to CUDA 1.3.0 with Python 3.13.9 +- Enhanced NVIDIA driver compatibility checks with user-friendly update prompts for portable installations +- Optimized Windows pinned memory limits to prevent system issues + +**API Node Enhancements** +- Migrated multiple API nodes (Luma, Minimax, Pixverse, Ideogram, StabilityAI, Pika) to new V2 client architecture +- Extended LTXV API nodes with 12s-20s duration options for longer video generation +- Fixed img2img operations in DALL-E 2 API node +- Enhanced Rodin3D nodes to return proper relative paths for 3D model outputs + +**Schema Conversion (V3)** +- Converted Recraft, Hypernetwork, and OpenAI nodes to V3 schema for improved consistency and maintainability + +**Updates** +- Embedded documentation updated to v0.3.1 +- Workflow templates updated to v0.2.11 +- Improved model loading status messages for better user feedback + + + **API Nodes** @@ -76,7 +111,6 @@ icon: "clock-rotate-left" - Fixed WAN2.2 cache VRAM leak - **API Nodes** From 54a93669f503577a9e6127f366c03c2a8c4430f3 Mon Sep 17 00:00:00 2001 From: coderfromthenorth93 <213232275+coderfromthenorth93@users.noreply.github.com> Date: Tue, 4 Nov 2025 20:51:34 -0500 Subject: [PATCH 2/3] Add/update zh-CN/changelog/index.mdx --- zh-CN/changelog/index.mdx | 36 +++++++++++++++++++++++++++++++++++- 1 file changed, 35 insertions(+), 1 deletion(-) diff --git a/zh-CN/changelog/index.mdx b/zh-CN/changelog/index.mdx index 8e471563..7d5f2ae2 100644 --- a/zh-CN/changelog/index.mdx +++ b/zh-CN/changelog/index.mdx @@ -1,9 +1,43 @@ --- title: "更新日志" -description: "追踪ComfyUI的最新功能、改进和错误修复" +description: "跟踪ComfyUI的最新功能、改进和错误修复" icon: "clock-rotate-left" --- + + +**内存和性能优化** +- 引入**混合精度量化系统**，优化模型加载和内存使用 +- 使用固定内存显著加速模型卸载，支持自动低RAM硬件检测 +- 新增**RAM压力缓存模式**，在资源受限条件下智能管理内存 +- 增强FP8操作，减少内存使用并修复torch.compile性能回归问题 +- 改进异步卸载速度，解决可能导致工作流损坏的竞争条件 + +**新的执行功能** +- **ScaleROPE节点**：为WAN和Lumina模型添加rope缩放支持，实现更好的分辨率适配 +- 增强子图执行功能，允许在单个工作流中多次运行 +- 改进缓存系统，正确处理字节数据和None输出 + +**便携版和驱动支持** +- 将稳定便携版更新至CUDA 1.3.0和Python 3.13.9 +- 增强NVIDIA驱动兼容性检查，为便携安装提供用户友好的更新提示 +- 优化Windows固定内存限制以防止系统问题 + +**API节点增强** +- 将多个API节点（Luma、Minimax、Pixverse、Ideogram、StabilityAI、Pika）迁移到新的V2客户端架构 +- 扩展LTXV API节点，支持12秒-20秒时长选项，实现更长的视频生成 +- 修复DALL-E 2 API节点中的img2img操作 +- 增强Rodin3D节点，为3D模型输出返回正确的相对路径 + +**架构转换（V3）** +- 将Recraft、Hypernetwork和OpenAI节点转换为V3架构，提高一致性和可维护性 + +**更新** +- 内嵌文档更新至v0.3.1 +- 工作流模板更新至v0.2.11 +- 改进模型加载状态消息，提供更好的用户反馈 + + **API 节点** From 4ea77274787e2486a18ff2558b29c8673c0770bc Mon Sep 17 00:00:00 2001 From: ComfyUI Wiki Date: Wed, 5 Nov 2025 18:04:00 +0800 Subject: [PATCH 3/3] Simplify changelog --- changelog/index.mdx | 32 ++++++++++++-------------------- zh-CN/changelog/index.mdx | 35 ++++++++++++++--------------------- 2 files changed, 26 insertions(+), 41 deletions(-) diff --git a/changelog/index.mdx b/changelog/index.mdx index e5ee11d5..1f698a08 100644 --- a/changelog/index.mdx +++ b/changelog/index.mdx @@ -6,36 +6,28 @@ icon: "clock-rotate-left" -**Memory & Performance Optimizations** -- Introduced **Mixed Precision Quantization System** for optimized model loading and memory usage -- Significantly accelerated model offloading using pinned memory with automatic low-RAM hardware detection +**Performance & Memory Optimizations** +- Introduced **Mixed Precision Quantization System** for optimized model loading - Added **RAM Pressure Cache Mode** for intelligent memory management under resource constraints -- Enhanced FP8 operations with reduced memory usage and fixed torch.compile performance regressions -- Improved async offload speeds and resolved race conditions that could cause workflow corruption +- Accelerated model offloading using pinned memory with automatic low-RAM hardware detection +- Enhanced FP8 operations: reduced memory usage and fixed torch.compile performance regressions +- Improved async offload speeds and resolved race conditions -**New Execution Features** -- **ScaleROPE Node**: Added support for rope scaling on WAN and Lumina models, enabling better resolution adaptation +**New Nodes & Execution Features** +- **ScaleROPE Node**: Rope scaling support for WAN and Lumina models - Enhanced subgraph execution allowing multiple runs within single workflows - Improved caching system with proper handling of bytes data and None outputs -**Portable & Driver Support** -- Updated stable portable to CUDA 1.3.0 with Python 3.13.9 -- Enhanced NVIDIA driver compatibility checks with user-friendly update prompts for portable installations -- Optimized Windows pinned memory limits to prevent system issues - **API Node Enhancements** -- Migrated multiple API nodes (Luma, Minimax, Pixverse, Ideogram, StabilityAI, Pika) to new V2 client architecture -- Extended LTXV API nodes with 12s-20s duration options for longer video generation -- Fixed img2img operations in DALL-E 2 API node -- Enhanced Rodin3D nodes to return proper relative paths for 3D model outputs - -**Schema Conversion (V3)** -- Converted Recraft, Hypernetwork, and OpenAI nodes to V3 schema for improved consistency and maintainability +- Migrated API nodes to V3 client architecture: Luma, Minimax, Pixverse, Ideogram, StabilityAI, Pika, Recraft, Hypernetwork, OpenAI +- Extended LTXV API nodes with 12s-20s duration options +- Fixed img2img operations in DALL-E 2 node +- Enhanced Rodin3D nodes to return proper relative paths **Updates** - Embedded documentation updated to v0.3.1 - Workflow templates updated to v0.2.11 -- Improved model loading status messages for better user feedback +- Fixed Windows pinned memory allocation issues diff --git a/zh-CN/changelog/index.mdx b/zh-CN/changelog/index.mdx index 7d5f2ae2..5cd781c4 100644 --- a/zh-CN/changelog/index.mdx +++ b/zh-CN/changelog/index.mdx @@ -5,37 +5,30 @@ icon: "clock-rotate-left" --- +**模型支持** +- 支持 [ChronoEdit 14B](https://research.nvidia.com/labs/toronto-ai/chronoedit/) 模型 -**内存和性能优化** -- 引入**混合精度量化系统**，优化模型加载和内存使用 -- 使用固定内存显著加速模型卸载，支持自动低RAM硬件检测 +**性能与内存优化** +- 引入**混合精度量化系统**，优化模型加载 - 新增**RAM压力缓存模式**，在资源受限条件下智能管理内存 -- 增强FP8操作，减少内存使用并修复torch.compile性能回归问题 -- 改进异步卸载速度，解决可能导致工作流损坏的竞争条件 +- 使用固定内存加速模型卸载，支持自动低RAM硬件检测 +- 增强FP8操作：减少内存使用并修复torch.compile性能回归问题 +- 改进异步卸载速度，解决竞争条件问题 +- 修复Windows固定内存分配问题 -**新的执行功能** -- **ScaleROPE节点**：为WAN和Lumina模型添加rope缩放支持，实现更好的分辨率适配 +**新节点与执行功能** +- **ScaleROPE节点**：支持WAN和Lumina模型的rope缩放 - 增强子图执行功能，允许在单个工作流中多次运行 - 改进缓存系统，正确处理字节数据和None输出 -**便携版和驱动支持** -- 将稳定便携版更新至CUDA 1.3.0和Python 3.13.9 -- 增强NVIDIA驱动兼容性检查，为便携安装提供用户友好的更新提示 -- 优化Windows固定内存限制以防止系统问题 - **API节点增强** -- 将多个API节点（Luma、Minimax、Pixverse、Ideogram、StabilityAI、Pika）迁移到新的V2客户端架构 -- 扩展LTXV API节点，支持12秒-20秒时长选项，实现更长的视频生成 -- 修复DALL-E 2 API节点中的img2img操作 -- 增强Rodin3D节点，为3D模型输出返回正确的相对路径 - -**架构转换（V3）** -- 将Recraft、Hypernetwork和OpenAI节点转换为V3架构，提高一致性和可维护性 +- 扩展LTXV API节点，支持12秒-20秒时长选项 +- 修复DALL-E 2节点中的img2img操作 +- 增强Rodin3D节点，返回正确的相对路径 **更新** -- 内嵌文档更新至v0.3.1 +- 更新核心节点文档翻译（繁体中文，阿拉伯语，土耳其语） - 工作流模板更新至v0.2.11 -- 改进模型加载状态消息，提供更好的用户反馈