Skip to content

Commit 710215e

Browse files
comfyui-wikicomfyui-wiki
committed
Update changelog for version 0.6.0, adding new model support, features, improvements, and bug fixes in both English and Chinese.
1 parent 087fbb3 commit 710215e

File tree

2 files changed

+38
-60
lines changed

2 files changed

+38
-60
lines changed

‎changelog/index.mdx‎

Lines changed: 19 additions & 30 deletions
Original file line numberDiff line numberDiff line change
@@ -6,40 +6,29 @@ icon: "clock-rotate-left"
66

77
<Update label="v0.6.0" description="December 24, 2024">
88

9-
**Major Model Support & VAE Improvements**
10-
- Added support for Qwen Image Layered diffusion model with dedicated empty latents node for enhanced image layering workflows
11-
- Improved VAE compatibility for Wan/Qwen models with automatic channel trimming/padding, enabling seamless integration of diverse model architectures
12-
- Enhanced nested tensor support for denoise masks and custom sampler nodes, providing better control over complex denoising operations
13-
14-
**New Nodes & Workflow Tools**
15-
- Added LatentCutToBatch node for advanced batch processing workflows
16-
- Introduced Flux2MaxImage API node for high-resolution image generation
17-
- Implemented unified jobs API with /api/jobs endpoints for better workflow monitoring and management
18-
19-
**Text Encoder & Attention Enhancements**
20-
- Implemented Jina CLIP v2 and NewBie dual CLIP support for improved text encoding capabilities
21-
- Added sliding attention in Gemma3 models for better long-sequence processing
22-
- Disabled prompt weights on NewBie text encoder to prevent compatibility issues in existing workflows
23-
24-
**Performance & Memory Optimizations**
25-
- Refactored trainer implementation with resolution bucketing for more efficient training workflows
26-
- Changed skip_load_model parameter to force_full_load for clearer model loading behavior
27-
- Enhanced AMD GPU support with optional MIOpen acceleration via environment variable
9+
**New Model Support**
10+
- Qwen Image Layered diffusion model with dedicated empty latents node
11+
- Improved VAE compatibility for Wan/Qwen models with automatic channel handling
12+
- Enhanced support for Jina CLIP v2 and NewBie dual CLIP text encoders
2813

29-
**API & External Service Updates**
30-
- Extended Kling O1 StartEndFrame node duration range for more flexible video generation
14+
**New Features**
15+
- LatentCutToBatch node for batch processing workflows
16+
- Flux2MaxImage API node for high-resolution image generation
17+
- Unified jobs API with /api/jobs endpoints for workflow monitoring
18+
- Sliding attention in Gemma3 models for better long-sequence processing
19+
20+
**Improvements**
21+
- Refactored trainer with resolution bucketing for better training efficiency
22+
- Enhanced AMD GPU support with optional MIOpen acceleration
23+
- Extended Kling O1 StartEndFrame node duration range
3124
- Updated Topaz 4k video upscaling functionality
32-
- Disabled watermark generation by default in API nodes for cleaner outputs
33-
- Integrated new Nano Banana custom endpoint for improved processing capabilities
25+
- Disabled watermark generation by default in API nodes
3426

35-
**Bug Fixes & Stability**
36-
- Fixed mask concatenation issues in ZImageFunControlNet when running in --gpu-only mode
27+
**Bug Fixes**
28+
- Fixed mask concatenation issues in ZImageFunControlNet in --gpu-only mode
3729
- Resolved batch processing errors with NewBie model integration
38-
- Fixed Context Windows logging errors that could interrupt workflow execution
39-
- Corrected Gemma quantization configuration to apply only to relevant models
40-
41-
**Workflow Templates**
42-
- Updated workflow templates to v0.7.63 with latest node configurations and best practices
30+
- Fixed Context Windows logging errors
31+
- Corrected Gemma quantization configuration
4332

4433
</Update>
4534

‎zh-CN/changelog/index.mdx‎

Lines changed: 19 additions & 30 deletions
Original file line numberDiff line numberDiff line change
@@ -6,40 +6,29 @@ icon: "clock-rotate-left"
66

77
<Update label="v0.6.0" description="2024年12月24日">
88

9-
**主要模型支持与 VAE 改进**
10-
- 新增 Qwen Image Layered 扩散模型支持，配备专用空潜在向量节点，增强图像分层工作流
11-
- 改进 Wan/Qwen 模型的 VAE 兼容性，支持自动通道裁剪/填充，实现不同模型架构的无缝集成
12-
- 增强去噪掩码和自定义采样器节点的嵌套张量支持，提供对复杂去噪操作的更好控制
13-
14-
**新节点和工作流工具**
15-
- 新增 LatentCutToBatch 节点，用于高级批处理工作流
16-
- 引入 Flux2MaxImage API 节点，用于高分辨率图像生成
17-
- 实现统一作业 API，提供 /api/jobs 端点，改善工作流监控和管理
18-
19-
**文本编码器和注意力机制增强**
20-
- 实现 Jina CLIP v2 和 NewBie 双 CLIP 支持，提升文本编码能力
21-
- 在 Gemma3 模型中添加滑动注意力机制，改善长序列处理
22-
- 在 NewBie 文本编码器上禁用提示词权重，防止现有工作流出现兼容性问题
23-
24-
**性能和内存优化**
25-
- 重构训练器实现，添加分辨率分桶功能，提高训练工作流效率
26-
- 将 skip_load_model 参数更改为 force_full_load，使模型加载行为更加清晰
27-
- 增强 AMD GPU 支持，通过环境变量提供可选的 MIOpen 加速
28-
29-
**API 和外部服务更新**
30-
- 扩展 Kling O1 StartEndFrame 节点持续时间范围，提供更灵活的视频生成
9+
**新模型支持**
10+
- Qwen Image Layered 扩散模型，配备专用空潜在向量节点
11+
- 改进 Wan/Qwen 模型的 VAE 兼容性，支持自动通道处理
12+
- 增强 Jina CLIP v2 和 NewBie 双 CLIP 文本编码器支持
13+
14+
**新功能**
15+
- LatentCutToBatch 节点，用于批处理工作流
16+
- Flux2MaxImage API 节点，用于高分辨率图像生成
17+
- 统一作业 API，提供 /api/jobs 端点进行工作流监控
18+
- Gemma3 模型中的滑动注意力机制，改善长序列处理
19+
20+
**改进**
21+
- 重构训练器，添加分辨率分桶功能提高训练效率
22+
- 增强 AMD GPU 支持，提供可选的 MIOpen 加速
23+
- 扩展 Kling O1 StartEndFrame 节点持续时间范围
3124
- 更新 Topaz 4k 视频超分辨率功能
32-
- API 节点默认禁用水印生成，输出更清洁
33-
- 集成新的 Nano Banana 自定义端点，提升处理能力
25+
- API 节点默认禁用水印生成
3426

35-
**错误修复和稳定性**
27+
**错误修复**
3628
- 修复在 --gpu-only 模式下 ZImageFunControlNet 的掩码连接问题
3729
- 解决 NewBie 模型集成的批处理错误
38-
- 修复可能中断工作流执行的 Context Windows 日志记录错误
39-
- 更正 Gemma 量化配置，仅应用于相关模型
40-
41-
**工作流模板**
42-
- 工作流模板更新至 v0.7.63，包含最新节点配置和最佳实践
30+
- 修复 Context Windows 日志记录错误
31+
- 更正 Gemma 量化配置
4332

4433
</Update>
4534
<Update label="v0.5.1" description="2024年12月18日">

0 commit comments

Comments
 (0)