IPAdapter在复杂流程中的表现有些让人失望，试试ComfyUI原生的区域条件采样节点吧IPAdapter's performance in complex processes is somewhat disappointing. Try the ComfyUI native region conditional sampling node<hr />添加了一个提前确定构图区域的方法，Added a method to determine the composition area in advance,<hr />v1.0 regional conditioning_21个主体+背景1 subject + background<hr />v2.0 regional conditioning_32个主体+背景2 subject + background<hr />v3.0 regional conditioning_54个主体+背景4 subject + background如果你需要，区域还能被分割成更多的部分，前提是你的电脑能支持直接生成更大的图片，否则，太多小区域只会造成混乱，The area can be divided into more parts if you want, provided that your computer can directly generate a larger picture, otherwise, too many small areas will just cause confusion.<hr />v4.0 LatenComposite_43个主体+背景3 subject + background有别于使用区域条件采样节点的方式，LatenComposite可以更加灵活的在前期初步绘制分割的区域的内容然后合并成一个整体，缺点是容易因为提示词污染而让画面出现不可控的东西Unlike the use of regional conditional sampling nodes, LatenComposite can be more flexible in the early stage of the initial drawing of the content of the divided area and then merged into a whole. The disadvantage is that it is easy to make the screen uncontrollable due to the pollution of the prompt word<hr />v5.0 DenseDiffusion使用DenseDiffusioon节点的分区绘图功能，目前来看，效果还不错，区域间的污染比较少，融合也比较自然，缺点是，有时候尤其是写实类的风格看上去像是劣质的PS，不过在较小尺寸的图片中可以通过多次放大重采样来解决这个问题，总体来说还是很不错的。Using the partition drawing function of the DenseDiffusioon node, the effect is not bad so far, the pollution between regions is relatively small, and the fusion is more natural. The disadvantage is that sometimes, especially the realistic style looks like inferior PS, but in smaller size images, this problem can be solved by multiple enlargement resampling, which is generally very good.另外需要注意的是，当你想要绘制某种风格的图像时，需要在所有的文本编码框里填入同样的风格提示词，而不是仅仅在整体描述框内填入这些风格词In addition, it should be noted that when you want to draw a certain style of image, you need to fill in the same style prompt words in all the text coding boxes, rather than just filling in the overall description box

Regional Conditional Sampling

注意：目前DenseDiffusion这个插件有BUG，某些出图尺寸会出现张量错误，解决方法是换成可用的尺寸，比如704*1280 ， 512*2048等等，另外在进行放大时需要选择laten放大并选择合适的倍数进行放大，否则也会出现张量错误，经过测试发现，不同的出图尺寸需要不同的放大倍数才能正常进行放大，上面说过的2个尺寸，1.5倍和2倍没有问题，其他倍数请自行测试，二次放大时的倍数同样需要进行测试，比较尴尬，但是可行，至于像素放大目前我自己发现不可用，需要更多测试才能得出结论。Attention:The current DenseDiffusion plug-in has bugs, and tensor errors will occur in some drawing sizes. The solution is to replace it with available sizes, such as 704 1280, 512 2048, etc. In addition, when magnifying, you need to select laten to enlarge and choose the appropriate multiple to enlarge. Otherwise, tensor errors will also occur. After testing, it is found that different drawing sizes require different magnifications to enlarge normally. The above-mentioned 2 sizes, 1.5 times and 2 times are no problem. Please test other multiples yourself. The multiples at the time of secondary enlargement also need to be tested, which is awkward, but feasible. As for pixel enlargement, I have found that it is not available at present, and more tests are needed to reach a conclusion.为了方便画面统一风格，增加字符串合并节点以优化了工作流In order to facilitate the unified style of the screen, string merging nodes are added to optimize the workflow

Files with SHA256: 8d5c43184b3a89c9dd768f10fae8f42a19c70e53d3ce435e960a99416b34c5f4