<h3 id="english-is-not-good-dsmpbdpx2">English is not good！英语不好警告！</h3>For the basic text-to-image workflow, see version 1.0. The latter two versions are customized according to personal preferences.1.0是相对简洁的单文生图流程 后面的版本都是个人喜好一些思路分享<hr /><h3 id="2.01-personal-preference-workflow-sharing-skhq48z5w">2.01-Personal Preference Workflow Sharing</h3><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/d03fb095-3a7a-4ebd-898a-4e81a5f17ece/width=525/d03fb095-3a7a-4ebd-898a-4e81a5f17ece.jpeg" />好吧 看上去好像复杂了很多 之前有人问 我就整理了下发了出来 就当我是在分享一个思路吧Alright, I know this doesn't appear to be simple – please regard it as me sharing a conceptual approach.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/8ecf7ab7-3bf0-4d0b-b6c3-624ffa2818c3/width=525/8ecf7ab7-3bf0-4d0b-b6c3-624ffa2818c3.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/e08e0c7a-25b0-4e79-bd68-0654a07f686f/width=525/e08e0c7a-25b0-4e79-bd68-0654a07f686f.jpeg" />现在我是用三个模型来生成图 Noob-v预测 接力 MiaoMiao harem 最后放大完用ntr过一遍出图 流程如上。 因为我感觉Noob-v预测系列的光影明暗要好 Miaomiao的肢体 (Jio) 要好 以及我喜欢ntr的衣物跟表情的质感，当然 这只是纯个人见解。Currently I generate images using three models: Noob-v for initial prediction, relay to MiaoMiao harem, then re-generate with NTR after upscaling. Process flow as above. This is because Noob-v excels in lighting/shading control, MiaoMiao produces superior limb structures, and I prefer NTR's clothing/facial texture rendering - purely subjective perspective.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/a4ad2af7-af9e-4de3-881d-4281789fc078/width=525/a4ad2af7-af9e-4de3-881d-4281789fc078.jpeg" />单一模型生图→双模型接力 1model→ 2model我生图用的lora比较多 而且这里是三个模型的流程 所以lora看起来非常多 但你们也许并不需要这么多 所以都删掉也ok.I use numerous LoRAs for image generation, and since this is a three-model workflow, the LoRA list appears extensive. But you probably don't need this many - so it's perfectly fine to remove them all.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/0e2f8384-5f07-4a10-b1c4-10c695b0955c/width=525/0e2f8384-5f07-4a10-b1c4-10c695b0955c.jpeg" /><h3 id="-lttti8mus"><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/336844c9-0357-49ec-a35c-4e6254077fbd/width=525/336844c9-0357-49ec-a35c-4e6254077fbd.jpeg" />-</h3><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/bea9b18c-f4df-466a-a74e-2db612d84a42/width=525/bea9b18c-f4df-466a-a74e-2db612d84a42.jpeg" />生成步骤分的多 就像图生图那样 可以中间单独调整tag跟lora 我还是比较喜欢这样的。The multi-step generation process, similar to image-to-image workflows where you can adjust tags and LoRAs mid-process, is what I personally prefer.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/e8f4b7a7-b74b-4194-a4b0-14ede967d062/width=525/e8f4b7a7-b74b-4194-a4b0-14ede967d062.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/277561ff-d0f9-4453-b61f-7446f0903f13/width=525/277561ff-d0f9-4453-b61f-7446f0903f13.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/5141d5e2-cd8a-4c63-b076-65dba04d985e/width=525/5141d5e2-cd8a-4c63-b076-65dba04d985e.jpeg" />是的 我就是为了jio跟袜子 ///<h3 id="2.0-6vlmlmou6">2.0</h3><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/edc98a20-ba7c-41b3-8a4b-80f79fda50ad/width=525/edc98a20-ba7c-41b3-8a4b-80f79fda50ad.jpeg" /><h3 id="comfyui-weibu-j5bxn61xi">也用了段时间的Comfyui 感觉确实比weibu好用，这次突发奇想折腾出个新的放大方法感觉挺有意思就顺便分享一下工作流</h3>I have been using Comfyui for a while and I feel it is indeed easier to use than Weibu.This time, I came up with a new amplification method on a whim. I thought it was quite interesting, so I would like to share the workflow.<h3 id="-rgat4fx6k"><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/d94b2ab5-ed2c-4521-bd79-c59d5c07e048/width=525/d94b2ab5-ed2c-4521-bd79-c59d5c07e048.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/01f812d1-4557-4e12-a4ab-fdd308e09288/width=525/01f812d1-4557-4e12-a4ab-fdd308e09288.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/b1d30708-9f10-4733-86cb-e8259944bf7f/width=525/b1d30708-9f10-4733-86cb-e8259944bf7f.jpeg" /></h3><h3 id="tag-weibu-lora-5c9awzg8i">最右下角的那个是二次放大用的正面tag 可以自己修改折腾 就像weibu的图生图那样，最上面的是二次放大 基底模型串联起来的lora。</h3>The one in the lower right corner is the front tag for secondary enlargement. You can modify it by yourself, just like Weibu img2img-images. The top one is the lora connected in series with the base model for secondary enlargement. <h3 id="controlnet-....-qaq-bdanf6fus">重绘 跟controlnet那边就不细说了 说起来太麻烦 但我觉得这里也看得懂吧.... QAQ</h3>I won’t go into details about redrawing and controlnet. It’s too troublesome to talk about.But I think you can understand it here... QAQ<h3 id="-6na7frts7">这次主要是跑图的时候被气到了</h3><h1 id="very-angry-!-3ix3lo1ew">very angry !<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/b09eab03-c770-4cc5-900b-9e4bbc6f8f70/width=525/b09eab03-c770-4cc5-900b-9e4bbc6f8f70.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/46545ab0-279f-4488-be78-2bf02398e910/width=525/46545ab0-279f-4488-be78-2bf02398e910.jpeg" /></h1><h3 id="taglora-exs5d2ua5">何意啊！！ 底图的脚这么好 怎么放大就一坨了！ 我知道是我的画师tag跟lora有一定程度上的污染 所以我就想了个办法来解决</h3>feet good feet bad, unhappy,I knew that my artist tag and lora were contaminated to a certain extent, so I thought of a way to solve it.<h3 id="weibu-tag-lora-h1p0nvir6"> 类似于weibu的图生图流程 先给一个底图 然后图生图那边就可以增减tag 和lora来控制出图的风格跟细节</h3>Similar to Weibu's img2img-images, you first give a base map, and then you can add or subtract tags and loras to control the style and details of the drawing.<h3 id="trt-style-this-is-girl-body-without-trt-style-midm3c33y"><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/6a1c60ac-b3dc-4ffa-8f52-3aa2aa7e8fce/width=525/6a1c60ac-b3dc-4ffa-8f52-3aa2aa7e8fce.jpeg" />这是没用trt style的人体 This is girl body without trt style</h3><h3 id="trt-style-style-lora-ymvftrzyg">简单的一个例子 我想要trt style的人体 但是我觉得他有点油 我就可以用这种方法在高清放大的时候用其他style lora来换种风格</h3>A simple example: I want a trt style human body but I think it is a bit oily. I can use this method to change the style with other style lora when zooming in in high definition.<h3 id=".....-tagstyle-lora-rmisfnyet"> <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/6206d8c6-8f27-42a4-bbda-2e75ca7279a6/width=525/6206d8c6-8f27-42a4-bbda-2e75ca7279a6.jpeg" /> 你可以看到 脚趾的效果变好了..... 因为我去除了一些画师tag以及更换了别的style lora。</h3> You can see that the toe effect has improved... because I removed some artist tags and changed to other styles lora<h3 id="lora-j0n8pgpub">还有这种原lora风格化太强 导致放大后效果不佳的</h3>There is also this kind of original lora that is too stylized, resulting in poor effect after enlargement.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/a3766e3e-1d9b-4319-8d1d-7d0fcb50bd35/width=525/a3766e3e-1d9b-4319-8d1d-7d0fcb50bd35.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/2473940a-aefc-49d9-af3b-8860aafce7c6/width=525/2473940a-aefc-49d9-af3b-8860aafce7c6.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/643ceff1-14c2-4f92-bc48-28dd2a7e30a2/width=525/643ceff1-14c2-4f92-bc48-28dd2a7e30a2.jpeg" /><h3 id="lorastyle-lora-rycy4my5p">只是简单的在新建的二次放大节点把角色lora的强度调低了点并换了个style lora</h3>Just simply lower the intensity of the character lora in the newly created secondary amplification node and change the style lora<h3 id="emmm...darknesstagellen-joetag-abbgz21d3">emmm...以及最右边的我忘了把darkness的tag换成Ellen Joe的触发tag了 但也挺好看的</h3>emmm...and on the far right, I forgot to replace the darkness tag with Ellen Joe's trigger tag, but it's pretty cool too.<h3 id="-6ehbmlag0">所以我想大伙应该能够理解这种放大的好处以及应用在哪</h3>So I think you you and you should be able to understand the benefits of this amplification and where it is applied.<h3 id="-gk2ybnes8"> 我的工作流只想简单涩涩 嗯 就这样</h3>My workflow just wants to create simple sex pictures, that's it <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/9924e450-4d2d-4864-8ab5-8a44e7273250/width=525/9924e450-4d2d-4864-8ab5-8a44e7273250.jpeg" /> ------------------------------------------------------------------------------------------------------<h3 id="1.0-pmnbuhvmi">1.0</h3>I recently learned comfyui and sometimes I just want some very simple features, but the workflows uploaded by others on the site are too complicated for me as a newbies or don't include my needs, so I created this workflow.我就想要简简单单生图放大涩涩<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/e846c55b-05ce-45df-b49d-a1a17e8b2afb/width=525/e846c55b-05ce-45df-b49d-a1a17e8b2afb.jpeg" /><ul><li><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/122a1073-4990-42df-8cb4-f1557657a925/width=525/122a1073-4990-42df-8cb4-f1557657a925.jpeg" />A separate Negative tag is used to control magnification</li><li>放大用单独的tag就可以做到力大砖飞了 ((((((((BAD FOOT)))))))) 对....对吗....</li></ul>I love using more than one LORA, maybe you don't need that many, so you can delete the green part yourselflora用的多 可以删 删完记得连连看就行了<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/44d52f1e-6634-48cf-8d97-d6c619e98630/width=525/44d52f1e-6634-48cf-8d97-d6c619e98630.jpeg" />These two LORAs are used to control the character这两个放角色<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/2a928a2b-32ce-446a-9f6b-0ec7b8b0cd29/width=525/2a928a2b-32ce-446a-9f6b-0ec7b8b0cd29.jpeg" />If you find it OK, please upload pictures and let me know/////Reference workflow<a target="_blank" rel="ugc" href="https://civarchive.com/models/364333/comfyui-face-eyes-and-lips-adetailer-flow-tutorial">https://civarchive.com/models/364333/comfyui-face-eyes-and-lips-adetailer-flow-tutorial</a>

A simple workflow - HiRes Fix + Various Detailer

aSimpleWorkflowHiresFix_v20.zip

V 2.0

v2.01

v1.0

<h3 id="english-is-not-good">English is not good！英语不好警告！</h3>注意：当使用多个角色+其他LoRA时，单次放大可能无法获得较好的效果。最好像我工作流一样 在放大完后再添加一个K采样生成一次。（可以是你原来的模型（应该？） 我是用的别的模型。如果生成或者放大出现报错 大概率是分辨率设置的问题 需要设置成64的倍数。我在这里分享的所有图片都是原图，你可以拖入到你的comfyui查看tag以及其他信息。tips：When using multiple LoRA models and generating multiple characters, a single upscaling pass may not yield optimal results. It's best to follow a workflow like mine: perform upscaling first, then connect a K-sampler for the final generation.If there is an error message, it should be due to the resolution setting issue. The resolution needs to be set to a multiple of 64.All the pictures I'm sharing here are the original ones. You can drag them into your comfyui to view the tags and other information.<hr />2.0 更新：默认区域增加到4个 可根据自身使用需求继续增加或关闭。根据使用体验在流程里添加了CN的深度跟tile节点来稳定构图。2.0 Update: Default rendering zones have been extended to four units, with scalability for customization or deactivation per operational requirements.Through empirical optimization, ControlNet Depth and Tile have been embedded within the processing chain to maintain correct image upscaling operations.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/4b18a84e-256f-4b71-90e5-90e42c3a58d3/width=525/4b18a84e-256f-4b71-90e5-90e42c3a58d3.jpeg" />1:第一个区域用来写大概的构图tag ，例如这张图 需要明确主体：3girls 场景：living_room,kitchen,wooden interior,winter,morning,sunlight 。 如果有需求还可以添加上视角tag。1: The first area is used to write the general composition tags. For example, in this picture, the main subject needs to be clearly identified: 3 girls. The scene: living room, kitchen, wooden interior, winter, morning, sunlight. If necessary, perspective tags can also be added,like from above,from below.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/26513f25-bff5-47f0-a73a-529308d5541f/width=525/26513f25-bff5-47f0-a73a-529308d5541f.jpeg" />2 3 4处的框，在前面的lora节点添加进你想要生成的人物角色lora。框里的内容就是你生成的角色触发词以及你想要的动作内容。The boxes at 2, 3, and 4, add the character lora you want to create to the previous lora node. The content inside the boxes is the trigger words for the character you generate and the actions you desire.5：这里我选择辅助场景生成，用粉色的遮罩画出桌椅，所以框里面的tag就写桌椅就行了。5: Here, I choose to assist in scene generation. I use a pink mask to draw the table and chairs, so the tags inside the frame can simply be written as "table and chair". <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/f24d94d0-d206-4b86-b778-5865a63b765b/width=525/f24d94d0-d206-4b86-b778-5865a63b765b.jpeg" />经过生成我们得到了最后的图像 看起来不太行对吧... 是的 这时候我们就要用到局部重绘了。After the generation process, we obtained the final image. It doesn't look good, does it... Yes, at this point, we need to use local redrawing.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/9ad959c2-004f-45b3-9057-6817d4fbd372/width=525/9ad959c2-004f-45b3-9057-6817d4fbd372.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/95c69f14-79da-4d90-9883-e0b427d79e5d/width=525/95c69f14-79da-4d90-9883-e0b427d79e5d.jpeg" />通过右键加载图像 打开遮罩编辑器 在里面对你需要修改的地方进行涂抹，像这种情况就是三个角色了。当然，你也可以涂抹你的场景进行重绘修改。Load the image via the right-click menu, then open the mask editor. Within it, make your modifications to the areas you need to change, such as in this case, there are three characters. Of course, you can also paint over your scene to re-draw and modify it.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3aecbfe5-6f81-4a74-b23b-01165f20b0b4/width=525/3aecbfe5-6f81-4a74-b23b-01165f20b0b4.jpeg" />哦对了，生成图片的时候ControlNet不一定要全部开启，一切以辅助出图效果为准，毕竟开的多生成速度会变慢。Oh, by the way, when generating images, ControlNet doesn't necessarily need to be fully enabled. It's enough to use it to assist in achieving the desired image effect. After all, if it's enabled too much, the generation speed will slow down.最后，关于多角色底图的选择。像这张图我是先用上面说的主体跟场景跑了没使用3个角色lora的底图，再用它来垫在下面描出的角色遮罩。所以实在没灵感想法还是把它交给ai去生成吧，Finally, regarding the selection of the multi-character background image. For this image, I first used the main subject and scene as described earlier without using the 3-character Lora background image. Then, I used it to lay underneath and draw the character mask. So, if I really don't have any ideas, I'll just leave it to the AI to generate it.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/65c59bb5-72a2-4a52-8d38-dfeedbfba215/width=525/65c59bb5-72a2-4a52-8d38-dfeedbfba215.jpeg" />-----更新一下经验分享：Update the experience sharing ：<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/d8d87c00-cb05-4e53-a391-e2f576f16ef9/width=525/d8d87c00-cb05-4e53-a391-e2f576f16ef9.jpeg" />线稿这里的值不能给太高，要不然就会变成这样......The value of this line drawing cannot be set too high; otherwise, it will turn out like this...<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/82ab25da-f7a3-47a0-9e5c-802a6a6d136a/width=525/82ab25da-f7a3-47a0-9e5c-802a6a6d136a.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/16eb0727-276a-4b7a-96c2-d4326ceabe2f/width=525/16eb0727-276a-4b7a-96c2-d4326ceabe2f.jpeg" />还有如果分区多角色 但是只有单角色 可以尝试在框里写上多个女孩 并用,,隔断 ，这东西对提示词还是太敏感了，需要动动脑写的比较符合才能出来好的效果。Also, if there are multiple characters in a zone but only a single character is present, you can try writing multiple girls in the box and using commas to separate them. This method is still too sensitive to the prompt words and requires some thinking to write in a way that produces a good result.<hr />1.0：添加方式如下：Adding method:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/07af5c13-cd36-4e83-92da-81171d084a7a/width=525/07af5c13-cd36-4e83-92da-81171d084a7a.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/1ed87cf5-7d0a-44e5-b405-33b6baf76a84/width=525/1ed87cf5-7d0a-44e5-b405-33b6baf76a84.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/a2a3d902-05ae-40d3-961f-158070e41f69/width=525/a2a3d902-05ae-40d3-961f-158070e41f69.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/44055c8c-53a0-49bb-b20e-8118567433b6/width=525/44055c8c-53a0-49bb-b20e-8118567433b6.jpeg" /><hr />参考/Related article：<a target="_blank" rel="ugc" href="https://www.bilibili.com/video/BV1TK21Y8EuL/?spm_id_from=333.1007.top_right_bar_window_history.content.click">https://www.bilibili.com/video/BV1TK21Y8EuL/?spm_id_from=333.1007.top_right_bar_window_history.content.click</a>

Files with SHA256: 9e7903c859d2bd9daa855c8f5fcb0d30348f8be7e3fb371a1a71be8e3c714a92