Workflow Explained (in Chinese):
Flux.1_v2.0: https://www.bilibili.com/video/BV19GDjYiErK/
Flux.1_v1.0: https://www.bilibili.com/video/BV16nSaYpE5A/
Features
This workflow adopts an iterative approach, automatically determining the necessary parameters and process based on the image size. It upscales by a factor of 2 each time, repeating the process until the target size is reached.
Solved Issues
Compared to common upscaling workflows, this workflow primarily addresses the following issues:
When the original image size is too small, forcing an upscale results in significant deviations from the original.
Different sizes of original images require specific parameter adjustments or even different upscaling processes to achieve optimal results.
When using prompts to guide AI in Tiled upscaling, the AI may sometimes experience "hallucinations" generating unwanted elements.
Requirement
Models
Any Flux.1 unet model as the base model
.\ComfyUI\models\unet\
VAE: ae.safetensors
.\ComfyUI\models\vae\
CLIP: t5xxl_fp8_e4m3fn.safetensors, clip_l.safetensors
.\ComfyUI\models\clip\
Upscale: 2xNomosUni_span_multijpg_ldl
.\ComfyUI\models\upscale_models\
Torbo LoRA: FLUX.1-Turbo-Alpha
Download and rename the file, put it into the path:
.\ComfyUI\models\loras\Flux.1\
(Optional) LoRA that makes the visual style more realistic, it can be replaced by other LoRA: zyd232ChineseGirl_Flux1D_v1_2
.\ComfyUI\models\loras\Flux.1\1_Character-Female\
ControlNet: Flux.1-dev-Controlnet-Upscaler
Download and rename the file, put it into the path: .\ComfyUI\models\controlnet\Flux.1\
Custom Nodes
After importing the workflow for the first time, please use the "Install Missing Custom Nodes" button in ComfyUI Manager menu to get all the nodes needed.
Tips
Generally, you only need to adjust the "Target Size" in the "Input" group and select the image to upscale.
The higher the "Similarity" value, the closer the result is to the original image; the lower the "Similarity" value, the more detailed the result, but with reduced resemblance to the original. Generally, when upscaling portraits, the "Similarity" value should be between 0.97 and 0.99 to preserve facial features. In other cases, you can try lower values, as 0.85 to 0.98 can still yield good results.
__________________________________________________________________
工作流详解：
Flux.1_v2.0: https://www.bilibili.com/video/BV19GDjYiErK/
Flux.1_v1.0: https://www.bilibili.com/video/BV16nSaYpE5A/
特点
这个工作流采用了循环的办法，根据图片的尺寸，自动判断放大时所需的参数和流程，每次放大2倍，不断循环，直至目标尺寸。
解决的问题
对比目前常见的放大流程，这个工作流主要解决了以下问题：
原图尺寸太小时，若强行一次性放大，结果和原图的偏差较大。
对于不同尺寸的原图，都要做针对性的调参，甚至要用不同的放大流程，才能得到较好的结果。
使用提示词引导，并进行Tile分块放大时，AI有时候会产生“幻觉”，绘制出我们不想要的东西。
必需
模型
任意 Flux.1 unet 模型
.\ComfyUI\models\unet\Flux.1\
VAE: ae.safetensors
.\ComfyUI\models\vae\
CLIP: t5xxl_fp8_e4m3fn.safetensors, clip_l.safetensors
.\ComfyUI\models\clip\
放大模型: 2xNomosUni_span_multijpg_ldl
.\ComfyUI\models\upscale_models\
加速 LoRA: FLUX.1-Turbo-Alpha
下载并重命名，放到以下路径中:
.\ComfyUI\models\loras\Flux.1\
(可选) 使风格更加真实的 LoRA，可以用其他 LoRA 代替: zyd232ChineseGirl_Flux1D_v1_2
.\ComfyUI\models\loras\Flux.1\
ControlNet: Flux.1-dev-Controlnet-Upscaler
下载并重命名，放到以下路径中:
.\ComfyUI\models\controlnet\Flux.1\
自定义节点
首次导入工作流程后，请使用 ComfyUI Manager 菜单中的 “Install Missing Custom Nodes” 按钮获取所需的所有节点。
提示
通常来说，你只需要在“Input”组中修改“Target Size”，并选择要放大的图片。
“Similarity”值越大，结果和原图越接近；“Similarity”值越小，结果的细节越多、但是会降低和原图的相似性。通常来说，对人物进行放大时，“Similarity”值应为0.97-0.99，以保持人物的相貌。其他情况下，你可以尝试更低的值，0.85-0.98 都可以得到不错的结果。
Description
Batch processing for v2.0.
Please change the Load Path and Output Path.
_________
v2.0 图片批处理。
请先修改载入路径和输出路径。
FAQ
Comments (26)
up主，这次更新的版本加载有问题，大部分节点被对齐挤到左侧的一堆，返回：TypeError: Cannot read properties of undefined (reading 'type')，问了ChatGPT是Easy-Use的节点无法读取 undefined 的 type 属性，
注：我的Easy-Use已经更新最新版本
1.0版本的工作流没有加载问题
ComfyUI 内核更新导致的问题（我记得是前天的版本有这个问题），更新一下ComfyUI 就行了
ComfyUI 内核已更新至12月6日最新版本，Easy-Use更新至12月4日版本，也是一样的问题，Easy-Use的节点无法读取 undefined 的 type 属性
@lss2024888458 ComfyUI 目前的最新版本应是12月04日的 v0.3.7。而你说的是12月6日的版本，是不是用了别的 commit？我这边一切正常，也让网友测试过、都没问题。我也查了一下你的报错，大概率并非easy-use导致的问题，建议看一下是不是某些前端相关的辅助类插件的问题（比如翻译、连线优化、可视化等等）
可能是有什么插件的依赖冲突了，多谢up主
@zyd232 我也遇到了一模一样的问题，我们估计都是comfyui portable的使用者，而非秋叶版本，目前直接把最新的2.0工作流拉入浏览器就会出现 巨长的一列错误
Loading aborted due to error reloading workflow data
This may be due to the following script:
/extensions/ComfyUI-Easy-Use/assets/extensions-29B2pS6_.js
一潭可以试试用官方版本试验一下，这肯定不是楼主一人的个例
@jingtianli 我这边用的是官方的源（https://github.com/comfyanonymous/ComfyUI），没有用整合包啥的。
@zyd232 https://github.com/yolain/ComfyUI-Easy-Use/issues/578
此bug经过Easy-Use 更新以后已经可以正常在python 3.12中使用
With the last updates of ACN you may want to correct the workflow to avoid the following error : ACN_ScaledSoftControlNetWeights 'int' object has no attribute 'copy'
To do so, unplug the connector from "Loop Upscale Path A" named "<= Base_shift - If else" connected to "Scaled Soft Weights ACN" cn_extras.
By removing this connector the workflow will operate again.
A big thanks to KosinKadink for the solution. https://github.com/Kosinkadink/ComfyUI-Advanced-ControlNet
Thank you for the information.
I have uploaded the fixed workflow for v2.0.
ComfyUI Portable 官方最新版出现如下错误，无法导入工作流
Loading aborted due to error reloading workflow data
TypeError: Cannot read properties of undefined (reading 'type')TypeError: Cannot read properties of undefined (reading 'type') at e.onConnectionsChange (http://127.0.0.1:8188/extensions/ComfyUI-Easy-Use/assets/extensions-29B2pS6_.js:1:100022) at e.configure (http://127.0.0.1:8188/assets/index-CoOvI8ZH.js:29240:35) at LGraph2.configure (http://127.0.0.1:8188/assets/index-CoOvI8ZH.js:39063:17) at LGraph$1.configure (http://127.0.0.1:8188/assets/index-CoOvI8ZH.js:88029:26) at LGraph.configure (http://127.0.0.1:8188/extensions/ComfyUI-Custom-Scripts/js/reroutePrimitive.js:14:29) at ComfyApp.loadGraphData (http://127.0.0.1:8188/assets/index-CoOvI8ZH.js:88529:18) at async app.loadGraphData (http://127.0.0.1:8188/extensions/ComfyUI-Manager/components-manager.js:773:9) at async reader.onload (http://127.0.0.1:8188/assets/index-CoOvI8ZH.js:88898:11)
This may be due to the following script:
/extensions/ComfyUI-Easy-Use/assets/extensions-29B2pS6_.js
这里建议up可以把SetNode统一修改为comfy自带的setnode，来避免此类问题
这也许是官方整合包python版本过高是3.12导致的原因？
https://github.com/yolain/ComfyUI-Easy-Use/issues/578
此bug经过Easy-Use 更新以后已经可以正常在python 3.12中使用
I installed all the missing nodes and even deleted all my nodes and reinstalled everything but the connections (e.g. "Size Target", "Image Input",...) keep framed red and I get the error "Cannot execute because a node is missing the class_type property.: Node ID '#21'"
Install Easy-Use from github: https://github.com/yolain/ComfyUI-Easy-Use
And make sure ComfyUI and Easy-Use both are the newest version.
up你好！有没有可能只是变清晰，而不需要放大或者放太大，同样清晰的情况下，又能节约时间？（我试了小数值的放大效果清晰度不行，少了很多细节）
Thank you for the workflow! It is excellent! Thank you for your work!
Something bad is happening in the Loop Upscale Path A group. I will post in the gallery, but setting the similarity to .95 and .97 the image is not recognizable as the same image. If this worked as intended, I think this would be a great tool for the upscale arsenal. Hopefully this is an easy fix. This appears to happen with all images.
I can't get it to output anything for some reason. It just finishes almost instantly, and has no output image.
got promptWARNING: PlaySound.IS_CHANGED() missing 1 required positional argument: 'self'Prompt executed in 0.84 secondsSame here. Removed play sound and still finished in 0.27 seconds with no output.
realy nice, but how to get rid of the fluxish look? maybe some grain lora? add-flux-detail loras does not help for me
It seems that the workflow is no longer working with the latest versions of ComfyUI and Transformers.
Steps to Reproduce:
Fresh install of ComfyUI 0.3.45
Installed all missing models using the ComfyUI Manager
Running the script with Florence2-Flux-Large fails due to an issue in the LayerStyle_Advance node, specifically the LayerUtility: Florence2 Image2Prompt (Advance) step, which causes the process to crash.
Error Explanation:
The crash is caused by GenerationMixin no longer being included in Transformers (version 4.50 and above).
Temporary Solution:
To resolve this issue, install Transformers version < 4.50 (but will not work in the end).
I attempted to modify modeling_florence2.py by changing the class definition:
class Florence2LanguageForConditionalGeneration(Florence2LanguagePreTrainedModel):
to:
class Florence2LanguageForConditionalGeneration(Florence2LanguagePreTrainedModel, GenerationMixin):
And I added:
from transformers.generation.utils import GenerationMixin
at the top of modeling_florence2.py, after the comments section.
Next Issue:
After making the above modifications, I then tried to update florence2_ultra.py, but I’m currently stuck and unsure about what exactly needs to be updated to reflect the new behavior of Transformers.
Additional Info:
After downgrading to Transformers 4.49, Florence2 works, but nothing else seems to function. The process ends without generating an image, and no error message is provided.
Summary:
The issue arises from GenerationMixin being removed in Transformers 4.50.
Downgrading Transformers to version < 4.50 resolves the issue for Florence2, but the overall process still fails with no image output or error.
I filled in both Load Images For Loop and Output Path, but it always find only 1 image and, after FW finished, no image is saved...