My stuff

  • My Workflows

  • Liked Workflows

  • Following Workflows

Go to OpenArt main site
Upload workflow

MJ image refiner (using flux nf4+ dreamshaper xl)

5.0

0 reviews
17
3.6K
831
3
Description

---------------------10.21---------------------

第一版中xl部分是在flux原尺寸图生图后,使用xl直接拉扯到三倍,这并不合理,可能出现细节不足的情况。故本次更新将放大分为两步进行:使用flux放大1.5倍,再用xl放大2倍。

In the initial version, the XL part was directly stretched to three times the original size after generating the image with Flux. This approach is not reasonable and may result in insufficient detail. Therefore, this update divides the enlargement process into two steps: first, enlarging the image by 1.5 times using Flux, and then enlarging it by 2 times using XL.

此版本时间略长,但可以获得更好的细节

This version takes longer time, but will get better details

---------------------original version---------------------

This workflow was originally created for my personal use. I noticed that while MidJourney generates fantastic images, the details often leave much to be desired. This workflow allows me to refine the details of MidJourney images while keeping the overall content intact.

Although this workflow is not perfect, it is sufficient for use on social networks.

This workflow is roughly divided into two parts: image to image generation with nf4 and upscale with xl, resulting in final images with a long side of 4032 pixels. It takes about 2 minutes to process one image using my setup with a 4090 GPU. The reason why both steps aren't performed using flux is that doing so would result in excessive detail loss and make the images appear overly artificial. Of course, you can adjust this workflow according to your own needs

这个工作流程一开始是搭来我自己用的。MJ生成的图像虽然艺术性观赏性不错,但细节完全没法细看。这个工作流的作用就是维持画面整体不变的同时提升细节。

尽管这个工作流并不完美,但用来发发某书某圈啥的还是够的。

这个工作流大致分成两个部分:nf4图生图+xl放大,最后生成的图像长边为4032,我4090一张图大概是2分钟。为什么不两步都用flux呢?因为这样会损失过多细节并导致图片看起来”很假“。当然大家也可以根据自己实际使用需求去修改。

Discussion

(No comments yet)

Loading...

Author

1
825
17
3.6K

No reviews yet

  • - latest (a year ago)

  • - v20241012-081401

Primitive Nodes (28)

Anything Everywhere (1)

CheckpointLoaderNF4 (1)

FluxGuidance (1)

GetNode (9)

Image Comparer (rgthree) (3)

LayerMask: LoadFlorence2Model (1)

LayerUtility: Florence2Image2Prompt (1)

ModelSamplingFlux (1)

Note (6)

SetNode (4)

Custom Nodes (47)

ComfyUI

  • - CLIPTextEncode (4)

  • - ControlNetLoader (1)

  • - ImageUpscaleWithModel (2)

  • - ImageScaleBy (2)

  • - CheckpointLoaderSimple (1)

  • - CLIPSetLastLayer (1)

  • - LoraLoader (1)

  • - SaveImage (2)

  • - ControlNetApply (1)

  • - LoadImage (1)

  • - UpscaleModelLoader (1)

ComfyUI Easy Use

  • - easy promptReplace (2)

  • - easy imageSize (1)

  • - easy float (4)

  • - ImpactQueueTriggerCountdown (1)

  • - KSampler //Inspire (2)

  • - TilePreprocessor (1)

  • - ConstrainImage|pysssss (1)

  • - MathExpression|pysssss (6)

  • - ShowText|pysssss (2)

  • - SeargePromptCombiner (2)

  • - GetImageSize (1)

  • - VAEDecodeTiled_TiledDiffusion (2)

  • - TiledDiffusion (2)

  • - VAEEncodeTiled_TiledDiffusion (2)

  • - Load Image Batch (1)

Checkpoints (2)

flux1-dev-bnb-nf4-v2.safetensors

真实系\dreamshaperXL_sfwV2TurboDPMSDE.safetensors

LoRAs (1)

DetailedEyes_V3.safetensors