Image GEN (FLUX/SD3/SDXL/PONY/SD1.5) with ControlNet, SUPIR, Tile Upscale, Face Swap版本v1.2 (ID: 684460)

ComfyUI Workflow for single image generation

• Needed Custom Node: https://github.com/Rvage0815/ComfyUI-Rvaged.git

• I've updated the node today (2024.08.11) so make sure you have the latest version. its not visible in the manager so check How to install custom nodes by hand if you don't know.

• check the comments and the images in there (1st Step, 2nd Step) how to run this WF in 2 steps

Everything you need is linked in Links /Resources. e.g.

• FLUX Schnell/Dev

• FLUX ControlNet

• SUPIR Checkpoint

• SDXL All4One ControlNet

• SDXL Tile ControlNet

• Upscaler Links

• ...

Changes:

v3.5

• IMPORTANT: i have updated the rvaged node (again)

• i've revised the controlnet section so that:

  • you can use sdxl union in the correct way.

  • it is now possible to load the images from a directory and use their poses or depthmaps etc.

  • no controlnet updates for flux right now sorry.

• the remove text group has been reworked again, i think it works well now and has also shrunk a bit...

• for no particular reason i have added a latent upscale sampler, i'm just kidding someone asked for it.

• the sampler settings are now also available for the upscale checkpoint.

• the sam model can be switched on/off.

• for the detailers i added a wildcard node now, for the prompt but more important for the loras in the prompt. this prevents them from changing more of the image than they should but only the recognized area of the detailer. e.g. the loras for eyes and teeth.

• i have tidied up some groups a little so that they look nicer, but i would recommend hiding the noodles, it is much more pleasant to work without them.

• in front of each row is an fast groups bypasser. the first one contains all groups while the others only contain the specific groups for that row.

• the loras used in the detailers are also saved into the generation data of the image (when using the save gen-data group)

v3.0

IMPORTANT: i have updated the rvaged node

• added some passer and a different scheduler selector (its smaller)

reworked some groups

• flux/sd3 share one group

• prompt area: switched from the 3 crea nodes to the single one

  • added one more wildcard node, its like a suffix that will be added after the crea node.

• added sd prompt reader to get the image infos when using upscale image folder

• shrinked the wf a bit

• fixed the issues when running the workflow in 2 steps. (Image Gen-Data and Loras will be save to the file now)

the archiv contains a preset for sdxl and one for flux, i've also added the fashion wildcards again.

v2.6

• some small optimizations

  • they way the image flows -> 1st sampler or upscale image groups -> remove text -> 2nd sampler (to fix visible areas)

  • added a prompt node in the detailer groups (collapse to edit the prompt, ran a bit out of space)

    • requires the installation of this node: https://github.com/abyz22/image_control

  • in the 2nd sampler its possible to diable the upscale model and/or the downscale node.

v2.5

• IMPORTANT: Update the Rvaged Node

  • i've added a guider switch to switch between the basic guider (flux no neg cond) and the cfg guider (all other).

• reworked some groups

• reordered the model groups, loras, model charger, sampler settings

• using the custom sampler as standard now

• added a lora stack for the upscaler checkpoint

• added bookmark for faster jumping around in the workflow (0-9)

  • you have to disable the image chooser shortcuts in the comfyui settings (comfyui panel -> the gear top right)

• added some notes here and there

• added the fashion wildcards to the archiv. the others can be found here:

  • -> luxy lingery

  • -> 200wildcards

v2.0

• added flux support

  • Check how to install here

  • Update ComfyUI

• added face swap

Features:

• Every Group can be disabled,

• i've added a new version of Remove Text to remove watermarks in each corner.

  • the size of the masks can be changed.

  • each corner can be switched on/off

• you can use 2 checkpoints (one for the initial render and one for the upscale)

  • current preset is: FLUX (Init Render) and dreamshaperXL lightning (upscale)

• it also allows to upscale already existing images (Load Image Node or Load Images from Folder)

  • disable both Sampler Groups and the checkpoint loader that you don't want to use.

With the new node set "Crea-Prompt" this workflow is heavily using random prompts. this can be an issue if you want to change things and restart the generation, but the last prompt is shown in the ShowText Node and also in the console. i've left a description of how to re-generate the last image in the workflow.

it has 3 upscalers:

• 1 after the first sampler that is mainly there to improve the initial render.

• 2 is a combination of supir (can be turned off) and tiled diffusion with a tile controlnet.

• 3 is basicly a copy of 2 without supir. (can be turned off)

• i've added 3 detailers, they do what you set them up for. if you put a hand model its a hand detailer. default setup is face, eye, hand.

there are some filters, LUTs and 2 Watermark Groups that i've already used in the magic moments workflow.

it saves the generation data into the image file.

• there are 2 groups to save the image

  • 1. WAS Node Save Image

    • if you don't want to share the gen-data when uploaing images.

  • 2. Save Image with Gen-Data

    • its preset NOT to embed the workflow because i don't want to share it everytime when i upload an image but it is saved as json file using the same name as the image file.

描述:

• reuploaded again

  • added lora stack to string so the used loras are saved within the image gen-data.

• reuploaded to lower the file size (was 34.5 mb oO)

• added a group to load the images from a folder (one by one with auto queue)

  • it could be used to upscale the output of the vid2vid workflows BUT it takes almost 300s for me each image AND im not sure if the result is smooth enough for a video. the changes might be to drastic. the image quality is of cause good but the added details and the possible changes in lighting might be visible in the video if you stich the images together.

• both groups: Upscale Image and Upscale Folder are able to use the 2nd Sampler to improve details, quite cool but you have to leave the first checkpoint loader, prompt and stuff enabled. both of them using the text remove group too, if enabled. it allows you to remove watermarks, copyright and/or object from an image or already rendered images. just put the mask to the correct location.

• the eye detailers doing 2 cycles now by default. i haven't tested it with the hand detailer but its something to consider...

• in the preview you can see an image from jadewind that i've used for testing the groups, changing eye colors, add more details and stuff like this...

• i don't think that there will be an update tomorrow but who know ;)

训练词语: