Simple Krea2 Dataset Creator
LC123 Nodes: https://github.com/lonecatone23/ComfyUI_LC123_nodes
You will need to update. I just uploaded them to ComfyUi. If you get errors, pull them from teh repo
Instagram: https://www.instagram.com/synth.studio.models/
Buy me a☕ https://ko-fi.com/lonecatone
This represents many hours of work that I give away for free. If you enjoy it, Please make sure to 👍like and post to my workflows. That is how you support me. Also, feel free to ⚡tip 😉
There's nothing better out there for making character datasets.
V8.0
While messing with my flagship Krea workflow, I realized that Remove backgroud helps Identity edit.
Added remove background
added preview windows to verify
changed the prompt selector📝 switch to prompt list names instead of numbers for ease of use
⚠️ you will need to update the LC123 nodes. If you have an error, that is because ComfyUi is slow in updating. Pull the pack directly from the Repo
V7.0
Removed the anchor prompt from the final Text output
Thanks @artyclaw for pointing it out 👍
minor start setting tweak
V6.0 Pro Grade
Upgraded 🎨Post Production Nodes
Added denoise
Added headshot and action shot datasets
Blackwell Attention Fix is here: https://github.com/lonecatone23/comfyUI-BlackwellAttentionFix
LC123 Nodes: https://github.com/lonecatone23/ComfyUI_LC123_nodes
V5.0 (because I can't ever stop messing with stuff 🤔)
Got irritated with the standard Image compare node taking the comparison numbers and spreading them out across the top of the screen, making them useless, so made my own.
Added a quality modifier and clothes swap anchor⚓ to make it so you can drop any image in and get different clothing, plus better skin.
Redid the prompts to add colors to every piece of clothing 👗👚. Huge difference in outfit variation now.
Updated the notes to include all the changes.
Added a link to my nodes (which should be in the manager).
I also added the folder in the .zip file
V4.1
rebuilt the filename node so that the number aligns with the name of the image without having to change it.
renames teh Power LoRA loader to add "Character" to it.
Moved the "play sound 🔊" to the end of the workflow.
V4.0
Added newer version of ID Edit LoRA
added link to the Repo
Added 🎨 Post processing Suite
Contrast adaptive sharpen
Color Match to Image 🖼️
Image adjust
Added a better aspect ratio control
updated notes
V3.1: addresses the ComfyUi list selector bug.
Uses Identity Edit technology for a consistent image
Creates a prompt tag set with each image
Dozens of seperate prompts to choose from
Dual image 🖼️ option for better variance between facial expresions
comprehensive notes and tips.
Blackwell attention fix node: Deals with old architecture on newer machines. (Can be deleted if not needed)
The folder is included in the zip file. The entire folder gets placed in the custom_nodes folder.
https://civarchive.com/models/2806105/krea2-identity-edit-rtx50xx-blackwell-attention-fix
Repo is here: https://github.com/lonecatone23/comfyUI-BlackwellAttentionFix
LC123 Nodes: https://github.com/lonecatone23/ComfyUI_LC123_nodes
也有中文说明
I take a lot of time to get these to work. If there are issues, please let me know so that I can correct them.
Description
removed Anchor prompt from final text file
updated post processing nodes
FAQ
Comments (22)
I'm loving this but I havea few bits of feedback.
1) The post-processing is better being turned off
2) And this is a real issue for me. There is a real issue with the Identity Edit Method - the square chin. The whole facial structure is not really recognized by Identity Edit. This is a Krea2 problem not a you problem.
3) Again this is an Identity Edit Method issue - facial expressions are too subtle
4) and this si probably obvious. This is basically female only
I ran a male dataset through it. I customized the prompts accordingly. It worked fine.
asjust the two strength sliders below the image. It recognizes facial structure significantly better with only one image, but totally loses facial expression variation. It's a toss up.
I also have switches and settings for the post processing. You have to dial it in. It was set for a blonde with pale skin. I'm actually getting really dissappointed that people aren't liking it 😔
@lonecatone23 Oh dude. I love this! Just some really minor spoiled brat points TBH. Your work is amazing. I'll give it a shot with the male prompts. And yes I saw the toggles for the post-processing and my feedback really is minor on that. The thing about the facial structure isn't a you problem at all - it's just the limit of the magic Identity Edit can do.
Dude seriously this is great work and thank you for it and for the reply :)
@Zeddy456 I just tweaked it.
@lonecatone23 I'm so glad to have come across this. Was doing something on my own and this is really much more efficient. I'm currently doing my first run and will tinker, tune, and maybe post my own result/review here. Thanks for creating it and don't be too discouraged. It looks like an incredible amount of time went into this.
I'm finally able to run this WV.
Making some tests with it. I also find post processing not good.
It leave the image without skin details and it make it more blurry with more shadows.
Then adjust it. Duh
@lonecatone23 I tried many variations, it just doesn't worked out for me.
Is there a way to make this work with fantasy characters? I tried with one of mine and it nearly completely stripped everything that made her, her.
I did not create the ID Edit LoRA nor its features, and I have not tried it to be able to tell you. Ask the author. His link is in the suggestions at the bottom of my page.
btw, if you want to use more than two reference images, you can just stitch together 2 images for image 1 and/or 2 and use those as ref. I tried 2+2 and it seems to work pretty well.
wait.... just pre-stitch them and add them as one image, or stitch the latents?
@lonecatone23 Load two images, stitch them side by side (if portrait), and use them instead of your original image loaders.
working good for me but eyes sometimes not great. I think if you could add an adetailer or some faceyolo maybe would fix. post process seems to make worse...
in general for people writing prompts need to have them facing forward. side views for example loses face accuracy.
Post process works absolutely fine and if I add a detailer on here, it'd take hours to process. You probably need higher quality images.
@lonecatone23 i figured it out. was me being dumb. settings wrong for the checkpoint i was using. post works now and lut looks good. eyes still not perfect. running faceyolo fixes but i lose accuracy to the input ref face... only take a 1min per image to run the faceyolo.
@dontforg3t2137 well that's good to know 🤔. that's an additional 20 minutes per dataset. Plus, a good checkpint grabs it without issue. I'd say a better approach would be an eye LoRA. I'll look into it, but this is the first I'm hearing about it
@lonecatone23 i had similar issue with eyes in your dataset for some reason in v5.0
@andreiskezznr1315 That's a quality issue. Low resolution, gguf, int8, and fp8 models, turbo LoRAs, and low quality images produce that. Not a fault of the workflow, model, or ID LoRA.
@lonecatone23 yea it seems in general krea 2 needs some faceyolo. other workflows i tried had same issue. Also keep in mind my background is in retouching so I am more picky lol. Thank you for putting up with me.
@dontforg3t2137 I want perfection as well. I get it. The tech will catch up. That's why I keep updating
First, I want to say thank you, this workflow is awesome.
I've ran this now about 8 times, but each time, regardless of which number I select, I still get the same prompted images. Whether I choose 1-8, each time it still gives the same poses. If you could give an indication of what change may be necessary to fix that, it would be much appreciated.