ComfyUI 28 Wonderful: Kohya + Self Awareness + Upscale (free workflows), Stable Diffusion
ฝัง
- เผยแพร่เมื่อ 25 ก.พ. 2024
- High quality images can be made with a combination of Self-Attention Guidance and Kohya Deep Shrinkplus Upscaling. Lol, in my enthusiasm on the intro screen and in the video I mixed up the term ... it is called the Self-Attention Guidance ... not Self-Awareness.
With that out of the way ... some updated Stable Diffusion ComfyUI workflows are available for download. They have a lot of options, yet are very simple to operate.
- Size selector
- Style selector
- LoRA slots
- Kohya Deep Shrink
- Self Awareness Guidance
- Latent Upscale
- Face Detailer
- Image Upscale
The workflow from the video is named Default (Turbo).
The folder that contains the workflows plus a README:
drive.google.com/drive/folder...
Upscalers: openmodeldb.info/
Impact Pack: github.com/ltdrdata/ComfyUI-I...
Kohya Deep Shrink: • ComfyUI 27 Incredible ...
Salf Awareness Gyuidance: • ComfyUI 26 The Node Th...
14 keywords for better image: • ComfyUI 25 - Keywords ... - แนวปฏิบัติและการใช้ชีวิต
Thanks for your feedback, it's appreciated.
Amazing! Great job with the workflows. Simple and straight forward instead of all the noodles, I find these so much easier to follow. Thank You so much!
You're welcome. Thanks for your enthusiastic feedback, it's appreciated.
Thank you! All your videos are incredibly helpful for those who are just starting to learn how to use ComfyUI.
You're very welcome!
This is new for me and following your tutorials from the start. It's easy to follow thank you very much.
You're welcome, and thanks for the feedback, it's appreciated.
Your tuto's are always usefull, clear and ... it's so nice to see someone not speaking like a machine gun !
Fantastic workflow! Thanks for sharing and your hard work.
Thanks for your feedback, it's appreciated.
I like your videos! You are amazing! thanks!
@Rudy: Love your videos, keep 'em coming!
Thanks for your nice feedback.
Wow man you really good at this i does a lot of work flow for my self but your stuff it another level respect bro keep it up.
Thanks for your feedback, it's appreciated.
I have just found your channel. Your content is great. Subscribed immediately.
Thanks and welcome.
Excellent! Liked and subbed!
Thanks, it's appreciated.
Usually renaming is not an issue, but it's worth a try to use the original name, or install, via the ComfyUI manager. Is it just these two custom nodes that have issues ... all others work well?
Lol I love how the first upscaler added 15 years to the woman AND the dog. It was a young women with a puppy and then it was mature woman with a mature dog. AI is wild sometimes, it's getting a lot better. But sometimes it's just like wtf bro. Great video and thank you for including a link for the workflow.
Lol, yes, weird things can happen, like that dog also Is a factor 2 too large and the latex dress starts to melt. AI will get better though ... developments go fast.
I tried generating some pics of a young woman in a snowy park with a husky (different workflow). Got about 3 that did that, got about 10 that gave her husky ears lol.
@@runebinder Lol ... yes such can easily happen, just like color bleeding. Was this with Kohya on or off? It seems to me that Koyha sometimes more easily mixes things up. There's a reason why it still is in the 'for testing' folder.
@@rudyshobbychannel, hi. I was just referring to AI in general. Was in ComfyUI but was a basic workflow I put together that’s not a patch on your stuff :)
Dankjewel Rudy!!!
Graag gedaan, bedankt voor je reactie.
kom wel een paar nodes tegen die ik nog niet gebruikte. deze al eens gebruikt? easy styles Selector Comfyui Easy Use.
I go on in EN for now ;) thnx again. not on discord??
Just the hands... Always the HANDS !!! grrr
Are you Dutch mate ?
nice videos !
Yeah, Dutch. Eyes, teeth, hands, feet ... problem areas.
Hey great work. Love your videos and this one is very usefull. Is it also possible to have an image/latent input at the beginning of the pipeline instead of using text2image. Could you use this flow as an enhance/upscale flow? Any tips on this?
Yes, of course, and the workflow is already included in the folder where the download link points to, the workflow is called img2img. It does not yet have the Sag and Kohya nodes in, but you can easily add them yourself. And the Latent Upscale node on top of the 2nd pipe sampler can be deleted.
the 1k image with the dog is better because the shadows are correct. The upscaled img looks weird with the wrong shadows.
also add the perp-neg node to your negative prompt output connected to SAG, and use the fabric sampler advanced with the additional conditioning hookups (pos prompt splits, and goes to pos and neg-null, and negative prompt splits and goes to neg and pos-null), and if you're using sdxl/hires upscales, add freeu v2: Model -> Loras -> IpAdapter/ControlNet -> SAG -> Perp-Neg -> FreeU -> Sampler
Great video for upscaling. Thanks for your knowledge sharing. Is it possible to upscale with InstantID, apply the face with high quality on full body image? Hope can see your video about this in the future.
I haven't tried but I'd say this should be possible because instantID takes place in the image generation step(s) ... the Latent and/or image upscale takes place from there.
I love the fact that you've created a node section with turning the upscaling on and off. Do you have a tutorial on how you do that? I'd like to do that with other functions such as refining and adding details, etc... Thanks
I appreciate you and the work you share.
The switch is a node that comes with the rgthree custom nodes set github.com/rgthree/rgthree-comfy. All you have to doe is put a collection of nodes inside a group. The group is automatically added to the switch.
@@rudyshobbychannel Awesome. I have doen that and now worked that into al lof my workflows. Genius. Thank you sir
It already is in the download folder.
Great worklow Rudy, this details are getting crazy, i have a question, it is about the concept of "Latent Upscale" , to my understanding it is he process of having the AI reimagine the pixels of an image, i see a lot of people using the term"creative upscale" but i think it is lame in comparison to "latent upscaling" which in my mind is what i have been learning.
Can you give me your definition? You are using the term very confidently
@hologramghost5331. My definition may very well be overly simple, but I use the term 'Latent Upscale' when there is no internediate pixel image involved. Like in my workflow, the first Sampler creates a 1024px image, but that's only shown to be able to see what it has been doing. The second sampler is fed with the Latent of the first, and with a multiplication factor of 1.5 it puts out a 1536px image ... I call that Latent Upscaling, because I fed it a latent. Maybe, under the hood, internally, the ttN pipe-sampler node secretly first goes to pixel space, I can't tell. But I don't think so, because when I used the stand alone Latent Upscale node that's available in ComfyUI and fed the upscaled latent to the ttN pipe-sampler, the final result was the same. In summary, I call it a Latent Upscaler if the input was latent ... as simple as that.
@@rudyshobbychannel That makes sense, and would be an accurate description, thank you!
Very interesting video. Thanks, mate! Could you make your edited "sdxl_styles_base.json" file available for download?
It is available in the download folder.
I'm enjoying your videos, thank you as always. I have a question. When I run render, no image is output from workflow png 512, final upscaled image jpg. There is no error statement, just a broken image file icon. any idea why?
It was caused by a different file storage location. Fixed. It was my mistake, thank you.
Would love to test this workflow! Is this SD1.5 or XL? Can you come up with a similar workflow for creative img2img upscaling? Many thanks for your time
Both 1.5 and XL can be used. The download folder also contains an img2img workflow with the same architecture.
Your workflow is great. However, I am using A1111, I have a hard time understanding how to use ComfyUI, I have a huge request! Could you post your best workflows in one workflow - with a link to exactly that? A workflow with Facedetailer, Kohya, Self Awareness and Upscale built in? I see, there are different versions of this, depending on the use case... But maybe we could have three different outputs and then we'd have them all! Thanks for your excellent work!
Exactly that workflow is the one used in the video. It is in the download folder and it is called Default (Turbo).
@@rudyshobbychannel Hey, I'm really new to ComfyUI. Thanks for the reply!
Placing the image size selector in the custom_nodes folder is not working. ComfyUI doesn't see it. And the pipe nodes are'nt seen neither by ComfyUI after installing them trough Manager. I guess they are bugged atm? (btw: another amazing vid. Thank you Rudy.)
That is strange. They are 'normal' custom nodes without any mods. Of course you have closed and restarted ComfyUI, I guess ... what more you can try is to clear your browser cache. It depends on which browser you use what the proper shortcut is, may have to Google that.
@obbychannel I just found out what the prob was: renaming the file.
Can you please tell me where you got the cpu etc..gauge in the menu side bar? I must have it!
Those are included in the 'Crystools'.
Is it possible to put the workflows in a git repository? I can do that if you allow me.
Yes, to put them on Github is on my TODO list.
Done: github.com/RudyB24/ComfyUI_Workflows/tree/main
missing "text box" node anyone know where to get this installed??
It's part of the Rgthree node pack. You can also change the 3 inputs it is connected to into widgets and then delete the node, since it's only function is to provide a filename for saved images.
I also have 8GB VRAM.😢 Every image needs more effort in fine tuning the WF to gain as much performance as possible from my machine.
Turbo Checkpoints can help gain speed. I use Dreamshaper Turbo with great pleasure, I think it generates great images.
I always get two oder more girls on the picture -_-
When i lauch your workflow, i get a bunch of errors ... Prompt outputs failed validation
SDXLAspectRatio:
- Value not in list: aspectRatio: '1024 x 1024' not in ['1:1 - 1024x1024 square', '2:3 - 832x1216 portrait', '3:4 - 896x1152 portrait', '5:8 - 768x1216 portrait', '9:16 - 768x1344 portrait', '9:19 - 704x1472 portrait', '9:21 - 640x1536 portrait', '3:2 - 1216x832 landscape', '4:3 - 1152x896 landscape', '8:5 - 1216x768 landscape', '16:9 - 1344x768 landscape', '19:9 - 1472x704 landscape', '21:9 - 1536x640 landscape']
SDXLPromptStyler:
- Value not in list: style: 'Neg 2' not in (list of length 107)
- Value not in list: log_prompt: 'False' not in ['No', 'Yes'] ... ... ... hum hum 🤨
That is due to the changed list of image sizes that I use. The Readme file in the download folder has more info. Otherwise you can just selstc an image size that is available in your dropdown.
github.com/throttlekitty/SDXLCustomAspectRatio
Thanks for your feedback, it's appreciated.