- Jan 4, 2022
- 233
- 865
- Jan 4, 2022
- 233
- 865
image have PNG info, where you can check it:
You don't have permission to view the spoiler content.
Log in or register now.
You can use more right prompt , not my "cartoon, Disney style ,line art"
And your prompts inspired me IMG
Last edited:
- Oct 29, 2017
- 812
- 2,167
If you can't do what you want to do with 5 LORAs you are probably doing something wrong.
edit: not meant to be an asshole, I mean it's probably some tool or compositional bad practice.
edit: not meant to be an asshole, I mean it's probably some tool or compositional bad practice.
Last edited:
- Jan 24, 2020
- 1,401
- 3,793
Using many loras tend to make the image into a big mess, a monstrosity. But if you can avoid this, you can always add lora's in the prompt. It's only the Additional Network extension that is limited to 5 afaik. I agree though with fr34aky, try to make the most out of the ones you are using and try to limit yourself to as few as posible. Learn to weight the prompt correctly and you will get much more out of it even before adding lora's.
Example: (huge ass:1.4)
You have obviously put more emphasis on a huge ass.. There's no reason you can't go higher than this, like 1.8 for example.
- Mar 11, 2023
- 46
- 132
Do you have any image examples of the type of 1990s disney art you mean? It'll make it easier to identify which models or styles will do what you want.
Since so many models have been trained on anime/manga - it might actually be easier if you find animes from the 90s that are closer to what you want, and you can generate something based on those styles.
EDIT: You might like to try using this model/checkpoint. I don't vouch for its safety because it's a .cpkt not a .safetensor but lots of people have used it. It was used to make these disney style pieces:
You must be registered to see the links
Link:
You must be registered to see the links
Last edited:
- Jul 2, 2017
- 1,188
- 4,171
Mr-Fox Painting a cell needs a type of ink that changes colours and line art in a distinct way that made them feel a little bit eerie and as soon as they got captured with the camera they became quite noisy. Especially the 90s combined that feel with high saturation. And not just Disney did that.
- Jan 4, 2022
- 233
- 865
Node Based UI
Like a pro!
If you change the last part of the graph only the part you changed and the part that depends on it will be executed.!
You must be registered to see the links
Like a pro!
If you change the last part of the graph only the part you changed and the part that depends on it will be executed.!
Last edited:
- Dec 21, 2020
- 1,523
- 3,589
This frontend looks so good, I even updated the original post with the link to it.
- Dec 21, 2020
- 1,523
- 3,589
A completely random fluke in a set:
I was going rather for:
I was going rather for:
You don't have permission to view the spoiler content.
Log in or register now.
Last edited:
- Jan 24, 2020
- 1,401
- 3,793
I have only tryed
You must be registered to see the links
but I was so underwhelmed that I never tryed it again. So I can't be of much help.You can always use some royalty free stuff, there are many sources for this.
You must be registered to see the links
is fairly well known, or try
You must be registered to see the links
.
- Jan 24, 2020
- 1,401
- 3,793
This song "
You must be registered to see the links
" from freemusicarchive is not half bad. At least some sections are useful.*I had completely forgot about this:
You must be registered to see the links
Check out this song:
You must be registered to see the links
(some sections is pretty good)
Last edited:
- Dec 21, 2020
- 1,523
- 3,589
- Jan 24, 2020
- 1,401
- 3,793
Now you have worked on the prompt and are satisfied with it. You have adjusted the lora's and have found a seed that you like.
The next step is to add some post processing. Go to settings/ post processing and "Enable postprocessing operations in txt2img and img2img tabs ". Choose both GFPGAN and Codeformer. You will then adjust the visibility of GFPGAN to 1 in text2img and do a test generation, do the same with codeformer. You can even try them together. See wich one you like the best if any.
Now it's time to enable hi res fix. Start out with "uppscale by" 1.5 since you are hardware limited and if it works increase to 2.
If this worked continue with 0.1 increases until you get cuda memory error. Now you have found your limit.
Something you can try to increase your limit slightly, is to add
"set PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.6,max_split_size_mb:64" to webui-user.bat
without "" of course.
Different upsamplers gives slightly different results, SwinIR_4x is a good choice
or you can download one from the
You must be registered to see the links
.Next thing is to adjust hi res steps, 10-30 is recomended and then denoising strength, 0.4-0.6 is a good start.
You don't have permission to view the spoiler content.
Log in or register now.
In the end this is only a demo and your "mileage" might vary etc. Euler a doesn't seem to gain much with Hi Res Fix.
We are not only looking for higher resolution but higher "fidelity", sharper image over all and better details etc.
Here I used DPM2 a instead of euler a, Hi Res Fix gives much more gains with other samplers it seems.
Though I didn't reach any perfect result I hope I demonstrated how we can use Hi Res Fix to enhance our images and take it to next level. After this we can still achieve a lot more with photoshop. Take a look on this post I made about this.
You must be registered to see the links
is a great alternative if you don't have photoshop.You might ask. Why bother with Hi Res Fix? Why not simply use a higher resolution?
Here's why:
You don't have permission to view the spoiler content.
Log in or register now.
Last edited:
Model :
Prompt : ((Actress)),((princess)),((nsfw)), Mangalsutra,((topless, nude, naked)), (((45-year-old gorgeous curvy indian women wearing flowers))),((( 45-year-old a beautiful seductive curvy Indian woman wearing flowers ))),(( big breast)),woman with straight hair and wearing a red panty and red blouse with detailed seductive alluring eyes in in a beautiful palace, best quality masterpiece, photorealistic, detailed, 8k, HDR, shallow depth of field, broad light, high contrast, backlighting, bloom, light sparkles, chromatic aberration, sharp focus, RAW color photo,smiling,clothed, in luxurious fantasy castle, decoration with flowers,
You must be registered to see the links
Prompt : ((Actress)),((princess)),((nsfw)), Mangalsutra,((topless, nude, naked)), (((45-year-old gorgeous curvy indian women wearing flowers))),((( 45-year-old a beautiful seductive curvy Indian woman wearing flowers ))),(( big breast)),woman with straight hair and wearing a red panty and red blouse with detailed seductive alluring eyes in in a beautiful palace, best quality masterpiece, photorealistic, detailed, 8k, HDR, shallow depth of field, broad light, high contrast, backlighting, bloom, light sparkles, chromatic aberration, sharp focus, RAW color photo,smiling,clothed, in luxurious fantasy castle, decoration with flowers,
- Nov 11, 2017
- 2,311
- 6,079
Of course, HiRex Fix can alter the actual composition from the original output, depending on denoising and other settings.
I use the Extras tab for when a larger version of an output is needed, sometimes after img2img fixup. By this point, I have different combinations of upscalers and face reformers that help with specific circumstances to handle from the original image.
- Jan 24, 2020
- 1,401
- 3,793
The problem with extra tab is that you loose the texture and definition. Everything becomes smoothed out. So imo hi res fix works much better. Yes the composition is altered very slightly, the reason it changed so much in my demo is because I used a different sampler wich I was very clear about.
- Dec 21, 2020
- 1,523
- 3,589
Edit: solved with: "pip install safetensors". Ahaha FML.
Also, just in case, I am attaching my *.bat file I use to start the gitclone version, instad of running "python main.py" everytime. It includes the "git update", so you are guaranteed to always stay uptodate with whatever the latest version is.
---
Gents, anyone tried installing ComfyUI using git clone? I did, but it aint running safetensors:
Also, just in case, I am attaching my *.bat file I use to start the gitclone version, instad of running "python main.py" everytime. It includes the "git update", so you are guaranteed to always stay uptodate with whatever the latest version is.
---
Gents, anyone tried installing ComfyUI using git clone? I did, but it aint running safetensors:
Code:
E:\Utilities\StableDiffusion\ComfyUI>git pull
Already up to date.
Set vram state to: NORMAL VRAM
Using xformers cross attention
Could not import safetensors, safetensors support disabled.
Set vram state to: NORMAL VRAM
Starting server
To see the GUI go to: http://127.0.0.1:8188
Last edited:
- Dec 21, 2020
- 1,523
- 3,589
It feels like ComfyUI's ability to integrate two checkpoints and area control brings MidJourney's capability to our desktops.
Somehow it feels faster, like twice as fast as WebUI.
But! It feels like I still havent figured out how to do the "restore face".
By the way, it will break your prompts. Yet, I feel this whole pipeline concept is absolutely worth giving it a go:
And I imagine the prompt is not immediately compatible:
Somehow it feels faster, like twice as fast as WebUI.
But! It feels like I still havent figured out how to do the "restore face".
You don't have permission to view the spoiler content.
Log in or register now.
And I imagine the prompt is not immediately compatible:
You don't have permission to view the spoiler content.
Log in or register now.
Last edited: