If you find the video useful and would like to tip you can buy me some electricity for all these image generations. It is greatly appreciated! ko-fi.com/kleebztech
Ok i liked this umbrella challenge! Very simple when using your earlier tutorials! I had used the same seed and entered into prompt "woman walking in rain with wet hair" Negative prompt " (umbrella:1.5), cover". As you mentioned in early videos using a thesaurus for alternative words helps a lot, thank you Kleebz
Oh then you may want another tip... "default_max_image_number": 1000 can be added to your config.txt file so you can have even more. lol. I actually use 100 as my default but have used 1000 in the past. I am not at the computer to paste the link to the video but I do have one on customizing if you are not familiar with modifying that file.
@@KLEEBZTECH I have not edited the config file, but I will look for it and shouldn't have any problem. If you have a video of it with other tips, just let me know when you can. I will pull up your channel and take a look after work. Thank you.
Great tutorial! I was able to do this by describing everything but the rain and it came out great with the following prompt. - wet woman walking on a city street with wet hair and clothes on a dark cloudy day
I tried over a hundred in a row with different seeds. It really comes down to just writing a good normal prompt and it will be gone. Continuous generation I actually discovered while making the video. 😂
Yes, it is a combination of positive and negative prompts to achieve the desired effects. Seed make a difference, but so will the dataset used to train the checkpoint and any loras used. @@KLEEBZTECH
I tried to replicate your results and got exactly the same - couldn’t get rid of the umbrella. However if I changed the prompt to ‘a woman walking in the rain with her hands in her pockets” the umbrella disappeared. I guess it comes down to reforming the prompt to counteract the item you want to get rid of.
Yup. Always better to use the regular prompt to get results rather than depend on the negative. Funny thing is even when I had her do things like carry other stuff in her hands there would often be an umbrellla just floating. Also found that the closer to the camera the less likely for one. Further away you are more likely. Have her dancing in the rain instead of walking and no umbrella.
Thx for all these intresting videos for Fooocus, really appreciate them and gets me thinking! As you hinting at around 8 min mark, to put more into the positive prompt. I would say at times, for the positive prompt, try for example to describe, lets say the reality that happens in your wanted image. (Maybe also works for the negative prompt...) For your first example. add a description of wet hair, and the umbrella never gets created for the woman!, but can at times be visible on other people in the image. a woman walking in the rain, wet long black hair So thinking how to describe something, so the unwanted thing have no logic reason to be created in the first place! If the umbrella is created, there can't be wet hair :), ofcourse in AI anything is possible and sure happens :) It really is a fun challenge!
ah ok the wonderful world of AI It's so odd, that different ppl with same type of workflow gets the outcome so different. I just did maybe all in all 50 fast tests directly when I saw your video, and I didn't get a single umbrella, not even a hint of one near the woman :) I got just a couple on other ppl if the there were other ppl in the image. But then I added, people with wet hair in the background on a city street, and umbrellas were gone :) But hey, the developement of AI is so extremely fast, I don't think I have seen another field in many decades that have this, I am sure it will only accelerate even further as time goes :) "what a time to be alive" :)
I went and tried your exact phrase and did get less umbrellas although ended up with much more zoomed in shots. I had phrased it a little different. Once I try to force it back to wide angle then the umbrellas start coming back. But it does come back to the idea of getting what you want with the regular prompt instead of trying to use the negative. Just have to be creative.
Did over 100 random ones. And yes usually the best approach is to try to use the regular prompt to get what you want. I was looking for examples to show that the negative prompt is not usually the best way to go and you want to use the regular prompt preferably. Just like with a house on a hill it would be best to mention an empty field instead of using negative prompt to not have trees.
Thank you! There is some info on GitHub but also much comes down to trial and error and lots of testing. Also reading many of the discussions on GitHub.
Another good way to compare is to put the images to compare in the same folder with nothing else and you can flip between them to notice any very subtle changes. That is useful when the changes are very slight.
So to summarize, while in some cases short negatives might have an impact with some models and some seeds, for the most part the answer is "usually no"?
So far with my testing some things do and some don't. But everything will have some sort of impact even if not the intended. Which is why I prefer not to include unless I know it does something.
Do you unselect Foocus styles by default (V2, Enhance) for the tests ? If not, you don't know what text is exactly in your final prompt. And by the way, there are some negative embeddings available on Civitai.
If you don't change the prompt and any settings yes. But otherwise no. Imagine if you were to change the prompt to walking on the beach on a sunny day. It would be totally different.
It will influence the picture but blocking things completely it doesn't do well with. I find it's good for changing the size or amount of something. Usually you have to give it extra weight.
The best way to test if they're being ignored completely is to turn off random seed. Then generate an image with nothing in the negative prompt and then do it again with the negative prompt filled in. If they look the same double check that they actually are by opening up one of the images in the photo app or something and flipping back and forth between the two. Even if slight differences you will notice the shift in pixels.
Then try different things and only have one thing in the negative prompt when testing. And apply extra weight to it. Something that will affect the size I find is often a good thing to test. Sometimes you have to rework your prompt so it doesn't need to rely on the negative.
@@dunkydog1676 or maybe fooocus just isn't as robust as comfy and auto1111 so it's obvious you're the one with the weak creativity especially after seeing the video on your page...
If you find the video useful and would like to tip you can buy me some electricity for all these image generations. It is greatly appreciated! ko-fi.com/kleebztech
Ok i liked this umbrella challenge! Very simple when using your earlier tutorials! I had used the same seed and entered into prompt "woman walking in rain with wet hair" Negative prompt " (umbrella:1.5), cover". As you mentioned in early videos using a thesaurus for alternative words helps a lot, thank you Kleebz
That generate forever info was a tip nobody else has brought up. Thank you! Sometimes, 30 just isn't enough to get what you want.
Oh then you may want another tip... "default_max_image_number": 1000 can be added to your config.txt file so you can have even more. lol. I actually use 100 as my default but have used 1000 in the past. I am not at the computer to paste the link to the video but I do have one on customizing if you are not familiar with modifying that file.
@@KLEEBZTECH I have not edited the config file, but I will look for it and shouldn't have any problem. If you have a video of it with other tips, just let me know when you can. I will pull up your channel and take a look after work. Thank you.
Great tutorial! I was able to do this by describing everything but the rain and it came out great with the following prompt. - wet woman walking on a city street with wet hair and clothes on a dark cloudy day
Your test may have been biased by the seed you fixed, as the seed itself may include an umbrella. Great tiip on the continuous generation.
I tried over a hundred in a row with different seeds. It really comes down to just writing a good normal prompt and it will be gone. Continuous generation I actually discovered while making the video. 😂
Yes, it is a combination of positive and negative prompts to achieve the desired effects. Seed make a difference, but so will the dataset used to train the checkpoint and any loras used. @@KLEEBZTECH
I tried to replicate your results and got exactly the same - couldn’t get rid of the umbrella. However if I changed the prompt to ‘a woman walking in the rain with her hands in her pockets” the umbrella disappeared. I guess it comes down to reforming the prompt to counteract the item you want to get rid of.
Yup. Always better to use the regular prompt to get results rather than depend on the negative. Funny thing is even when I had her do things like carry other stuff in her hands there would often be an umbrellla just floating. Also found that the closer to the camera the less likely for one. Further away you are more likely. Have her dancing in the rain instead of walking and no umbrella.
Thx for all these intresting videos for Fooocus, really appreciate them and gets me thinking!
As you hinting at around 8 min mark, to put more into the positive prompt.
I would say at times, for the positive prompt, try for example to describe, lets say the reality that happens in your wanted image. (Maybe also works for the negative prompt...)
For your first example. add a description of wet hair, and the umbrella never gets created for the woman!, but can at times be visible on other people in the image.
a woman walking in the rain, wet long black hair
So thinking how to describe something, so the unwanted thing have no logic reason to be created in the first place!
If the umbrella is created, there can't be wet hair :), ofcourse in AI anything is possible and sure happens :)
It really is a fun challenge!
I agree... But the wet hair didn't impact the results when I tried. I was thinking that as well. And often that will work.
ah ok the wonderful world of AI
It's so odd, that different ppl with same type of workflow gets the outcome so different.
I just did maybe all in all 50 fast tests directly when I saw your video, and I didn't get a single umbrella, not even a hint of one near the woman :)
I got just a couple on other ppl if the there were other ppl in the image.
But then I added, people with wet hair in the background on a city street, and umbrellas were gone :)
But hey, the developement of AI is so extremely fast, I don't think I have seen another field in many decades that have this, I am sure it will only accelerate even further as time goes :) "what a time to be alive" :)
I went and tried your exact phrase and did get less umbrellas although ended up with much more zoomed in shots. I had phrased it a little different. Once I try to force it back to wide angle then the umbrellas start coming back. But it does come back to the idea of getting what you want with the regular prompt instead of trying to use the negative. Just have to be creative.
Will the negative prompt work with a random seed? How about using the prompt "Woman getting wet walking in the rain"?
Did over 100 random ones. And yes usually the best approach is to try to use the regular prompt to get what you want. I was looking for examples to show that the negative prompt is not usually the best way to go and you want to use the regular prompt preferably. Just like with a house on a hill it would be best to mention an empty field instead of using negative prompt to not have trees.
great turtorial, how you know all the shortcut's? gives a manuel?, i was on github and this maunal guide don´t have this informations
Thank you! There is some info on GitHub but also much comes down to trial and error and lots of testing. Also reading many of the discussions on GitHub.
@@KLEEBZTECH thank you☺
Nice tutorial. But what did you click to get the comparison windows?
That is the Photos app for Windows. You can click on the bottom left corner to bring up the filmstrip and then check off the ones to compare.
Another good way to compare is to put the images to compare in the same folder with nothing else and you can flip between them to notice any very subtle changes. That is useful when the changes are very slight.
So to summarize, while in some cases short negatives might have an impact with some models and some seeds, for the most part the answer is "usually no"?
So far with my testing some things do and some don't. But everything will have some sort of impact even if not the intended. Which is why I prefer not to include unless I know it does something.
Excelente explicación!😁
If I would treat it as a riddle, the solution would be: put in the negative prompt:
hat, umbrella, umbrellas
Try it.
Won't usually work. Even with added weight. The best solution is to create a better regular prompt. Mention soaking wet or something similar.
Do you unselect Foocus styles by default (V2, Enhance) for the tests ? If not, you don't know what text is exactly in your final prompt.
And by the way, there are some negative embeddings available on Civitai.
Yes when testing I usually do with and without. Best to limit what impacts the results. Same as limiting any extra text.
dosent the seed play a role in consistent generation? by keeping the seed the same dosent that make sure that the umbrella will ALWAYS be there?
If you don't change the prompt and any settings yes. But otherwise no. Imagine if you were to change the prompt to walking on the beach on a sunny day. It would be totally different.
@@KLEEBZTECH ah ok
so using the typical negative prompts like "bad hands, deformed bodies, ..." does not have an impact?
Very little or not at all.
Negative prompts never made a difference to me, I have tried sometimes and it never blocked what I put there.
It will influence the picture but blocking things completely it doesn't do well with. I find it's good for changing the size or amount of something. Usually you have to give it extra weight.
@@KLEEBZTECH I'll try that, I haven't tried this weight thing. Thanks
Maybe you should try " a woman in the rain without an umbrella" in your prompt?
Mentioning umbrella in the regular prompt would make it worse. It doesn't understand without which is the reason for negative prompts.
@@KLEEBZTECH Not worth a try?
Everything is worth a try. 👍
So here's how to get rid of umbrella: a woman,rain(prompt), umbrella (negative prompt).Random has to be checked@@KLEEBZTECH
for me it works with negative prompt "no umbrella" at the first try. only with "umbrella" it dont works.
lol. Figures.
I got it with "hat, umbrella, umbrellas" but this works too!
Great
Cant wait for me to access my PC again my wife hardly ever lets me use my PC. Played with foookus for an hour.
it feels like negative prompts are being ignored in my generations maybe the problem is the sampler
The best way to test if they're being ignored completely is to turn off random seed. Then generate an image with nothing in the negative prompt and then do it again with the negative prompt filled in. If they look the same double check that they actually are by opening up one of the images in the photo app or something and flipping back and forth between the two. Even if slight differences you will notice the shift in pixels.
Then try different things and only have one thing in the negative prompt when testing. And apply extra weight to it. Something that will affect the size I find is often a good thing to test. Sometimes you have to rework your prompt so it doesn't need to rely on the negative.
Just type: a woman walking in the rain without an umbrella lol
lol. Mentioning umbrella will just add one.
@@KLEEBZTECH i found a solution
dont use fooocus. i typed black man and i get a white guy. racist ai
fooocus is so limited. it's great for getting started though.
I guess that depends on what you need. I do not find it that limiting if you enable some of the advanced stuff. Much easier and less time consuming.
Maybe your creativity is weak
@@dunkydog1676 or maybe fooocus just isn't as robust as comfy and auto1111 so it's obvious you're the one with the weak creativity especially after seeing the video on your page...
@@KLEEBZTECH as stated it's still the less robust of the other offerings.
@@dunkydog1676 yes after watching your video i can see how weak my creativity is. 🤣