Thanks for the info. I played around a little with Playground through their site and one thing I liked was it was great for very short, random prompts and parsed them very well. One of my songs has "cosmic horror" generated using Playground's parser and it gave me some really nice, very odd and quite creative images just from those two words. Seeing your Octopus Man test reminded me of how well it did. I will definitely be trying this!
What a great video. Thank you for that. And I admit, I am often so captivated and focused that I even forget to like (I always catch up on that)😄. It's so refreshing that with you, not everything is described as mind-blowing, sensational, unprecedented, revolutionary, etc. 🙃
Yeah it snuck in with the last small update. And if wildcards didn't work I would really not be happy testing since that is how I run my test prompts. In theory those should work with anything in Fooocus.
@@KLEEBZTECH I have found that if you ONLY select one style (uncheck everything else) it does make subtle changes towards that style but nothing like normal. I am creating macro shots of small alien creatures and after 2 to 300 images... I really like it for that. Sharp and crisp photos. Will try something else tomorrow but it is definitely cool to use for some things. You can really hit a wall when trying to do something like an elf girl and unable to use any loras though. Just can't pull in the pointed ears and seems very unrealistic when going for photo with a human vs a creature. I was using long prompts. Maybe I should test much shorter ones for going for humans.
Great video, btw - I often have trouble with people's natural eyes on Fooocus when I generate photos of people. Sometimes it's OK, sometimes it's very artificial, regardless of the model...
Yes eyes are one of the things that is often messed up with any AI software. Usually because training data does not have enough detail unless very close. Why I usually use the improve detail function in the inpainting area to improve those things after I upscale and such.
I am using fooocus primary as a inpainting and controlnet tool for photorealistic images, so playground is useless for me. But thank you very much for information.
PG 2.5 has a quite nice contrast, but bit of a messy output. Sure, it's probably better than sdxl base, then again what's not? I don't agree with their claim of superior aesthetics either. Again, maybe better than sdxl base, but there are literally at least a dozen trained models that are objectively better than Playground 2.5. Again, it's just an another research paper hype crap with lots of smoke without fire. Just like FreeU was.
hey great videos! do you have any tips for generating consistent stills from video stills? i have tried putting up pyracanny&imageprompt to 80% duration with weight 1, even with changing the seed to be the same seed for each frame the background can change a lot
@@Skunkmail have you tried using the vary upscale section? With a decent prompt and adjusting an option in the debug menu that I can't recall the name of and not near computer. Has to do with overriding the value for vary. Don't know if it mentions denoise or not.
@@Skunkmail Forced Overwrite of Denoising Strength of "Vary" is the setting I was talking about. I am not positive it will do what you want since I am not 100% what you are doing.
@@KLEEBZTECH this was perfect btw, seems like below 0.5 and it returns really close to the original. i feel like a bit of a fool trying to get cpds/pyracanny/imageprompt trying to do what this would have done for me :p
With Playground it is limited to just the prompting. You can in theory use inpainting if you go into the debug menu and turn off the Fooocus inpainting engine but does not work great. Very hit or miss.
No that will not change. I don't think anyone has developed those tools for Playground and even if they did the amount of work to implement would probably mean it not happening. It is just not popular enough.
I am so confused... How do people get clean results... No matter what i do i get weird lines or shitty weird details especially in foliage... And the new enhance feature does not work...
I have tested some of the Pony models and they can work but are a little tricky to get good results. I have not used them much though. Really need to understand how the prompting works from what I know and probably want to disable any styles.
Be sure to read the information on prompting to get results with Pony. It is very specific about using certain terms or else you will get garbage. For example You really need to start the prompt with "score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, " as shown on the CivitAI site.
@@glaudeo2908 U can use it, 24 steps enough, cfg higher than 4 will make img more contrast and vibrant, inpainting works good only when in dev options U take "none" (like in this video), Pyro and cdps working unstable, so for better result I place both with low parameters and its work better for me.
Usually Fooocus should auto update. You may need to install a fresh copy. Or are you using it in something like Stability Matrix which has an update button?
It should update itself unless there is an issue with your install. If any code has been modified then you most likely will not get updated. Often if any issues it can be easier to download a fresh install and have it update and then move things like your config and any other things like models. I personally keep my models in different locations since I have several installs and then point to the folders in the config so when I install fresh it only takes a couple minutes to get back running.
I get the same image error in invoke ui when I install sd1.5 models with size greater than 2gb, like with cetus mix whalefall2, it's around 4gb, I just don't know what to do
Have you checked to see if any specific recommended settings for those? I find when that happens it often has specific needs like a specific scheduler.
Since I am a noob, how do I download and install this new version if I already have the old one... Any comand prompt ? Or I need to do it manually? I am really lost on it 😢
Yikes. They have been on version 2 for over 9 months now. Go to the Github and download and install fresh in a new folder. Best option. github.com/lllyasviel/Fooocus
Nothing really that I have found that does well. Most of your online ones are not going to be truly free because of the resources needed. And I have not seen any open source offline ones that do well. But I don't follow the video space a ton right now so I could be wrong. I know the new Luma one online does decent but only has limited free generations per month.
![SahBabii - Viking [Official Music Video]](http://i.ytimg.com/vi/D37cL92JX0I/mqdefault.jpg)
Thanks for the info.
I played around a little with Playground through their site and one thing I liked was it was great for very short, random prompts and parsed them very well. One of my songs has "cosmic horror" generated using Playground's parser and it gave me some really nice, very odd and quite creative images just from those two words. Seeing your Octopus Man test reminded me of how well it did. I will definitely be trying this!
I was out mowing the lawn today and listening to my RUclips music when one of your songs came on. lol.
@@KLEEBZTECH hah! Double weirdness, I was also mowing the lawn today. 😆
Your videos are great. Look forward to seeing if Fooocus eventually supports SD3
I expect that it will.
What a great video. Thank you for that. And I admit, I am often so captivated and focused that I even forget to like (I always catch up on that)😄. It's so refreshing that with you, not everything is described as mind-blowing, sensational, unprecedented, revolutionary, etc. 🙃
You are so welcome! And thanks for the feedback.
@@KLEEBZTECH I keep getting "Error unsupported image type", any help with this?
@@TheWuCepticon1981 You will have to be more specific. Where are you getting that error and what are you doing when you do?
Great news here. Thanks for sharing this update of fooocus.
You bet!
Wildcards work, great video, I had no idea this was in there so thanks
Yeah it snuck in with the last small update. And if wildcards didn't work I would really not be happy testing since that is how I run my test prompts. In theory those should work with anything in Fooocus.
I'll continue to use SDXL for now, but thanks a lot for the detailed explanation nonetheless!
I will for most things as well but I can see using for certain things.
Very informative, Thanks! Let's see if we can get something better out of it than in ComfyUI, as it lacks the specific sampler.
Great video, lots of good info! Thanks for sharing!
Thanks for watching!
I love some of the things I am doing with playground 2.5. It is giving much better results for what I am trying out this weekend for sure.
Good to know. I find it is good for some things.
@@KLEEBZTECH I have found that if you ONLY select one style (uncheck everything else) it does make subtle changes towards that style but nothing like normal. I am creating macro shots of small alien creatures and after 2 to 300 images... I really like it for that. Sharp and crisp photos. Will try something else tomorrow but it is definitely cool to use for some things.
You can really hit a wall when trying to do something like an elf girl and unable to use any loras though. Just can't pull in the pointed ears and seems very unrealistic when going for photo with a human vs a creature.
I was using long prompts. Maybe I should test much shorter ones for going for humans.
macro shots of small alien creatures sounds interesting!
Great video, btw - I often have trouble with people's natural eyes on Fooocus when I generate photos of people. Sometimes it's OK, sometimes it's very artificial, regardless of the model...
Yes eyes are one of the things that is often messed up with any AI software. Usually because training data does not have enough detail unless very close. Why I usually use the improve detail function in the inpainting area to improve those things after I upscale and such.
Great video as usual. I hope Flux comes to Fooocus, its supposedly the hotest new model.
There is certainly a chance if it takes off and gets more abilities.
Thanks man great video
Glad you enjoyed it!
Wow, Playground is ideal model for creating art, not realism, but it great
I am using fooocus primary as a inpainting and controlnet tool for photorealistic images, so playground is useless for me. But thank you very much for information.
PG 2.5 has a quite nice contrast, but bit of a messy output. Sure, it's probably better than sdxl base, then again what's not? I don't agree with their claim of superior aesthetics either. Again, maybe better than sdxl base, but there are literally at least a dozen trained models that are objectively better than Playground 2.5. Again, it's just an another research paper hype crap with lots of smoke without fire. Just like FreeU was.
I find it is good a certain things but overall I don't think it would replace most of what I use.
But generating with Playground and upscaling with something else can lead to some decent results.
hey great videos! do you have any tips for generating consistent stills from video stills? i have tried putting up pyracanny&imageprompt to 80% duration with weight 1, even with changing the seed to be the same seed for each frame the background can change a lot
@@Skunkmail have you tried using the vary upscale section? With a decent prompt and adjusting an option in the debug menu that I can't recall the name of and not near computer. Has to do with overriding the value for vary. Don't know if it mentions denoise or not.
@@KLEEBZTECH interesting, i haven't! thanks for the response
@@Skunkmail Forced Overwrite of Denoising Strength of "Vary" is the setting I was talking about. I am not positive it will do what you want since I am not 100% what you are doing.
@@KLEEBZTECH this was perfect btw, seems like below 0.5 and it returns really close to the original. i feel like a bit of a fool trying to get cpds/pyracanny/imageprompt trying to do what this would have done for me :p
Yeah but trying all different things is how you find new ways of doing things.
Is this available when you use foocus in colab on browser?
I just tested and it works in Colab. I did launch it with the playground_v2.5 preset so it downloaded the required model first to make it easier.
@@KLEEBZTECH actually i am asking as we use it without downloading, using Google colab fooocus url
You may have missed the first part of my comment. Yes I tested and works with Colab.
Hi! Do you think will be possible to run Deforum (Stable Diffusion) on Fooocus?
I would not expect that to be possible since video is not anything they have planned.
I know this is now in Fooocus, the problem is it does not utilize any of the advance features such as faceswap or pyracany, any thoughts?
With Playground it is limited to just the prompting. You can in theory use inpainting if you go into the debug menu and turn off the Fooocus inpainting engine but does not work great. Very hit or miss.
@@KLEEBZTECH Do you know if they are going to fix that, it's a pity it can't use the extras
No that will not change. I don't think anyone has developed those tools for Playground and even if they did the amount of work to implement would probably mean it not happening. It is just not popular enough.
I am so confused... How do people get clean results... No matter what i do i get weird lines or shitty weird details especially in foliage... And the new enhance feature does not work...
The person who runs this group has some guides on using playground. facebook.com/groups/fooocus/?ref=share&mibextid=lOuIew
As for the enhance feature I don't think it'll work with playground model
Should i use lora sd_xl_offset with playground 2.5 or use no lora at all?
None.
@@KLEEBZTECH Tks! enjoying all your videos n tips!
I still can't figure out why my config and flash file keeps changing.
Have you tried posting on the Fooocus Github discussion section? It is pretty active there for troubleshooting.
@@KLEEBZTECH will have to find it and see if they can help.
I just updated and it now does not change my changes.
Can I use Pony V6 XL in Fooocus?
I have tested some of the Pony models and they can work but are a little tricky to get good results. I have not used them much though. Really need to understand how the prompting works from what I know and probably want to disable any styles.
Be sure to read the information on prompting to get results with Pony. It is very specific about using certain terms or else you will get garbage. For example You really need to start the prompt with "score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, " as shown on the CivitAI site.
@@KLEEBZTECH Thank you
@@glaudeo2908 U can use it, 24 steps enough, cfg higher than 4 will make img more contrast and vibrant, inpainting works good only when in dev options U take "none" (like in this video), Pyro and cdps working unstable, so for better result I place both with low parameters and its work better for me.
Question could I download the model and use it in ForgeUI?
I have not tried so can't say if it supports it. I don't use ForgeUI often.
i'm using focus V2, how can i update that to this latest version
What version shows in browser tab?
@@KLEEBZTECH V2 2-2-831 which is released on Dec 13, 2023
Usually Fooocus should auto update. You may need to install a fresh copy. Or are you using it in something like Stability Matrix which has an update button?
Does fooocus ai updates itself or have to download latest version manually?
It should update itself unless there is an issue with your install. If any code has been modified then you most likely will not get updated. Often if any issues it can be easier to download a fresh install and have it update and then move things like your config and any other things like models. I personally keep my models in different locations since I have several installs and then point to the folders in the config so when I install fresh it only takes a couple minutes to get back running.
I get the same image error in invoke ui when I install sd1.5 models with size greater than 2gb, like with cetus mix whalefall2, it's around 4gb, I just don't know what to do
what do you mean by image error?
@@KLEEBZTECH I get a colored noisy image
@@KLEEBZTECH the same results you got when you changed the base model to playground from the model selector without changing to the proper scheduler
Have you checked to see if any specific recommended settings for those? I find when that happens it often has specific needs like a specific scheduler.
@@KLEEBZTECH I tried the recommended scheduler with no success, I actually found a pruned 2gb version that actually worked for me
Since I am a noob, how do I download and install this new version if I already have the old one... Any comand prompt ? Or I need to do it manually? I am really lost on it 😢
Updates are usually automatic. It should show the version number in the browser tab for example. 2.4.3 is the current latest version.
@@KLEEBZTECH yes but mine is not 😔 I am still at 1.8.4, any recommendations how to upload?
Yikes. They have been on version 2 for over 9 months now. Go to the Github and download and install fresh in a new folder. Best option. github.com/lllyasviel/Fooocus
Thank you highly Appreciate it @@KLEEBZTECH
❤❤❤
😁👍
Is in Google colab?
Should be but I have not tested to see how it works in Colab.
Try Kaggle
Hey is there any free Ai text to video generator?
Nothing really that I have found that does well. Most of your online ones are not going to be truly free because of the resources needed. And I have not seen any open source offline ones that do well. But I don't follow the video space a ton right now so I could be wrong. I know the new Luma one online does decent but only has limited free generations per month.
This is the Luma one I mention. Free to try but not cheap to do many. lumalabs.ai/dream-machine/creations
@@KLEEBZTECH Thanks for such a prompt reply.
Use the same seed for exact comparison.
When comparing things with the same model I use the same seed.
1st
Faceswap bug? Resolved?
Nothing changed with faceswap.