Fooocus 2.3.0 - New Preset Feature & Lightning Performance (Stable Diffusion)
HTML-код
- Опубликовано: 8 июн 2024
- Another UPDATE! that brings even more quality of life updates to Fooocus including a new preset selection panel, a lighting performance setting, wildcards in order checkbox, and more. Improving more on Fooocus, Stable Diffusions best SDXL AI Image Generator. The Inpaint\Outpaint is unmatched!
Official Notes: github.com/lllyasviel/Fooocus...
If you want to help support this channel! www.buymeacoffee.com/jumpintoai
Check Out my Full Fooocus Tutorial Playlist!: • Fooocus!
Want the easiest SDXL Stable Diffusion software? Go to the Fooocus download page: github.com/lllyasviel/Fooocus
-----
00:00 Intro
00:13 New Features
02:05 Troubleshooting
----- 
Наука
Fooocus download page: github.com/lllyasviel/Fooocus
Is there a difference between Fooocus & InstantID? Whitch is better?
Hello, I can only have the 2.1.831 version. Is it normal?
@@henridusasca8077 it should update automatically. You didn't change any of the run.bat files? They have the command arg that starts the update. Has it never updated? Perhaps firewall is blocking it?
@@JumpIntoAI ho ! OK.
Actually, when I clicked on the bat, it opens the commande windows, but nothing happened then. I assume my graphic card isn't powerfull enough.
@@henridusasca8077 Ahh, yes this can be the case if low Vram and or an AMD GPU.
It depends on the error given.
You are the best best Fooocus focused channel my guy, just thought you should know. Thanks for distilling this great software for the rest of us.
Thanks! I started this all on a whim expecting it to go nowhere. Happy to be wrong.
I like watching your videos. Easy to follow.
Awesome, thanks !
It's nice to see "read wildcards in order" feature... Previously I've used to create special wildcards with arrays written in them in order to get all items in that wildcard.
Agreed! I loved wildcards but being able to order them how I want really opens up the doors on some things. I can now setup hundreds or even thousands of generations to run. My GPU will be working overtime while i sleep and work.
Thanks more videos about Fooocus please, like
When I click on run and the download is complete, the foocus window does not open, and I do not know why. It only hangs at 100%, and there is nothing underneath it, like the rest of the videos.
Thanks for the update on the fooocus update 🔽,is there a way to batch upscale a whole folder of images?
No, it can't batch upscale.
Finally!
2:15 How does the forest in the background look so realistic? Are there any settings for this?
Nope that was luck and probably due to the 3 default fooocus styles. My wildcard for that only describes the woman.
That was on the dreamshaper lightning model
What's the best way to achieve results similar to headshotpro using Foocus? Take some irl photos of yourself from different angles and different lightning conditions, create a big grid and then do some manipulation? I'd like to achieve both something realistic and something more anime-like.
Hi. Thank you for the video! One question:
Hi, there’s no way of adding multiple word prompts with commas as wildcards like:
Red hair, long hair, curly hair
Pink hair, short hair, straight hair
So the red hair will always be long and curly and the pink hair always short and straight?
yes, actually entering that into a wildcard just like that will work.
Brother when I upload my real photo to Image prompt when I generate the image not the same face of mine
I already Uncheck random but result not my real face
Do you know how to run it on kaggle? Thanks for your great content
1:30 Is there a way to use the Lightning models in Fooocus? For some reason I thought the earlier versions of Fooocus didn’t allow Lightning XL models to use.
You can use lightning checkpoints in fooocus, you just need to adjust the settings for them to work correctly.
on their civitai page it will tell you the settings but most of them use a guidance scale of 1 to 2
the steps 4 to 6
and the sampler on dpmpp sde.
I noticed that when choosing the Lightning preset it sets the Guidance Scale = 4, Sampler = dpmpp_2m_sde_gpu, and the sampling steps to -1. In other videos and in lightning model note I've seen the recommended settings as 1.5, dpmpp_sde, and 4. I wonder why they chose their default values? Also, you can run the Lightning preset on a non-lightning model and it still works but probably doesn't give the best results.
The lightning performance setting is meant to be used with non-lightning models, its the 4 step lightning lora that bytedance released and thats how its used.
The settings that are greyed out when using that setting are not reflecting the settings its using. if you look in the command window it actually shows the settings it change and what is used when you switch to the lightning preset and generate an image.
[Parameters] Adaptive CFG = 1.0
[Parameters] Sharpness = 0.0
[Parameters] ADM Scale = 1.0 : 1.0 : 0.0
[Parameters] CFG = 1.0
[Parameters] Sampler = euler - sgm_uniform
[Parameters] Steps = 4 - 4
Hay man, may I ask your pc specs? Your generations are really fast! or its just an editing speed up? I have a 1660 ti 6GB and 0 cores and my goodness it is taking 6 minutes to generate 1 single image...so I'm looking wich will be my new video card. thanks and thanks for the video
most of the generations are sped up for the video, but mine take like 10-15 seconds an image on the speed setting.
4070 ti 12gb
If specifically wanting a card for Stable Diffusion, stick with NVIDIA and go for the highest VRam you can reasonably afford.
Have you tried lightning models? since they use 4-5 steps vs 30, it should improve your times.
@@JumpIntoAI Thanks a lot for your answer! - Yeah I tried with lightning and yeah they're a bit faster but you know nothing spectactular. So yeah, I'll take a look to the Telsa P4 and will see if fits in my motherboard If not I'll go and pick that 4070 that seems to be pretty fast. Thanks again!
Thanks so much for the solution with this **** sai
First of all tnx for video.
Correct for today I use Kaggle for run Fooocus. I see differences between local installation and on Kaggle. I don't see "custom" and "turbo" presets. By additional I don't see size "1536x1536". Is it by default or You changed config file?
Those are all my creations and additions. resolutions added in config.txt and the presets are .jsons I created in the fooocus\presets folder.
Hey, I've noticed SDXL inpaint models dont work for me in fooocus, is it the same for you too? I tried with RealVisv3inpaint model
No they aren't compatible, But the Inpaint model built into Fooocus is probably one of the best there is.
I've noticed that before update inpaint worked waay better. Now on the same images works way worse.
@@procentpisz Please share your settings.
While using the lightning mode, is it possible to add negative prompts at the positive prompt box?
the lightning performance setting disables the negative prompt similar to the extreme speed setting. I believe the reason for this is the lightning setting uses a low CFG (guidance scale) of 1. and at 1 the negative prompt is ignored.
Thank you! Could you please tell me how you added the categories to the Preset and Performance sections?
Please watch previous episodes in this channel.
In previous episodes, version 2.3.0 was not covered, so the Preset and Performance categories were not discussed.@@hmm-m
The custom presets are in the fooocus\presets folder. You can create a copy of any of the .jsons in there and alter the settings inside them however you like.
ruclips.net/video/DRcxsqnhjws/видео.html
This shows how to make a preset out of dreamshaper lightning model but it can be applied to anything once you understand the basics.
Any .json in the preset folder will show up in the new preset section in fooocus.
The performance section settings are all part of Fooocus itself.
@@b.s.design Presets were discussed before and they existed before the latest update.
Thank You!@@JumpIntoAI
how do we know our current version, and how do we upgrade to the new version .. can you run a tutorial about the update and version up for newest update
when loading the program you can see the version mentioned in the command window "Fooocus version: 2.3.1"
Also the browser tab that fooocus is open in should list the version next to its name.
🎯 Key Takeaways for quick navigation:
00:28 *Focus 2.2.0 update introduces new output formats: JPG and WebP, alongside PNG.*
01:09 *Checkboxes added for activating/deactivating Luras without changing weights individually.*
01:36 *Default Max weights for Luras can now go beyond +/-2, helpful for certain models like Detail Tweaker Xcel.*
02:49 *New debug options include "Disable Intermediate Results" and "Disable Seed Increment".*
03:43 *Introduction of support for arrays in prompts allows for quick generation of multiple variations of images.*
06:17 *Ability to save metadata to images for easy reloading of settings used to create them.*
08:08 *Users can now specify additional paths for models and Luras in the config file, enabling organization and easier access.*
Made with HARPA AI
🎯 Key Takeaways for quick navigation:
Version 2.3 of Focus brings new preset selection panel and lightning performance setting.
Lightning setting allows using regular models with lightning speed, similar to extreme speed.
Style tab reformatted for easier reading with two columns.
Debug menu now supports reading wildcards in order and subdirectories for wildcards.
Some presets might need adjustments due to changes in the update.
Issues may occur when switching between presets, requiring manual adjustments.
Clicking on a preset that requires a model download starts the download immediately; workaround involves moving or deleting the file.
Removing a preset while Focus is open requires hitting the refresh button to remove it.
Bug reported with config text parameters not coming through correctly.
New presets can be disabled by adding a line to the Run.bat file.
Made with HARPA AI
How does one update it?
It seems multiple wildcards are not working, only the first one in prompt. Anyone else having this issue?
Hmm, multiple wildcards still working ok for me.
Make sure your wildcard.txt file is all lowercase (like fruit.txt vs Fruit.txt) I just noticed some issues with that for me after the update.
How do you get the dark mode appearance?
if you are running Fooocus on your machine and in windows, you can set windows to default apps to darkmode
Settings > Personalization > Colors, and then choose "Dark" under "Choose Your Default App Mode."
For everything else you can add ?__theme=dark to the url, like 127.0.0.1:7860/?__theme=dark
or even add --theme dark to whatever run.bat file you start
like this if you use run_realistic.bat
.\python_embeded\python.exe -s Fooocus\entry_with_update.py --preset realistic --theme dark
i not found file (sai json)
How do I Update the exisitng version of Fooocus?
tx
oh, it automaticaly updates
Hello, I am using Fooocus, watching your videos, thank you for sharing. Amazing, I am seeing in CMD. message like this "You are using gradio version 3.41.2, however version 4.29.0 is available, please upgrade." I downloaded new version. but I don't know how to upgrade.
Fooocus currently uses the older version of Gradio right now. Thats something on the developers side to upgrade. So don't worry.
@@JumpIntoAI Thank you so much!
We want a video on 2x zoom out in fooocus colab❤❤❤❤
How do you activate the dark mode?
if you are running Fooocus on your machine and in windows, you can set windows to default apps to darkmode
Settings > Personalization > Colors, and then choose "Dark" under "Choose Your Default App Mode."
For everything else you can add ?__theme=dark to the url, like 127.0.0.1:7860/?__theme=dark
or even add --theme dark to whatever .bat file you start
like this if you use realistic.bat
.\python_embeded\python.exe -s Fooocus\entry_with_update.py --preset realistic --theme dark
I am the only one whose loras do not appear, even though they are correctly inserted in the preset file?
Mine work but I had to add the new format in my preset file. if you look in the config_modification_tutorial.txt you will see the new "true" parameter with the loras. also the amount of lora default setting lines needs to match the "default_max_lora_number" if you bumped it up from 5-6 you will need to add another line like:
[
true,
"None",
1.0
]
Why do most image generators not have 1920x1080 included? This is the most common format for video and have to keep adding it back after every update.
The image generators aren't trained on that resolution so it's more likely to get image artifacts, multiple limbs, heads, clones, and other issues.
But for me it can be great on landscapes.
And apart from what JumpIntoAI said, it's not really an issue at all. You have outpaint and upscale, you can use it and then crop in a photo editor as you wish.
@@JumpIntoAI this doesn't compute. AI image generators create images based on pixels not images. Yes the images are used to train the models on objects around us but they know nothing of aspect ratios as far as I know. If they did then we would be limited to 4:3 and 3:4 images.
I keep getting runtime error trying to use it in colab
That usually means the free Vram has run out.
how many can you generate before it runs out?@@JumpIntoAI
@@JumpIntoAI how many photos till you run out?
Maybe, maybe, but *config.txt* doesn't work in 2.3.0!(((
I know, super buggy at the moment, there is an issue thread on it and hopefully fixed soon.
@@JumpIntoAI maybe it will not be fix, author delegated params to presets.(
Sad to see the lady loss her leg in third generation better inpaint her before the blood loss kicks in good video do you plan on doing forge ui videos it does help with generation speeds
Yes, I want to transition a little into Forge and cover the things I use it for that Fooocus can't, then go from there. It's just lack of free time that slows me.
And dont worry, I used the random magically appearing strawberry from the other image to heal her leg. But her twisted around hand is another story.
@@JumpIntoAI Here hoping SD 3 reduces those problems especially with holding objects like weapons
Its unfortunate the Faceswap/Inpaint combo STILL doesnt work very well..
why did you make such difficulties??it was better than the previous version
I could not follow because the background music is disturbing.
How does one update it?