When I watch your videos I feel like rather than giving us a fish, you are teaching how to catch one. For that I want to thank you for all Fooocus-oriented videos , they are very informative and helpfull for people like me who never tried local AI image generation. As a humble request, can you review some usefull LoRAs that you use? Like that zoom-slider LoRA that you featured in one of your videos?
This question is a little bit strange but here we go: do you think that SD FOOOCUS will add some EMOTE or LTX or SORA functions in the near future? A mean do you think that will can create a woman singing when she's moving or talking? I'm so excited for the infinite possibilities about AI mixing FOOOCUS with the rest of AI text to video coming soon, and i thought that maybe SD will include this functions in their packages. I know that im out of my mind😂. Thabk you for the video. Very well explained.
I don't think any animation type features will ever be added directly. One of the main features of Fooocus is it's light VRam requirements. Adding anything like that would bloat the requirements beyond majority of what the average user has. But It's best to think of it as one step in the workflow. I create my images and inpaint as needed and use those images in my Animatediff/SVD projects.
I would like to ask if anybody else is experiencing "error" issues with the latest version of Fooocus. I've been unable to run this program and that is, the error message informs me that I don't have the latest Nvidea drivers, which I do. I have no idea how to resolve this problem and I'm hoping for some thoughts that may be helpful. Thank you.
Elevenlabs. But it probably sounds good because I've been using my actual voice more now and spot checking with the AI synthesized version when I can't record and need corrections. Trying to move away from it entirely.
I noticed there was nothing saved in yesterdays output. There was a history log but it did not get saved to outputs. I read it could've been due to quality performance over speed. Changed to speed performance and I'm noticing that while there is a history log, there doesn't seem to be anything saved to output. How do I ensure the generations are saved to the output folder thanks
hmm, is this the local version or the google colab online version? I think the online version doesn't save the outputs once the session ends. So you need to download them before you quit. If this is on your computer I am not entirely sure. Did you change the output path at some point in the config.txt?
Yes! you can use any SDXL model as a base model and any SDXL and sd 1.5 as a refiner. Just download them and put them into the fooocus\models\checkpoints folder.
Does anyone know how to make arrays of multiple words ? Like something : [[ (sun,water), (night, fire) ]] i want to alternate between sun water and night fire in this example.
if you use coma as you said it will not count () and take 4 elements in array. you can use wild cards now it has feature so it will read wild card in order or for small prompt you can just use 'and' like [[ sun and water, night and fire]] so it will generate 2 images in total
After I go to Debug Mode, "Save Metadata to Images" is not ticked after I put the two lines to the config file and it doesn't work. Moreover, it messes up my other settings, like the number of images to generate or outputs folder... strange. I put the two lines (metadata - yes, format - fooocus) at the end of the config file, like u do in the vid... // In my opinion you can do vids about major updates of Fooocus, I liked this one very much. The new options would suprise me without this vid that I watched before using new ver of Fooocus.
are these changes in a preset json file or the config.txt file? If using a preset in the command window at the top do you see it saying something like "Load preset C:\Fooocus\presets\custom.json] failed" ?
@@JumpIntoAIThe problem was of course a comma before the added lines, so after closure of previous section by a square bracket... Sorry for trouble...
What operating system and GPU do you have? This does have some requirements which you can see here: github.com/lllyasviel/Fooocus?tab=readme-ov-file#minimal-requirement
@@JumpIntoAI Nah it's fine, it's on me. I spent waaaaaaaaay to much time on generic music platforms over the last few years for corporate videos soundtracks. I now have a kind of trauma when I hear it, they sound all the same 😅. That said, the content of your video was good enough for me to quickly ignore it, which is saying a lot!
I used foocus ai on public link not install in local but I’m always get error connection time out brother ? If I install on local still have this matter or yet ? My pc Core i7 12700kf Ram 32 Gpu 1660S 6Gb Can I run this app on local ?
There is a Colab version on the developer page. colab.research.google.com/github/lllyasviel/Fooocus/blob/main/fooocus_colab.ipynb Keep in mind most advanced features will cause memory to run out if using on a Free colab account.
@@dongytuetinh I think thats the git install version at the top, that colab is updated to the 2.2 version of fooocus. checking myself, it says in the command lines Fooocus version: 2.2.1
lol, well in this the point was to use the same seed using the new array support so it will look similar. But in general short undescriptive prompts will end up like this. And sometimes the default fooocus styles lean into portraits. Better Prompts and pyracanny/CPDS poses is the way to go.
edit one of the run.bat files and add the line --always-cpu after Fooocus\entry_with_update.py so it will look like this: .\python_embeded\python.exe -s Fooocus\entry_with_update.py --always-cpu
@@eveekiviblog7361 On my 1050 Ti it is very slow, So I want to try it on my i5 13400F, the method --always-cpu not working for me. I tried several times, but it give me "the system cannot find the path....error.
@@hanynagy8969 sounds like a path is missing or in there incorrectly, one misplaced comma or a capitalized letter can mess things up, without knowing a lot more details it's hard to help. But if you copy your command window text and create a post on github.com/lllyasviel/Fooocus/issues It would be easier to help.
Thank you for the efforts, it is very well explained. I have a question, I have a clothing store and I would like to create AI models who wear my creations, when I use a prompt image of my dresses, it always generates images with dresses a little different from those that I added to the prompt. I tried the negative prompt but to no avail. please how to make it so that it keeps the same dress with its small details (drawing, embroidery...)? Thanks
The way stable diffusion works, It wont be able to transfer the exact details of an item into an image without some work. I have seen others create a LoRA based on an item, but that might not be efficient if you have multiple items. Another way which is easier with certain products, maybe not clothes. Is taking a good image of the clothes on someone else, and then basically using the inpaint and masking everything else but the clothes themselves and generating a new person wearing said clothes. Obviously this is locked to the lighting and shape of the original image but can work. Products that can be set on a table or simple wearables (like a purse) can be done easier this way.
Fooocus is a great tool, and your videos about it are excellent. I love how organized, informative and well thought out they are. I just want you to know that I really appreciate all the work you put into making these vids. You're my favorite AI-oriented channel by far. Btw, do you know why Fooocus (since the update) has been stuck on "Loading models ..." for several minutes every time I hit generate? Before the update it took like 10 seconds to start the actual generation...
Assuming this is on your computer and not the colab? is it still working just taking longer? If so sounds like the Vram is being used up. Or does it stay stuck? If it stays stuck it might be a file path error, might check the cmd window and see if it failed to load model.
Can you make a video on how to create good vector graphics in Fooocus, like brands, line art, coloring books and vector assets. I think there a great amount of potential in that area that most miss for the oversaturated photorealistic genre.
I agree! The want for creating "models" overshadows a lot of the other capabilities of SD. There are even some amazing tools coming out recently like Layer Diffusion for Forge. www.reddit.com/r/StableDiffusion/comments/1b1w6tt/transparent_image_layer_diffusion_using_latent/ Of course it is gated behind having to use Forge. But something to check out if you haven't. I have a couple half finished projects and one is for logos. It's just lack of time that stops me.
New RUclipsr add on my list with Matt wolfe and theorically media. Thanks! 👍 I hope you'll add me in the near future. I created the first AI photorealistic singer influencer. We are on TikTok and IG right now, coming soon YT and Spotify. With her first album.
Hello. Thank you for the review. Could you suggest the solution: I'm the t-shirt online store owner. I'd like to create images of different people in different poses with my t-shirts (with logo) on them. What ai tool can help me with that?
Sadly, Stable diffusion can't transfer specific items especially with logos into a scene without manipulation (photoshop) and or creating a LoRA for that specific item. There are ways to get products in a scene with stable diffusion but wearables like shirts are the hardest and need the most pre and post processing. Eventually I will get a vid out about this to explain better.
yes! It's a bit hard to find at first but go to advanced>Advanced tab>check developer debug mode>in debug tools tab scroll down till you see "Forced Overwrite of Sampling Step" and put in whatever steps you like. -1 will put it back to auto.
I have an NVIDA 4070 TI 12gb GPU. AMD GPUs struggle a bit since stable diffusion was made for nvidia architecture (cuda) but there are some workarounds You can check out the developer discussion page to see some methods like here: github.com/lllyasviel/Fooocus/discussions/488
you can add /?__theme=dark to the end of the address bar. like 127.0.0.1:7865/?__theme=dark If you are running it locally, you can add --theme dark to the run.bat like this to always load it that way: .\python_embeded\python.exe -s Fooocus\entry_with_update.py --preset realistic --theme dark
They really, really, really have to add an erasure brush to the inpaint outpaint! It makes no sense having to step back 2,3,4,5, times or delete the whole mask, when you could just switch to the erasure bush.....and erase. Save a whole lota time and aggravation IMO.
That's exactly what I'm talking about buddy. It also would be nice if they let you manually change the dimensions in the outpaint (top, bottom, right, left} instead of having to use the default dimensions. I've been trying to send Illyasviel a suggestion or request on these matters. But I'm new to this Github stuff. Do you know how to send a message in the github web site? I'm registered,but how the heck do you send a message (discussions, pull request, issues etc. etc.)????? Thanks for any feedback my friend.
@@neonxp99 I would go to the discussion page on github: github.com/lllyasviel/Fooocus/discussions and post there with "new discussion" . Mashb1t is an incredibly active mod/collaborator there that could answer those questions.
There is a issue in Focus, I can't extract the zip with the .7, after 35% it get stuck, note - I have already installed automatic1111 on my system ??? Can they work together
This shows how to setup a turbo preset: ruclips.net/video/cCS_pBdh7ko/видео.html This shows how to do a lightning preset: ruclips.net/video/DRcxsqnhjws/видео.html But can be used to make any preset you want.
You can use your own resolutions, the ones presented are because of how the sdxl models are trained and those odd measurements are there to get the best results. But! you can put your own custom size in if you wish. go to advanced tab>developer debug mode> scroll down to the lines "Forced Overwrite of Generating Width" and "Forced Overwrite of Generating Height" and put any sizes you like.
@@MarcoEscobedo10 embeddings use a prompt, for example if you are trying to use a negative embedding like civitai.com/models/148131/ultimate-text-embeddings-sdxl-pack-styles-negative you would download that to embedding folder and then in the negative prompt type (embedding:ac_neg1:1.1)
Fooocus Update github thread: github.com/lllyasviel/Fooocus/discussions/2410
this is easy the best chanel for fooocus, thank you for your effort
When I watch your videos I feel like rather than giving us a fish, you are teaching how to catch one. For that I want to thank you for all Fooocus-oriented videos , they are very informative and helpfull for people like me who never tried local AI image generation. As a humble request, can you review some usefull LoRAs that you use? Like that zoom-slider LoRA that you featured in one of your videos?
Wow... such a kind and generous comment, how can i refuse? 😊
Thanks for the video!
This question is a little bit strange but here we go: do you think that SD FOOOCUS will add some EMOTE or LTX or SORA functions in the near future? A mean do you think that will can create a woman singing when she's moving or talking? I'm so excited for the infinite possibilities about AI mixing FOOOCUS with the rest of AI text to video coming soon, and i thought that maybe SD will include this functions in their packages. I know that im out of my mind😂. Thabk you for the video. Very well explained.
I don't think any animation type features will ever be added directly. One of the main features of Fooocus is it's light VRam requirements. Adding anything like that would bloat the requirements beyond majority of what the average user has.
But It's best to think of it as one step in the workflow. I create my images and inpaint as needed and use those images in my Animatediff/SVD projects.
I would like to ask if anybody else is experiencing "error" issues with the latest version of Fooocus. I've been unable to run this program and that is, the error message informs me that I don't have the latest Nvidea drivers, which I do. I have no idea how to resolve this problem and I'm hoping for some thoughts that may be helpful. Thank you.
Which ai tool you use making the voice over on the video? It's really cool
Elevenlabs. But it probably sounds good because I've been using my actual voice more now and spot checking with the AI synthesized version when I can't record and need corrections. Trying to move away from it entirely.
Damn I didn't know it was AI voice! 😍
I noticed there was nothing saved in yesterdays output. There was a history log but it did not get saved to outputs. I read it could've been due to quality performance over speed. Changed to speed performance and I'm noticing that while there is a history log, there doesn't seem to be anything saved to output. How do I ensure the generations are saved to the output folder thanks
hmm, is this the local version or the google colab online version? I think the online version doesn't save the outputs once the session ends. So you need to download them before you quit.
If this is on your computer I am not entirely sure. Did you change the output path at some point in the config.txt?
@@JumpIntoAI Thank you I found today and yesterday's output folder in the downloads folder for some reason (local)
how to update my version to this version?
The run.bat files have a command line in there to check for updates every time you start it. So should auto update.
how to save the preset.. i neet to change setting for lightning everytime
Check this out: ruclips.net/video/DRcxsqnhjws/видео.html
Can I use Civit AI models in Fooocus?
Thank you very much in advance.
If yes, how can I make them work?
Best.
Yes! you can use any SDXL model as a base model and any SDXL and sd 1.5 as a refiner. Just download them and put them into the fooocus\models\checkpoints folder.
Does anyone know how to make arrays of multiple words ? Like something : [[ (sun,water), (night, fire) ]] i want to alternate between sun water and night fire in this example.
Do like that [[sun,water]], [[night, fire]]
if you use coma as you said it will not count () and take 4 elements in array.
you can use wild cards now it has feature so it will read wild card in order
or for small prompt you can just use 'and' like [[ sun and water, night and fire]] so it will generate 2 images in total
Weird we still have a remove background option unlike in Stable Diffusion
After I go to Debug Mode, "Save Metadata to Images" is not ticked after I put the two lines to the config file and it doesn't work. Moreover, it messes up my other settings, like the number of images to generate or outputs folder... strange. I put the two lines (metadata - yes, format - fooocus) at the end of the config file, like u do in the vid... // In my opinion you can do vids about major updates of Fooocus, I liked this one very much. The new options would suprise me without this vid that I watched before using new ver of Fooocus.
are these changes in a preset json file or the config.txt file? If using a preset in the command window at the top do you see it saying something like "Load preset C:\Fooocus\presets\custom.json] failed" ?
@@JumpIntoAIThe problem was of course a comma before the added lines, so after closure of previous section by a square bracket... Sorry for trouble...
Is there any way to use reactor on foocus? I really like reactor on automatic1111 it works very well but I could find a way to work on foocus
Sadly no, just the built in face swap.
My fooocus always works on cpu instead of GPU, any idea how to fix that?
What operating system and GPU do you have? This does have some requirements which you can see here: github.com/lllyasviel/Fooocus?tab=readme-ov-file#minimal-requirement
AudioJungle music 😫, but hey, thank you so much for the video!
Motion Array! But I understand your pain. I apologize
@@JumpIntoAI Nah it's fine, it's on me. I spent waaaaaaaaay to much time on generic music platforms over the last few years for corporate videos soundtracks. I now have a kind of trauma when I hear it, they sound all the same 😅. That said, the content of your video was good enough for me to quickly ignore it, which is saying a lot!
I used foocus ai on public link not install in local but I’m always get error connection time out brother ?
If I install on local still have this matter or yet ?
My pc Core i7 12700kf
Ram 32
Gpu 1660S 6Gb
Can I run this app on local ?
What is public link? You mean colab notebook? It works fine, been using it for month
My fooocus is not updating, is there an option to force update?
hmm. It should auto update when started. Do you have it installed locally and are you starting it through one of the run.bat files?
will SD3 be included in Fooocus ?
There is no way to know before it's released. But we can hope.
this is style realistic in foocus can create 3D anime or cartoon or another style ?
Yes, this can create any of those with the right prompts or styles selected.
Do have a colab version, bro?
There is a Colab version on the developer page. colab.research.google.com/github/lllyasviel/Fooocus/blob/main/fooocus_colab.ipynb
Keep in mind most advanced features will cause memory to run out if using on a Free colab account.
@@JumpIntoAI thank you so much ❤️ but it still old version 1.12.2 🥲
@@dongytuetinh I think thats the git install version at the top, that colab is updated to the 2.2 version of fooocus.
checking myself, it says in the command lines
Fooocus version: 2.2.1
@@JumpIntoAI thank you so much ☺️
Foocus updates automatically?
Yes, it checks for updates every time you run it.
@@JumpIntoAI ok, thanks
WTF is with the simular framing used for EVERY generated inmage??!!
lol, well in this the point was to use the same seed using the new array support so it will look similar.
But in general short undescriptive prompts will end up like this. And sometimes the default fooocus styles lean into portraits. Better Prompts and pyracanny/CPDS poses is the way to go.
I have Focus 2.2.0 Using the GPU, but I want to use the CPU, So can you help me get it to work on CPU?
edit one of the run.bat files and add the line --always-cpu after Fooocus\entry_with_update.py
so it will look like this:
.\python_embeded\python.exe -s Fooocus\entry_with_update.py --always-cpu
@@JumpIntoAI I already did that, but in cmd it says system cannot find the path.......press any key to contiue.
But why?
@@eveekiviblog7361 On my 1050 Ti it is very slow, So I want to try it on my i5 13400F, the method --always-cpu not working for me. I tried several times, but it give me "the system cannot find the path....error.
@@hanynagy8969 sounds like a path is missing or in there incorrectly, one misplaced comma or a capitalized letter can mess things up, without knowing a lot more details it's hard to help. But if you copy your command window text and create a post on github.com/lllyasviel/Fooocus/issues
It would be easier to help.
Thank you for the efforts, it is very well explained. I have a question, I have a clothing store and I would like to create AI models who wear my creations, when I use a prompt image of my dresses, it always generates images with dresses a little different from those that I added to the prompt. I tried the negative prompt but to no avail.
please how to make it so that it keeps the same dress with its small details (drawing, embroidery...)? Thanks
The way stable diffusion works, It wont be able to transfer the exact details of an item into an image without some work. I have seen others create a LoRA based on an item, but that might not be efficient if you have multiple items.
Another way which is easier with certain products, maybe not clothes. Is taking a good image of the clothes on someone else, and then basically using the inpaint and masking everything else but the clothes themselves and generating a new person wearing said clothes. Obviously this is locked to the lighting and shape of the original image but can work.
Products that can be set on a table or simple wearables (like a purse) can be done easier this way.
Fooocus is a great tool, and your videos about it are excellent. I love how organized, informative and well thought out they are. I just want you to know that I really appreciate all the work you put into making these vids. You're my favorite AI-oriented channel by far.
Btw, do you know why Fooocus (since the update) has been stuck on "Loading models ..." for several minutes every time I hit generate? Before the update it took like 10 seconds to start the actual generation...
Assuming this is on your computer and not the colab? is it still working just taking longer? If so sounds like the Vram is being used up. Or does it stay stuck?
If it stays stuck it might be a file path error, might check the cmd window and see if it failed to load model.
Can you make a video on how to create good vector graphics in Fooocus, like brands, line art, coloring books and vector assets. I think there a great amount of potential in that area that most miss for the oversaturated photorealistic genre.
I agree! The want for creating "models" overshadows a lot of the other capabilities of SD. There are even some amazing tools coming out recently like Layer Diffusion for Forge. www.reddit.com/r/StableDiffusion/comments/1b1w6tt/transparent_image_layer_diffusion_using_latent/
Of course it is gated behind having to use Forge. But something to check out if you haven't.
I have a couple half finished projects and one is for logos. It's just lack of time that stops me.
New RUclipsr add on my list with Matt wolfe and theorically media. Thanks! 👍 I hope you'll add me in the near future. I created the first AI photorealistic singer influencer. We are on TikTok and IG right now, coming soon YT and Spotify. With her first album.
very good update
What about writing a prompt to choose between things at random, like a cat or dog?
How do you keep track of the max value of each lora?
Hello. Thank you for the review. Could you suggest the solution: I'm the t-shirt online store owner. I'd like to create images of different people in different poses with my t-shirts (with logo) on them. What ai tool can help me with that?
Sadly, Stable diffusion can't transfer specific items especially with logos into a scene without manipulation (photoshop) and or creating a LoRA for that specific item.
There are ways to get products in a scene with stable diffusion but wearables like shirts are the hardest and need the most pre and post processing.
Eventually I will get a vid out about this to explain better.
Is there a way to control the number of steps to produce the image other than the three options?
yes! It's a bit hard to find at first but go to advanced>Advanced tab>check developer debug mode>in debug tools tab scroll down till you see "Forced Overwrite of Sampling Step" and put in whatever steps you like. -1 will put it back to auto.
what is your desktop specification for runnning fooocus? On AMD, it almos does not work..
I have an NVIDA 4070 TI 12gb GPU.
AMD GPUs struggle a bit since stable diffusion was made for nvidia architecture (cuda) but there are some workarounds
You can check out the developer discussion page to see some methods like here: github.com/lllyasviel/Fooocus/discussions/488
in your app the background is dark, but in mine it is white and I can't see anything :@
you can add /?__theme=dark to the end of the address bar. like
127.0.0.1:7865/?__theme=dark
If you are running it locally, you can add --theme dark to the run.bat like this to always load it that way:
.\python_embeded\python.exe -s Fooocus\entry_with_update.py --preset realistic --theme dark
I am a beginner. Could you tell me where I can find my generated images?
in the fooocus\outputs folder, in there it will be folders separated by date
Excelente videos. I love how you master Fooocus. Impressive. Just a note, this character " is called double quotes. This () are parenthesis 😉.
They really, really, really have to add an erasure brush to the inpaint outpaint! It makes no sense having to step back 2,3,4,5, times or delete the whole mask, when you could just switch to the erasure bush.....and erase. Save a whole lota time and aggravation IMO.
Right there with you on that, especially painful when you do everything in one long mouse click and mess up at the end.
That's exactly what I'm talking about buddy. It also would be nice if they let you manually change the dimensions in the outpaint (top, bottom, right, left} instead of having to use the default dimensions. I've been trying to send Illyasviel a suggestion or request on these matters. But I'm new to this Github stuff. Do you know how to send a message in the github web site? I'm registered,but how the heck do you send a message (discussions, pull request, issues etc. etc.)?????
Thanks for any feedback my friend.
@@neonxp99 I would go to the discussion page on github: github.com/lllyasviel/Fooocus/discussions
and post there with "new discussion" . Mashb1t is an incredibly active mod/collaborator there that could answer those questions.
Mission Accomplished : - )!! Hopefully they will address these issues soon.
Thanks for your help my friend👍✌
@JumpIntoAI , as i only have juggernautXL_v8Rundiffusion model ,how i install other Base Model
You can use any SDXL 1.0 checkpoint, find one on CivitAI and download it to your Fooocus\models\checkpoints folder
There is a issue in Focus, I can't extract the zip with the .7, after 35% it get stuck, note - I have already installed automatic1111 on my system ??? Can they work together
try redownloading the zip file perhaps? Fooocus is its own entity but it can share the models you have already downloaded with automatic 1111.
Well... how to update Fooocus without redownloading the archive from Github? Are there any args to bat-file?
No need to do anything. The bat file should have the update arg by default "entry_with_update.py" So it checks for updates every time you start it.
Where is the link to Presets tutorial? When I changed presets json files it defaults back to original setting when running Fooocus.
This shows how to setup a turbo preset: ruclips.net/video/cCS_pBdh7ko/видео.html
This shows how to do a lightning preset: ruclips.net/video/DRcxsqnhjws/видео.html
But can be used to make any preset you want.
Until now I am waiting for the option to be able to change the image sizes in a personalized way and not with the measurements suggested in Fooocus
You can use your own resolutions, the ones presented are because of how the sdxl models are trained and those odd measurements are there to get the best results. But! you can put your own custom size in if you wish.
go to advanced tab>developer debug mode> scroll down to the lines "Forced Overwrite of Generating Width" and "Forced Overwrite of Generating Height" and put any sizes you like.
@@JumpIntoAI Thank you so much! this has helped me. Do you know where I find the buttons in the program to use embeddings?
@@MarcoEscobedo10 embeddings use a prompt, for example if you are trying to use a negative embedding like civitai.com/models/148131/ultimate-text-embeddings-sdxl-pack-styles-negative
you would download that to embedding folder and then in the negative prompt type (embedding:ac_neg1:1.1)