Outstanding overview, as always! I can tell you had to put a lot of research into this, and it's appreciated. You're the only content creator out there who seems to take Fooocus seriously.
Thanks. I'm still experimenting and learning myself. It is definitely a different beast. But some of the newer ones are pretty decent and easier to use.
@@KLEEBZTECH totally. The satpony_BakedvaeREALISM model is pretty amazing for realism. Hands are so good but still seem to get the one face no matter what you prompt in lol
I have noticed that about the realistic models and the faces. But I have not played around enough with them yet. Have you tried using a none pony realistic model as a refiner? I have found that can be useful sometimes when set to only handle the last few steps.
Awesome video. Thanks. I’ve attempted to make realistic character loras with pony but haven’t had much success. Would love to see a video on creating loras for Pony or perhaps some tips
Love that 2.5.x added pony, I find that the results tend to be closer to the prompt, although most "realistic" pony models tend to still always have this slight animation appearance to them
Yes and also not much variety in the faces. But I am starting to see improvements overall. I've been doing some testing for an upcoming video on realistic Pony models
I really appreciate your channel. I am subscribed which is a rare thing for me. One request: do you have any advice on using Loras which are not specific to the model - such as an SDXL Lora with a Pony model or vice versa? I notice that CivitAI often has versions of the same Lora for XL, 1.5, Flux and Pony so I wonder whether there are any cross-model possibilities?
Excellent! I was having a ton of problems running pony models and now I know why. Great overview to get me started. I don't know why there aren't more resources for using pony models with focus, but I really appreciate your work. I look forward to your video on using realistic pony models.
Video is to show how to get pony to work with Fooocus specifically. I am sure there are a ton of videos covering it for Comfy or A1111 and I primarily use Fooocus as my default SD tool.
i've been sick for the last week, so probably coming this week(end). For the Enhance feature many lines of code had to be rewritten, which will take a bit of effort to merge all things correctly and test afterwards.
I may have missed something in past videos, but how to use checkpoint merge in fooocus? The problem is that when I try to use AutismMix SDXL, it refuses to generate
Which exact checkpoint are you using? If you have the exact filename or which one from here: civitai.com/models/288584?modelVersionId=324619 And are you getting any messages in the console/CMD window when you do?
@@KLEEBZTECH "File corrupted: /content/Fooocus/models/checkpoints/autismmixSDXL_autismmixPony.safetensors Fooocus has tried to move the corrupted file to /content/Fooocus/models/checkpoints/autismmixSDXL_autismmixPony.safetensors.corrupted' displays this error, did I download the model on the wrong path or what?
@@KLEEBZTECHFile corrupted: /content/Fooocus/models/checkpoints/autismmixSDXL_autismmixPony.safetensors Fooocus has tried to move the corrupted file to /content/Fooocus/models/checkpoints/autismmixSDXL_autismmixPony.safetensors.corrupted
@@KLEEBZTECH File сorrupted: /сontent/Fооосus/mоdels/сheckpоints/аutismmixSDXL_аutismmixPony.safеtеnsоrs Fооосus has tried to move the cоrruptеd file to /сontent/Fооосus/mоdels/сheсkpоints/аutismmixSDXL_аutismmixPоny.sаfetensоrs.соrruptеd
I would delete the file and try downloading again. It say the file is corrupted so it is trying to rename the file and it is not actually moving it. Just adding the file extension to signify it is corrupt it seems. Because a "checkpoint merge" is no different than a checkpoint. It just refers to how it was trained. I tested the AutismMix_pony model and no issues.
Thanks for the video :) I tried to use a specific model for PonyXL inpainting (ponydiffusionV6XL_v10-inpainting.safetensors) but I always get an error : AttributeError: 'NoneType' object has no attribute 'shape'
@@KLEEBZTECH At least, it works without that model by selecting "None" as Engine :) And the result is better than keeping the latest version of the engine :)
Hey just wondering, which gpu are you using? Looks like you are producing images about twice as fast as me. I have the 4060ti 16gb RAM. I thought it was fairly powerful, but your images seem to generate twice as fast.
Thank you for the video! As always, you bring up and thoroughly explain interesting topics. Your channel has been a real discovery for me. I've heard that PonyXL is great for fixing hands and feet in anime-style art using the inpaint tool. I've tried doing it myself, but it didn't turn out very well. It would be fantastic if you made a video about this, as there are still many challenges despite the numerous methods available to make hands and feet look normal.
Inpainting and Pony is still one of the things I am experimenting with a lot and trying to find the best way since the Fooocus inpainting engine does not work well with it. And Pony I feel does much better with hands. At least with some of the realistic ones I have tested it seems to be the case.
The new enhance feature in Mashb1ts fork (and probably coming to regular fooocus in near future) might help with that but I think it still has issues dealing with eyes right now.
I use a workaround like u mentioned for pony, Inpainting with default v.2 inpaint, after it produces the image with a distorted face, i inpaint it again, with improve detail - Highly detailed face.. it works better and matches the image with no quality loss
@@arjhunk1520 There are ways. You would have to get and install one of the older versions. But there should be no reason this should not work any more. If using the pony preset built in it that also disables the Fooocus inpaint engine. Not sure if that is what you are running into. Might want to go onto Github and ask since there are some new useful features in this version like the auto masking.
@@arjhunk1520 Will be very soon. I have done all the testing I needed to do and will give everything I have figured out. But will admit I still don't know everything about it yet.
If you need a good clean anime style in a standard pony checkpoint, then I can share my settings: Guidance Scale 7.0 Image sharpness 0 Sampler - ddpm Loras: Anime Lineart (pony diffusion) : 0,5 MeMaXL Flat Anime Style (AutismMix) : 0,5 Promt: score_9, score_8_up, score_7_up, score_6_up, source_anime, high resolution, anime screencap, masterpiece, (detailed face:1.1), pretty face, 1girl, [what you want].
Outstanding overview, as always! I can tell you had to put a lot of research into this, and it's appreciated. You're the only content creator out there who seems to take Fooocus seriously.
Thanks! I do try to do more on Fooocus since it is not covered much. And I do feel it is better than many give it credit for.
Finally a good overview of PONY models. I recently started using them in Forge and its great to see how the code tricks actually work
Thanks. I'm still experimenting and learning myself. It is definitely a different beast. But some of the newer ones are pretty decent and easier to use.
@@KLEEBZTECH totally. The satpony_BakedvaeREALISM model is pretty amazing for realism. Hands are so good but still seem to get the one face no matter what you prompt in lol
I have noticed that about the realistic models and the faces. But I have not played around enough with them yet. Have you tried using a none pony realistic model as a refiner? I have found that can be useful sometimes when set to only handle the last few steps.
@@KLEEBZTECH Ill give it a shot cheers
Awesome video. Thanks. I’ve attempted to make realistic character loras with pony but haven’t had much success. Would love to see a video on creating loras for Pony or perhaps some tips
Love that 2.5.x added pony, I find that the results tend to be closer to the prompt, although most "realistic" pony models tend to still always have this slight animation appearance to them
Yes and also not much variety in the faces. But I am starting to see improvements overall. I've been doing some testing for an upcoming video on realistic Pony models
does this model doesnt work with the pyra canny of fooocus?
Great vid, generation time is much faster, what spec you use man?
I really appreciate your channel. I am subscribed which is a rare thing for me. One request: do you have any advice on using Loras which are not specific to the model - such as an SDXL Lora with a Pony model or vice versa? I notice that CivitAI often has versions of the same Lora for XL, 1.5, Flux and Pony so I wonder whether there are any cross-model possibilities?
Yeah that really won't work.
Great video and lots of great info on pony models. Now I feel comfortable trying them since you explained them in the video. Thank you!
Realistic models coming soon.
Excellent! I was having a ton of problems running pony models and now I know why. Great overview to get me started. I don't know why there aren't more resources for using pony models with focus, but I really appreciate your work. I look forward to your video on using realistic pony models.
Why not Comfy or the OG Automatic1111?
Video is to show how to get pony to work with Fooocus specifically. I am sure there are a ton of videos covering it for Comfy or A1111 and I primarily use Fooocus as my default SD tool.
Of course many of the things discussed apply to any SD UI.
I personally prefer Fooocus UI, I was using Comfy in the past :)
We still don't have ADETAILER in Fooocus?
we have it , try mashb1t fork
Similar feature coming soon to regular Fooocus and already being tested in Mashb1t fork.
i've been sick for the last week, so probably coming this week(end). For the Enhance feature many lines of code had to be rewritten, which will take a bit of effort to merge all things correctly and test afterwards.
You are doing great work! I hope you feel better soon! I got to get back to messing with Enhance.
Just use the "enhance" bar, basicly the same
I may have missed something in past videos, but how to use checkpoint merge in fooocus? The problem is that when I try to use AutismMix SDXL, it refuses to generate
Which exact checkpoint are you using? If you have the exact filename or which one from here: civitai.com/models/288584?modelVersionId=324619 And are you getting any messages in the console/CMD window when you do?
@@KLEEBZTECH "File corrupted: /content/Fooocus/models/checkpoints/autismmixSDXL_autismmixPony.safetensors
Fooocus has tried to move the corrupted file to /content/Fooocus/models/checkpoints/autismmixSDXL_autismmixPony.safetensors.corrupted'
displays this error, did I download the model on the wrong path or what?
@@KLEEBZTECHFile corrupted: /content/Fooocus/models/checkpoints/autismmixSDXL_autismmixPony.safetensors
Fooocus has tried to move the corrupted file to /content/Fooocus/models/checkpoints/autismmixSDXL_autismmixPony.safetensors.corrupted
@@KLEEBZTECH File сorrupted: /сontent/Fооосus/mоdels/сheckpоints/аutismmixSDXL_аutismmixPony.safеtеnsоrs
Fооосus has tried to move the cоrruptеd file to /сontent/Fооосus/mоdels/сheсkpоints/аutismmixSDXL_аutismmixPоny.sаfetensоrs.соrruptеd
I would delete the file and try downloading again. It say the file is corrupted so it is trying to rename the file and it is not actually moving it. Just adding the file extension to signify it is corrupt it seems. Because a "checkpoint merge" is no different than a checkpoint. It just refers to how it was trained. I tested the AutismMix_pony model and no issues.
Thanks for the video :) I tried to use a specific model for PonyXL inpainting (ponydiffusionV6XL_v10-inpainting.safetensors) but I always get an error : AttributeError: 'NoneType' object has no attribute 'shape'
I tried as well. Not going to work.
@@KLEEBZTECH At least, it works without that model by selecting "None" as Engine :) And the result is better than keeping the latest version of the engine :)
Nice video with so many detail!
Besides, just wondering, is it possible to run Fooocus with PonyXL on a 4GB VRAM GPU device (ex: GTX970)?
Fooocus will run on those according to the requirements but will be very slow.
@@KLEEBZTECH Thanks, I'll still give it a try to see if I get the luck
If you have any issues check here: github.com/lllyasviel/Fooocus/blob/5a71495822a11bbabf7c889eed6d9b38b261bb96/troubleshoot.md#L4
Models size are around 6.6GB so you won't be able to put the whole model in VRAM :( I think there are some "lite" models that are less huge :)
@@kervala OH NO....
can you please do a video on running Pony realistic models?
Coming soon. Already working on putting things together.
Hey just wondering, which gpu are you using? Looks like you are producing images about twice as fast as me. I have the 4060ti 16gb RAM. I thought it was fairly powerful, but your images seem to generate twice as fast.
This video I am using a 4070 with 12GB VRAM. Images in the video were all 30 steps.
Thank you for the video! As always, you bring up and thoroughly explain interesting topics. Your channel has been a real discovery for me.
I've heard that PonyXL is great for fixing hands and feet in anime-style art using the inpaint tool.
I've tried doing it myself, but it didn't turn out very well. It would be fantastic if you made a video about this, as there are still many challenges despite the numerous methods available to make hands and feet look normal.
Inpainting and Pony is still one of the things I am experimenting with a lot and trying to find the best way since the Fooocus inpainting engine does not work well with it. And Pony I feel does much better with hands. At least with some of the realistic ones I have tested it seems to be the case.
Bro can you suggest any idea or LoRA's to get perfect eyes in juggernaut checkpoint
Have you used the improve detail feature for inpainting?
Yes but it doesn't provide what I really need and also it consumes time when I work on many images
The new enhance feature in Mashb1ts fork (and probably coming to regular fooocus in near future) might help with that but I think it still has issues dealing with eyes right now.
Ohk thanks
@@KLEEBZTECH
Thank you 👍👍👍
Welcome 👍
I use a workaround like u mentioned for pony, Inpainting with default v.2 inpaint, after it produces the image with a distorted face, i inpaint it again, with improve detail - Highly detailed face.. it works better and matches the image with no quality loss
Good to know! Thanks!
@@KLEEBZTECHThis method stopped working after the update, is there a way to use the previous fooocus version without updating?
@@arjhunk1520 There are ways. You would have to get and install one of the older versions. But there should be no reason this should not work any more. If using the pony preset built in it that also disables the Fooocus inpaint engine. Not sure if that is what you are running into. Might want to go onto Github and ask since there are some new useful features in this version like the auto masking.
@@KLEEBZTECHThank you, I just restarted fooocus and it worked like it used to 😅😅😅 please upload an in depth video on Enhance feauture. Thank you
@@arjhunk1520 Will be very soon. I have done all the testing I needed to do and will give everything I have figured out. But will admit I still don't know everything about it yet.
thanks a lot 💚💚
You're welcome 😊
If you need a good clean anime style in a standard pony checkpoint, then I can share my settings:
Guidance Scale 7.0
Image sharpness 0
Sampler - ddpm
Loras:
Anime Lineart (pony diffusion) : 0,5
MeMaXL Flat Anime Style (AutismMix) : 0,5
Promt:
score_9, score_8_up, score_7_up, score_6_up, source_anime, high resolution, anime screencap, masterpiece, (detailed face:1.1), pretty face, 1girl, [what you want].
WTH is a pny model? It' can't just mean My Little Pony, can it? Must be more to it than that.
It was created at first to do furry and cartoon images mostly. Then it grew from that.
Lucky. I have 64GB RAM, 8GB VRAM and 30GB Unified Shared VRAM, and Fooocus crashes by simply trying to load the model. The power of Python =)