So pleased I came back to this tutorial thank you for your work here. I look forward to both the 1.5 version of Photomaker and the "official" ComfyUI version of InstantID. For those that read this the nodes are changing quickly so everything won't be exactly as stated.
Like alaways you explain things from start, with detail and very clean, Thank you. I found IPAdapter miles ahead, specially the new FACEID V2 SDXL, it's heavy and sometimes it switch my comfyui to low v-ram mode (make it slow as hell and i have to restart my pc in order to get normal vram) but it give way more accurate results.
@@CodeCraftersCorner So ture, even FaceID SD 1.5 and v2 are way way better than this, we can get more accurate results and have way more control over the face and expression. but either way it is good to see all these new advancements in just a short time and for free, i am really thankful for all of them and hope they create even better tools in the future. Thanks man
Thanks for the video, I prefer this official implementation much more the ZHO-ZHO-ZHO one, since that one download other nodes and even checkpoints without asking. To be honest not so user friendly. BTW, there is a new model called InstantID just released, similar and seems stronger function, worth looking. AI researchers are just crazy, I remember one week ago it was FaceID v2 plus, now Photomaker, and InstantID might be the next, just in less than 10 days.
Thank you @Gabriecielo! Yes, I, too prefer the official implementation. For instantID, i just released a video on the channel. So far, no official implementation but we got a good "early access" by ZHO-ZHO-ZHO.
@Digital_Toolbox, Thanks! Yes, it is quite hard to get the generated images to be like the original with PhotoMaker. I suggest experimenting and once you get something good, save the settings for later use. Different checkpoints will have different optimal settings.
Hello @RonnieMirands, the project got a lots of updates lately as mentioned by @uk3dcom. It is possible that now it can take SD1.5 models and workflows no longer work. The developer mentioned to "Due to various node updates and changes, you may have to recreate the node." in their guide. Best to experiment and see what works for you.
Thank you for your work. I tried both Photomakers from (1) Zho-zho-zho and (2) Shiimizu. The first one was very slow, but produced great results even from a single photo. The second one never produced a decent result. I don't know what the problem could be.
Thank you, @VitaliyChaback! The Zho-Zho-Zho version is complete and can be used on its own. The Shiimizu version will require additional helper nodes to get decent results. You can try to combine it with ControlNet and using a second KSampler node to add details. You can try to look into the workflows I provided here: bit.ly/3Umkq9d. The post is from 6 days ago. Since the project is getting updates frequently, some node may have different names. Good Luck!
Hello, thanks for the explanations. Is it possible to use control net elements with this photomaker. For example, when we make an oil painting effect, is it possible to apply only the effect to the picture and keep the picture faithful to the original?
Hello @hudaicihandurmus5414! Yes you can use ControlNets. You have to download the SDXL Control Lora models and use these as the ControlNet v1.1 for SD1.5 will not work. I have updated the Patreon post (free) with an image and workflow file. The link is in the description. Thank you for watching.
In my Photomaker Node, there are only 2 menu items - the LoadPhotoMaker option is not available. Reinstalled twice (the SMZ version). Photomaker Encode and Apply Photomaker Style are OK, as is Prepare Images For ClipVision From Path.
Hello @JackTorcello, I just tested and installed the latest version successfully. All the nodes are showing. I see you have logged an issue in the Github repo. Hope the developers can assist you with that.
Thanks a lot for the video, very detailed! Did I understand correctly that you need to create your own folder for each person? maybe it would be better to create a temporary folder, and then delete it after completion.
Thank you for the feedback, @krupnikova! Yes, when loading images from a folder, you may have to create multiple folders for each person. If you do not need the input images after the image generation, using a temporary folder is a good idea for organization and cleanup.
Thank you for the support, @mehradbayat9665! There are many reasons to use multiple ksamplers. One example is, you can start generating an image and stop the generation short, (around 4-6 steps). The idea is to generate an image with a correct pose and frame. Then send the output to another ksampler with possible a different model or prompts to get the style. You can also do noise injection with this technique.
Thank you for the suggestion, @OliviaFromTrivia! I'm relatively new to creating UA-cam educational content, and I truly appreciate your input. I will certainly incorporate that to improve the viewing experience.
i tried it on my own photos, with several different models. the result is absolutely doesn't look like me :) I guess it works only on celebrities because their data is already in the models.
Hello @deviantstudio! Yes, if the model is already trained on the subject, then it will give better results. You can try using multiple images as input which can enhance the resemblance.
Very useful video! I am a little confused about what to do with the ZZZ version...uninstall? If so, how? I see when you are loading nodes, the ZZZ Photomaker version is no longer there, correct?
Thank you, @WhySoBroke! If you have successfully installed the official version by Shiimizu and it is running well, yes, you can uninstall the ZZZ version by deleting the folder from your custom nodes folder. Then assuming you are on Windows, go into your home directory. It should be something like C:\Users\your_username. In here, there is a folder named .cache. Go in and delete the model. This will help save some disk space.
Thank you @@CodeCraftersCorner I am having another issue. When I load your workflow, the following Node is not found: PhotomakerEncode. Only PhotomakerEncodePlus is present but that one is missing the clipvision input. Any idea how to fix that?
@WhySoBroke, The developer updated the node. Try the following: 1. go into the PhotoMaker folder inside the custom node folder. 2. open CMD here (type CMD in address bar and press enter) 3. type in: git pull. 4. (optional) if you get a "error: Your local changes to the following files would be overwritten by merge:", type in: git restore . ( the dot is important) 5. Then do: git pull 6. Now start ComfyUI 7. Replace the old nodes with the new nodes and connect them. 8. The PhotoMaker Encode Plus will take the clip from the model, the photomaker output from the Load PhotoMaker node, an image as input, the positive prompt. 9. The node will output a CONDITIONING which you can connect to the positive prompt. 10. You will not require a clipvision input for this one.
Hi @MrJohnerlandsson83. I do not have a Linux system to test. I am guessing you will have to install it from the command line. I saw an article suggesting sudo apt install nvidia-settings for Ubuntu 20.04. Be sure to double check the command and research it before running on your build. You may have a different app manager.
When i queue prompt i got very very long error starts like: Error occurred when executing PhotoMakerLoader: Error(s) in loading state_dict for PhotoMakerIDEncoder:
@goliat2606, You will have to download a new SDXL model. The ones you tried are all SD1.5 models. PhotoMaker will only work with SDXL models. Go to civitai website and use the filter to search for an SDXL model and try again.
@@CodeCraftersCorner now i tester on fresh portable ComfyUI and it works:). What can be wrong on my old installation:/? Any way to update packages or something?
Thank you @drmuradkhan, and well wishes! I currently have a private discord server for my Patreon supporters. I do not have the capacity to make it public just yet.
So pleased I came back to this tutorial thank you for your work here. I look forward to both the 1.5 version of Photomaker and the "official" ComfyUI version of InstantID. For those that read this the nodes are changing quickly so everything won't be exactly as stated.
Glad to hear, @uk3dcom!
Glad I found your channel. Excellent, comprehensive explanation. Thank you, and I'm looking forward to more of your tutorials.
Thank you for the positive feedback, @brgtubedev001!
Thank you for the kind explanation today. Thanks to you, I installed it successfully, and it's working well. You are truly intelligent.
Thank you very much, @KKsinsa! I truly appreciate your kind words.
Excellent video my friend. Leaving this comment to please the algorithim gods in your favor.
Thank you so much, @Douchebagus!
Nicely explained. Good work.
Thank you, @aamir3d!
Like alaways you explain things from start, with detail and very clean, Thank you.
I found IPAdapter miles ahead, specially the new FACEID V2 SDXL, it's heavy and sometimes it switch my comfyui to low v-ram mode (make it slow as hell and i have to restart my pc in order to get normal vram) but it give way more accurate results.
Thank you for the positive feedback, @0A01amir! I also prefer FACEID since I have more control over the final output.
@@CodeCraftersCorner So ture, even FaceID SD 1.5 and v2 are way way better than this, we can get more accurate results and have way more control over the face and expression. but either way it is good to see all these new advancements in just a short time and for free, i am really thankful for all of them and hope they create even better tools in the future.
Thanks man
@0A01amir, yes, very true!
Excellent tutorial ! Complete ! Thanks for the share . :O)
Glad you liked it!
I think the node broke again. I logged an issue on the repo. It was working fine and then there was an update which caused the load node to break.
Thank you for reporting, @michaelbayes802! I saw that the repo got updated again after your comment. Issue #17 was closed and resolved.
Brilliant descriptive video,thanks. Just one thing, i don't see a workflow example in the description?
Thank you for pointing that out, @SmallFox74! I have updated the video description. Here is the link: bit.ly/3Umkq9d
You explain very well!! thank you very much for sharing
Thanks for watching, @nicolaslozanovalencia! Appreciate the feedback.
Thanks for the video, I prefer this official implementation much more the ZHO-ZHO-ZHO one, since that one download other nodes and even checkpoints without asking. To be honest not so user friendly. BTW, there is a new model called InstantID just released, similar and seems stronger function, worth looking. AI researchers are just crazy, I remember one week ago it was FaceID v2 plus, now Photomaker, and InstantID might be the next, just in less than 10 days.
Thank you @Gabriecielo! Yes, I, too prefer the official implementation. For instantID, i just released a video on the channel. So far, no official implementation but we got a good "early access" by ZHO-ZHO-ZHO.
Great tutorial! I managed to make Photomaker work but for some reason I am getting images that doesn't match the original reference at all.
@Digital_Toolbox, Thanks! Yes, it is quite hard to get the generated images to be like the original with PhotoMaker. I suggest experimenting and once you get something good, save the settings for later use. Different checkpoints will have different optimal settings.
excellent work amigo, keep up the great work and keep going! subbed! :)
Thanks for the sub, @2PeteShakur! Glad you liked it.
why your comfyui, has a beautiful interface and more options? is there anything i am missing? really great!
Hello @RonnieMirands, yes, I have some custom nodes installed and I changed my ComfyUI settings. I may make a video on the topic in the future.
Many thanks, excellent video
Thank you very much, @johnlenoob6951!
weird, you said is not compatible with SD1.5 models, but when i load any SD1.5 model it works as well!
but loading the basic workflow it doesnt work at all, the image output has nothing to do with the input, its like broken
Hello @RonnieMirands, the project got a lots of updates lately as mentioned by @uk3dcom. It is possible that now it can take SD1.5 models and workflows no longer work. The developer mentioned to "Due to various node updates and changes, you may have to recreate the node." in their guide. Best to experiment and see what works for you.
Thank you for your work. I tried both Photomakers from (1) Zho-zho-zho and (2) Shiimizu. The first one was very slow, but produced great results even from a single photo. The second one never produced a decent result. I don't know what the problem could be.
Thank you, @VitaliyChaback! The Zho-Zho-Zho version is complete and can be used on its own. The Shiimizu version will require additional helper nodes to get decent results. You can try to combine it with ControlNet and using a second KSampler node to add details. You can try to look into the workflows I provided here: bit.ly/3Umkq9d. The post is from 6 days ago. Since the project is getting updates frequently, some node may have different names. Good Luck!
Hello, thanks for the explanations. Is it possible to use control net elements with this photomaker. For example, when we make an oil painting effect, is it possible to apply only the effect to the picture and keep the picture faithful to the original?
Hello @hudaicihandurmus5414! Yes you can use ControlNets. You have to download the SDXL Control Lora models and use these as the ControlNet v1.1 for SD1.5 will not work. I have updated the Patreon post (free) with an image and workflow file. The link is in the description. Thank you for watching.
Really good video, thanks !
Thank you @johnriperti3127! Glad you liked it.
In my Photomaker Node, there are only 2 menu items - the LoadPhotoMaker option is not available. Reinstalled twice (the SMZ version). Photomaker Encode and Apply Photomaker Style are OK, as is Prepare Images For ClipVision From Path.
Hello @JackTorcello, I just tested and installed the latest version successfully. All the nodes are showing. I see you have logged an issue in the Github repo. Hope the developers can assist you with that.
@@CodeCraftersCorner Thank you. You are an amazingly thoroughgoing and patient teacher. I can see that I will learn a lot from you!
@JackTorcello Thank you so much! Glad to have you as part of the community.
Same problem for me. Under Add Node: _for_testing:photomaker:PhotomakerLoader can be found! @@CodeCraftersCorner
@adelrosso, I'll create a new video with updated steps. The project changed since the video.
Tutorial is incomplete. Does not show complete workflow. No test runs.
Thanks a lot for the video, very detailed! Did I understand correctly that you need to create your own folder for each person? maybe it would be better to create a temporary folder, and then delete it after completion.
Thank you for the feedback, @krupnikova! Yes, when loading images from a folder, you may have to create multiple folders for each person. If you do not need the input images after the image generation, using a temporary folder is a good idea for organization and cleanup.
Love it. Liked. Only 1 question this time :-) - What's the point of going from one ksampler into another ksampler?
Thank you for the support, @mehradbayat9665! There are many reasons to use multiple ksamplers. One example is, you can start generating an image and stop the generation short, (around 4-6 steps). The idea is to generate an image with a correct pose and frame. Then send the output to another ksampler with possible a different model or prompts to get the style. You can also do noise injection with this technique.
Interesting, and what does noise injection do? And would you ever consider creating a video to go more in depth?@@CodeCraftersCorner
And why wouldn't you just do the pose and style in the first run? Do you get more consistency with two ksamplers?@@CodeCraftersCorner
@mehradbayat9665, It is just an option giving more flexibility. Two ksamplers does not guarantee a better result but does give more control.
you should add the "hook" to your videos, basically frontloading the best what to expect from the video, instead I have to search for it ....
Thank you for the suggestion, @OliviaFromTrivia! I'm relatively new to creating UA-cam educational content, and I truly appreciate your input. I will certainly incorporate that to improve the viewing experience.
i tried it on my own photos, with several different models. the result is absolutely doesn't look like me :)
I guess it works only on celebrities because their data is already in the models.
Hello @deviantstudio! Yes, if the model is already trained on the subject, then it will give better results. You can try using multiple images as input which can enhance the resemblance.
They have changed the name of the nodes to photomaker plus so it doesn't create issue with the zho zho version
Thanks for the info, @musicandhappinessbyjo795! Appreciate it!
It was not work whatever follow your step or download your workflow. Can you show how it's work in the video?
Sure, @AICreator1314! I'll create a new video with updated steps. Stay tuned!
How to use LORA in this workflow?
Hello, @maykehafemann9531, I am recording a new video for a follow on PhotoMaker.
Very useful video! I am a little confused about what to do with the ZZZ version...uninstall? If so, how? I see when you are loading nodes, the ZZZ Photomaker version is no longer there, correct?
Thank you, @WhySoBroke! If you have successfully installed the official version by Shiimizu and it is running well, yes, you can uninstall the ZZZ version by deleting the folder from your custom nodes folder. Then assuming you are on Windows, go into your home directory. It should be something like C:\Users\your_username. In here, there is a folder named .cache. Go in and delete the model. This will help save some disk space.
Thank you @@CodeCraftersCorner I am having another issue. When I load your workflow, the following Node is not found: PhotomakerEncode. Only PhotomakerEncodePlus is present but that one is missing the clipvision input. Any idea how to fix that?
@WhySoBroke, The developer updated the node. Try the following:
1. go into the PhotoMaker folder inside the custom node folder.
2. open CMD here (type CMD in address bar and press enter)
3. type in: git pull.
4. (optional) if you get a "error: Your local changes to the following files would be overwritten by merge:", type in: git restore . ( the dot is important)
5. Then do: git pull
6. Now start ComfyUI
7. Replace the old nodes with the new nodes and connect them.
8. The PhotoMaker Encode Plus will take the clip from the model, the photomaker output from the Load PhotoMaker node, an image as input, the positive prompt.
9. The node will output a CONDITIONING which you can connect to the positive prompt.
10. You will not require a clipvision input for this one.
@@CodeCraftersCorner nice one mate, worked a treat! ;)
👍
Thanks. The project has not been updated since January 2024.. I guess it must have been deprecated.
Yes, there has not been any new updates.
How do I set sysmem fallback policy in linux?
Hi @MrJohnerlandsson83. I do not have a Linux system to test. I am guessing you will have to install it from the command line. I saw an article suggesting sudo apt install nvidia-settings for Ubuntu 20.04. Be sure to double check the command and research it before running on your build. You may have a different app manager.
When i queue prompt i got very very long error starts like:
Error occurred when executing PhotoMakerLoader:
Error(s) in loading state_dict for PhotoMakerIDEncoder:
Hi @goliat2606! Can you try with a different model / checkpoint? In PyTorch, the state_dict is an object which holds the model parameter.
@@CodeCraftersCorner I tried v1-5-pruned-emaonly, realisticVisionV60B1_v60B1VAE and dreamshaper_8. Always the same :/.
@goliat2606, You will have to download a new SDXL model. The ones you tried are all SD1.5 models. PhotoMaker will only work with SDXL models. Go to civitai website and use the filter to search for an SDXL model and try again.
@@CodeCraftersCorner I tried on juggernautXL_v8Rundiffusion and realismEngineSDXL_v30VAE and the same ;/.
@@CodeCraftersCorner now i tester on fresh portable ComfyUI and it works:). What can be wrong on my old installation:/? Any way to update packages or something?
could you upload your pathflow, so that I can just download it and use it, rather than following this video. I cant seem to figure it out. Thanks
Sure, here is the workflow with some more examples: bit.ly/3HCXFX6
Thank you for the video and i wish you all the best. do you have any discord server ?
Thank you @drmuradkhan, and well wishes! I currently have a private discord server for my Patreon supporters. I do not have the capacity to make it public just yet.