***UPDATE*** Make sure you check out my GitHub for a new Python script that I created that gives you a nice interface for splitting these character sheets up. you can find it here: github.com/Xerophayze/splitter
First of all, thank you for this tutorial, this is something I was looking for a while. Have you tried generating a small rotation around you character, full body or portrait? If there is enough consistence, you can probably create a character that can be displayed in a 3D display. I want to try this technique with Instant ID to see if it is possible to generate a 3D portrait from a 2D photo.
@@AIchemywithXerophayze-jt1gg I thought you'll love it since you're into using interfaces like A1111. Stable swarm ui is comfy ui but with an A1111 like interface. You can also use comfy ui nodes in it
I just used my prompt generator to generate this prompt and it works great with Juggernaut. Wide view digital fantasy illustration, (Enormous Minotaur standing amidst ruins:1.2), Rule of thirds composition, Strong horns, Massive axe, (Ancient temple backdrop:1.3), Crumbling columns, Overgrown vegetation, (Eerie atmosphere:1.2), Moody lighting, Weathered textures, Dramatic shadows
I keep trying to apply the poses from an openpose sheet, but SD keeps acting like it's not even picking up controlnet even with the 'controlnet is more important' selected. Any ideas?
Yeah usually if it's doing that, the controlnet version or the model are not compatible. So whatever model or checkpoint you're using, you need to change it to something else that will be compatible with the controlnetmodel you're using
You would want to render the image initially as a PNG, then you can use a background removal tool to remove the background and convert it to a transparency, which means that it's a PNG with an alpha channel. That way you can import the image into another image and overlay it on top of another image.
Can these character full body and face sheets just be created unclothed in order to train a character embedding or lora so then the clothing and poses can be added later with openpose and prompting?
I would start by going to and installing stability matrix, and then installing the latest version of forge. I believe you can grab my templates from my Google share.
How do you get characters to have good faces in these model sheets? Whenever my caracters are created far away for model sheets, the faces are always warped.
@@AIchemywithXerophayze-jt1gg Sorry, I should have mentioned that I'm following the instructions on the video. Maybe I'm just getting unlucky if my results are so different. Could it be because you're using that A1111 forge and I'm using the standard A1111?
I've never used onetrainer before. Honestly I've never trained my own littl Lora before either. Maybe this is a good opportunity to jump into that and see.
Hi there, I installed Forge after watching your video, but I get an error like "TypeError: 'NoneType' object is not iterable". Do you have any suggestions for a solution?
You know I was getting that error message but only in specific situations. I can't remember exactly what it was though that was causing me to get that error. Join my discord and give me a few more details about your settings. discord.com/invite/F38wKFgP
This stuff seems helpful, but not exactly sure about who uses them. I hope you understand that beginners are watching this because they assume such models can help them with LoRA training. I mean, I can only imagine someone producing character rotation diagrams for training, but not inside such 'photo frames'. It would be helpful if you instead make a video about using the same seed to create full-resolution character datasets (diff. background, diff. cloths, but the same character in each individual image) that can be fed into Kohya or OneTrainer for training. I mean, some people are already doing that, but info on how to easily produce those can be hard to find or is available in bits here and there.
You inspired me. I did a whole other video today similar to this one and my other character sheet one, but this I go into detail about how to set it up and split the images so that you have an image set that you can use to train LoRa's. check it out later today: th-cam.com/video/A1IyMM9VO6A/w-d-xo.html
I'm confused, your advertising character consistency but your only showing a method on how to make character sheets instead of showing examples on how you can use these sheets to make your characters actually consistent in regular generations
This was more for character generation and keeping the characters consistent. In a normal generation with an entire scene, that is an entirely different process honestly.
![SAFEPLANET - ดาวเคราะห์ [ THE PLANET ]](http://i.ytimg.com/vi/4X1XL3I_k3c/mqdefault.jpg)
***UPDATE*** Make sure you check out my GitHub for a new Python script that I created that gives you a nice interface for splitting these character sheets up. you can find it here:
github.com/Xerophayze/splitter
You can use the open pose preview images. Just turn off the pre-processor.
Good to know!
Needed this so I can make some references for Blender. Thanks, gonna try it out.
Glad I could help!
Yeah but can you change small details like belts and holsters and piping like you can with Krita AI in your character creation
Yes within painting you can change a lot of that. As for being specific when you're rendering these character sheets out, not really.
First of all, thank you for this tutorial, this is something I was looking for a while. Have you tried generating a small rotation around you character, full body or portrait? If there is enough consistence, you can probably create a character that can be displayed in a 3D display. I want to try this technique with Instant ID to see if it is possible to generate a 3D portrait from a 2D photo.
I haven't yet
Great tutorial thank you! Where did you get the turbo scheduler (dpm++ 2m SDE turbo) though? Can't find it anywhere online
Not sure but I believe that it's only available if you're using automatic 1111 forge edition.
This is excellent. Do you have something for :
a) comfyui (and or)
b) uses an existing character to build on?
Sorry neither
I think you'll love using stable swarm ui
I've heard of it. Haven't watched anything on it yet. I'll look into it. Gotta finish the updates to XeroGen first.
@@AIchemywithXerophayze-jt1gg I thought you'll love it since you're into using interfaces like A1111. Stable swarm ui is comfy ui but with an A1111 like interface. You can also use comfy ui nodes in it
What is the recommended to use? I only have juggernaut and the default one. I am trying to generate a minotaur but i keep getting a full body human 😅
I just used my prompt generator to generate this prompt and it works great with Juggernaut.
Wide view digital fantasy illustration, (Enormous Minotaur standing amidst ruins:1.2), Rule of thirds composition, Strong horns, Massive axe, (Ancient temple backdrop:1.3), Crumbling columns, Overgrown vegetation, (Eerie atmosphere:1.2), Moody lighting, Weathered textures, Dramatic shadows
@AIchemywithXerophayze-jt1gg thanks but I am trying the minotaur in a character sheet
Ok make sure you watch my new video, it may help
I keep trying to apply the poses from an openpose sheet, but SD keeps acting like it's not even picking up controlnet even with the 'controlnet is more important' selected. Any ideas?
Yeah usually if it's doing that, the controlnet version or the model are not compatible. So whatever model or checkpoint you're using, you need to change it to something else that will be compatible with the controlnetmodel you're using
Thanks. I have a question, how do you extract the characters from such different angles and then use them?
You would want to render the image initially as a PNG, then you can use a background removal tool to remove the background and convert it to a transparency, which means that it's a PNG with an alpha channel. That way you can import the image into another image and overlay it on top of another image.
@@AIchemywithXerophayze-jt1gg Thanks. Can I use that to make my facial expressions and poses more diverse with a consistent character like Lora
Can these character full body and face sheets just be created unclothed in order to train a character embedding or lora so then the clothing and poses can be added later with openpose and prompting?
I don't see why not, it's just a matter of how you prompt it
um hi may i ask how do i install it it look like so confuse?
I would start by going to and installing stability matrix, and then installing the latest version of forge. I believe you can grab my templates from my Google share.
I’m having trouble with installing the comfy-UI on my computer to work properly.
I would recommend using stability matrix and then use that to install comfy UI.
how did u get the openpose xl2?
huggingface.co/thibaud/controlnet-openpose-sdxl-1.0/blob/main/OpenPoseXL2.safetensors
which openpose model are u using? i cant get openpose to work on SDXL
I'm using the openposeXL2 model.
huggingface.co/thibaud/controlnet-openpose-sdxl-1.0/blob/main/OpenPoseXL2.safetensors
How do you get characters to have good faces in these model sheets? Whenever my caracters are created far away for model sheets, the faces are always warped.
Whenever I generate these, I always include adetailer for the faces.
@@AIchemywithXerophayze-jt1gg Sorry, I should have mentioned that I'm following the instructions on the video. Maybe I'm just getting unlucky if my results are so different.
Could it be because you're using that A1111 forge and I'm using the standard A1111?
Can you also make a tutorial to use these workflows and make a unique character lora using onetrainer?
I've never used onetrainer before. Honestly I've never trained my own littl Lora before either. Maybe this is a good opportunity to jump into that and see.
I forgot to ask how would able to use that negative prompt?
Which negative prompt? Are you talking about the one I have in my styles drop down? You can find it at share.xerophayze.com
@@AIchemywithXerophayze-jt1gg yes
Great Tuto
Hi there, I installed Forge after watching your video, but I get an error like "TypeError: 'NoneType' object is not iterable". Do you have any suggestions for a solution?
You know I was getting that error message but only in specific situations. I can't remember exactly what it was though that was causing me to get that error. Join my discord and give me a few more details about your settings.
discord.com/invite/F38wKFgP
Great video , template link is not working though :)
Try now. Someone else pointed that out as well. It should be fixed now.
can the character be positioned in "A POSE"?
That would require utilizing an original character that's already in the pose and then using control nets open pose
This stuff seems helpful, but not exactly sure about who uses them. I hope you understand that beginners are watching this because they assume such models can help them with LoRA training. I mean, I can only imagine someone producing character rotation diagrams for training, but not inside such 'photo frames'. It would be helpful if you instead make a video about using the same seed to create full-resolution character datasets (diff. background, diff. cloths, but the same character in each individual image) that can be fed into Kohya or OneTrainer for training. I mean, some people are already doing that, but info on how to easily produce those can be hard to find or is available in bits here and there.
You inspired me. I did a whole other video today similar to this one and my other character sheet one, but this I go into detail about how to set it up and split the images so that you have an image set that you can use to train LoRa's. check it out later today: th-cam.com/video/A1IyMM9VO6A/w-d-xo.html
I'm confused, your advertising character consistency but your only showing a method on how to make character sheets instead of showing examples on how you can use these sheets to make your characters actually consistent in regular generations
This was more for character generation and keeping the characters consistent. In a normal generation with an entire scene, that is an entirely different process honestly.
program? name
If you're talking about what program I was using to create those, it's called WebUI Forge. It's like automatic 1111
Hello! Can I do this on google collab?
I know a lot of other people do it, I don't know how to set it up and I've never done that before.
Get a better microphone imo
Yeah it's been brought up before. Had the game turned down too much on my microphone.