The new RPG v6 Beta and all future updates have moved here.
Support for RPG v5:
User Guide here: RPG User Guide v4.3
Official Youtube User: @singularity-ai
Join me on Discord: https://discord.gg/PcvEs8eVpj
Leonardo AI: https://leonardo.ai/
RunDiffusion: https://rundiffusion.com/
StableHorde: https://stablehorde.net/
STATUS: RELEASE
VERSION 5.0
STATUS: RELEASE
VERSION 4.0
I have built a guide to help navigate the model capacity and help you start creating your avatar.
Download the User Guide v4.3 here: RPG User Guide v4.3
If you like the model, please leave a review!
This model card focuses on Role Playing Game portrait similar to Baldur's Gate, Dungeon and Dragon, Icewindale, and more modern style of RPG character.
The model is the result of various iterations of merge pack combined with Dreambooth Training. This is a work in progress. If you are interested to help, reach me via my Discord server, user Anashel.
Description
575 000 traning steps added
FAQ
Comments (54)
why are all of your v5 previews missing generation data?
As I mention in the image, i'll upload the meta this week! :)
@Anashel it's not so much that you haven't added the data, it's that the images you uploaded have had their metadata scrubbed, why?
@WeirdScienceX The workflow (i'll put that in the V5 user guide) was prompt2image > upscale > left outcrop. The left outcrop part end up removing all meta data, so I have to add it manually when I upload the image.
@Anashel ah, so you generate ~cinemascope ar images with the character in the middle and crop it to create those previews?
@eurotaku Not sure I understand what you mean. It's very basic; 1) create the main image with a second pass (if you use Vlad1111) or hi-rez (if you use Auto1111), 2) Then you uncrop to 1920 on the left of right side and autofill. There is some outcrop / outpainting plugin in auto, or you can use Photoshop Beta if you have the chance to have access to it. It's just one click uncrop no promp.
@Anashel Vlad is actually currently broken in terms of embedded metadata. This is why my images are missing resources when I upload them. https://discord.com/channels/1101998836328697867/1133120874111836362/1133242381651677294 says it will be fixed tomorrow.
@Anashel clearer now, so you first create the character only and then outpaint to get a landscape ar
@eurotaku Exact, I know there are now two pass control net plugin that let you do the landscape then the character so you can really have high-def portion of both, but this is the 'quick and dirty' version. :)
@demoran Ah! Didnt knew that, it's crazy how fast and involve he is. I asked for a QOL fixes and it was done in an hour... really he is insanely dedicated!
I added most of them! Very sorry for the long delay.
Where is version 6?
Says on their patreon, so probably a delayed release type thing.
Version 6 is still in project phase. The V5 is the version just released.
I have release RPG v5 24h ago. I am just starting training RPG v6.
@Anashel is v6 still gonna be trained on 1.5? will rpgxl be a completely seperate project?
I understand! Thank you!
@eurotaku No V6 Will be native SDXL. I will not train 1.5 anymore, so V5 will be the last 1.5 base model.
@Anashel Great to hear... even though SDXL models atm are a tad restricted/censored it's clear the quality they can produce even non fine tuned is far above what 1.5 was achieving this early in its lifespan.
for V5 are we getting an updated User Guide or is it like the same thing as V4.3 User Guide
I believe I saw him say on reddit that he is working an updated pdf guide.
@randallaware Yes you are right, I am working on a RPG v5 guide.
@Anashel Are there big changes or more of the same with some tweaks
@empessah925 For now I will keep it as v4.3, it work perfectly with V5 so far. I did added the prompt on most of the image here. I'll probably add also my control grid, around 30 prompts I used to test the model.
that's a lot of tokens to get any usable output
You don't even need a prompt to get usable output with any model. If the model is trained to output high-quality images, such as this model and most other models, it will, by default, output a high-quality image with zero prompt.
Sorry you have problem with the model. I made a user guide for the model, if this can be of help. I also added around 100 control net model to help you design various character. https://civitai.com/models/115223/rpg-v4-guide, it's a comprehensive 43-page PDF guide that includes sample prompts and numerous other tips!
Pose ControlNet seems to not work with this model for some reasons, I am using the Draw Things mac/ios app. Am I doing something wrong?
I have uploaded around 100 control net but I do only use depth, I never try it with pose. Is depth also not working for you?
Can't wait for V6 XL!
V6 is on pause right now, I struggle too much with the training. Am i yet to find a descent training tools that work well with SDXL. It's sad since I kept all my high-rez version of my original content before i scale it down for SD 1.5.
Too bad this project : https://github.com/jamessealesmith/continual-diffusion is being gate keeped. If the training code is released, it may be useful
Checkpoint doesn't understand well when you ask it to generate a man instead of a woman. It needs to write MALE MASCULINE MAN in the positive and FEMALE FEMININE WOMAN in the negative, then you will have at least 6 men and 4 woman for 10 generations.
Sorry, it's indeed a challenge with that model. I made a user guide for the model, if this can be of help. I also added around 100 control net model to help you design various character. https://civitai.com/models/115223/rpg-v4-guide, it's a comprehensive 43-page PDF guide that includes sample prompts and numerous other tips! But the use of negative to make sure you have the right gender is something you will have to do in some case.
man?
I like your model but I would like to ask what is the difference between version 4 and 5? thank you
More training :) You should get much better result from V5, more realistic style
Any updates on this model or others, it's been a while. @Anashel are you still planning/working on them? :)
My new project!! https://github.com/Anashel-RPG/echoai :)
@Anashel I like some of those demo images, definatley has potential. I don't suppose normal prompt info is available for those, such as: - https://github.com/Anashel-RPG/echoai/raw/main/readme_img/cycle38.jpg for example? - in case, I'll be watching with interest. ;)
@emmasteadman Yes, on windows just do properties on the image and go to the details panel. On mac, you will need a website like https://jimpl.com/
@Anashel I looked at your project, it looks really promising but sadly not for me though. On top of having my own PC/decent GPU, using another MidJourney like company with heavy policing and censoring filters is of no appeal for me.
@Anashel I looked at your project, it looks really promising but sadly not for me though. On top of having my own PC/decent GPU, using another MidJourney like company with heavy policing and censoring filters is of no appeal for me.
@Aerth It run with the API you want. So far I have integrated local SD auto1111, Leonardo, Prodia, CivitAI, Mijourney Beta and Dall-E. So, basically you run it on what you want. i'm working on a more advanced meta-data structure so you can trace the entire generation flow. Anyway, still a long way to go before release!
@Anashel Ah, thank you, glad to hear that :)
Very flexible and versatile, with very good photorealism, unfortunately sometimes buggy, but overall still a very decent model!
8.5mb ???
Only one race yet. Next one has two race and plate armor and it jumped to 30 mb. So by the time I have fully train all the high rez asset of RPG v5 + my new assets, it will be much lager.
@Anashel Their question is because of the fact that this isn't posted as a Lora, which I assume it is, it's instead posted under the category of being a trained checkpoint.
rpg v5 doesn`t work after 15aug. What`s wrong? SD can`t switch model now.
I did not change anything. You download it on your machine and it doesnt work?
I downloaded it before 15aug and it was working correctly. but after 15 aug sd cannot switch model to rpg. I downloaded the model again and it dont work (((
Can produce good photo portraits with the correct negative embeddings.
Please make this for sdxl it will be amazing in sdxl