📲 Commissions, Training, Consultation - Flux, Pony, SDXL + Characters, Styles, Faces + ❤️ Support 🫶 Discord
ℹ️ How was it made?
ℹ️ This model was trained on the first three episodes of the TV show Game of Thrones. 9k images focused on characters' faces (50 subjects in total), and 4k images from different scenes. Additionally, 30k images were used as regularization images - medieval-themed images and half of the ❤️🔥 Divas dataset.
ℹ️ Trigger word for style is "game of thrones" - using it negative prompt can help to make brighter images. Characters' names that can be used as trigger words:
ned stark, jon snow, catelyn stark, daenerys targaryen, robert baratheon, joffrey baratheon, tyrion lannister, jaime lannister, robb stark, petyr baelish, littlefinger, cersei lannister, viserys targaryen, jorah mormont, benjen stark, waymar royce, rakharo dothraki, will ranger, jeor mormont, doreah handmaid, irri dothraki, maester luwin, khal drogo, the hound, sandor clegane, lord varys, rast ranger, rodrik cassel, maester aemon, grenn ranger, gared ranger, theon greyjoy, ser barristan selmy, jory cassel, ros cook, yoren recruiter, royal steward, catspaw assassin, pypar steward, old nan, illyrio mopatis, ilyn payne, lancel lannister, renly baratheon, mikken blacksmith, alliser thorne, syrio forelℹ️ This model is based on ❤️🔥 Divas model - original training, remixed recipe, and half of the dataset used for regularization.
ℹ️ My goal was to achieve high accuracy of characters, which was challenging due to the limited amount of frames/angles/lighting/scenes of many characters. I had to pull a few tricks out of my sleeves, but I think that most of the characters look very accurate.
Description
ℹ️ This model was trained on the first three episodes of the TV show Game of Thrones. 9k images focused on characters' faces (50 subjects in total), and 4k images from different scenes. Additionally, 30k images were used as regularization images - medieval-themed images and half of the ❤️🔥 Divas dataset.
ℹ️ Trigger word for style is "game of thrones" - using it negative prompt can help to make brighter images. Characters' names that can be used as trigger words:
ned stark, jon snow, catelyn stark, daenerys targaryen, robert baratheon, joffrey baratheon, tyrion lannister, jaime lannister, robb stark, petyr baelish, littlefinger, cersei lannister, viserys targaryen, jorah mormont, benjen stark, waymar royce, rakharo dothraki, will ranger, jeor mormont, doreah handmaid, irri dothraki, maester luwin, khal drogo, the hound, sandor clegane, lord varys, rast ranger, rodrik cassel, maester aemon, grenn ranger, gared ranger, theon greyjoy, ser barristan selmy, jory cassel, ros cook, yoren recruiter, royal steward, catspaw assassin, pypar steward, old nan, illyrio mopatis, ilyn payne, lancel lannister, renly baratheon, mikken blacksmith, alliser thorne, syrio forelℹ️ This model is based on ❤️🔥 Divas model - original training, remixed recipe, and half of the dataset used for regularization.
ℹ️ My goal was to achieve high accuracy of characters, which was challenging due to the limited amount of frames/angles/lighting/scenes of many characters. I had to pull a few tricks out of my sleeves, but I think that most of the characters look very accurate.
FAQ
Comments (25)
actually a fun model
im actually waiting for next version on next season or something
Lol, don't hold your breath.
This looks great. Thank you.
Would love to see such checkpoints being created for such iconic TV series such as the Original Star Trek (Spock, Kirk, etc.), or the series VIKINGS.
Thank you for the suggestion.
Wow this is interesting, Curios question, How long did it take to train the whole model?
The model was trained in 3 cycles of training, 420 epochs in total. I was saving checkpoint every 5 epochs, which was happening exactly every hour. So if you only interested in GPU time, it took 3.5 days on RTX 4090. The whole process of preparations to create the dataset, defining settings, training, mixing and testing, took 3 weeks.
@alexds9 WOW 😮
Damn everything about this model is perfect, the Quality, subjects, size, Would be amazing seeing an article or guide how dataset is gathered and how this is trained, i didn't even thought this was possible, Thank
This is awesome. I also can't wait for you to expand it to scenes and compositions as well, include the dragons at different ages and make it SDXL. Just saying!!!!!!!!
This could be the model series that allows us to fix the last season :D :D
Great checkpoint, but how to trigger a different age for a character? For example, Aria and Bran changed a lot.
Currently, the model is trained on the first three episodes of the show, so you can't trigger older ages now.
If I make another model for a later season of GOT, only then you will be able to prompt for older ages.
@alexds9 ahh this makes sence
Great work done, but very specific, hopefully the target audience will appreciate it.
The source material from GOT is very challenging for training.
You can see here: how it was made.
Maybe a LoRA for this? Please and thanks!
Sorry for the delayed response, the site has issues with notifications sometimes...
Lora can hold much less content in comparison to a checkpoint.
Due to amount of characters involved (50), I don't think it's possible to train a Lora instead of a checkpoint in this case.
I'm giving Civitai generation service another chance. Hopefully, users will try to understand how to use SD before reviewing models.
[Original message:]
Civitai generation service has been blocked due to inexperienced users using it to generate low-quality images and leaving bad reviews.
this has to be one of the greatest checkpoints ever made... 10-out-of-10. just an incredible amount of work... well done!
Thank you very much.
🫶 Discord ❤️️
@alexds9 A bit late in the game, but is there any chance or hope for a Pony or Flux-based version of this checkpoint? This was among the best 1.5 checkpoints I've ever used and you appear to have a talent!
@xdsp
I would love to give more attention to my older models. Currently, I actually work on transferring a few of my models to Pony and Flux. Flux is working really great with training, but I haven't experimented yet with multiple subjects training, which needed for this model. I'll need it for other models too, so there is a good chance I will get to GOT too.
I'm late on this and unfortunately I myself have finally moved to SDXL/PDXL so I won't be using this, well maybe I'll backtrack on my laptop sometime. Anyway, I just want to thank you for being another of the "VERY" few people who do not just waste space on CivitAI with a whole screen of single character LoRA's. Great work!
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.