CivArchive
    Naoe & Yasuke [Assassin's Creed Shadows] Illustrious/Hunyuan/Flux/PonyXL - PonyXL v1.0
    NSFW
    Preview 12703757
    Preview 12703780
    Preview 12703786
    Preview 12703789
    Preview 12703805
    Preview 12703818
    Preview 12703829
    Preview 12703843
    Preview 12703848
    Preview 12703860
    Preview 12703934

    Trained on screengrabs from the recent AC: Shadows trailer and some screenshots. It naturally leans towards 3D/realism as the entire dataset is of that style.

    I mainly made this as an experiment as I was curious how well this would work with two subjects in one LoRA while using a fairly bad dataset, though the dataset is greatly improved after the first PonyXL v1.0 version.

    Hunyuan video usage and notes:

    • Only trained on Naoe.

    • Naoe prompt: Naoe woman

    • You can include wearing a ninja outfit if the AI is hesitant to add the outfit, though this isn't normally required.

    • For full prompt examples, you can download my video uploads and load them into ComfyUI.

    Illustrious usage and notes:

    • Naoe prompt: 1girl, ACNaoe, black hair, short hair, brown eyes, breasts, dark blue ninja outfit, [red sash::0.3], sword on back

    • Add hood for Naoe's hood.

    • Yasuke prompt: 1boy, ACYasuke, facial hair, dark-skinned male, brown hair, samurai armor

    • I suggest using the above tags as starting points and change them accordingly by removing tags you don't need.

    • Yasuke is undertrained compared to Naoe, so the model isn't as consistent with him.

    • For full prompt examples, look at metadata for the image uploads.

    PonyXL usage and notes:

    • Naoe prompt: 1girl, Naoe, brown eyes, short black hair, japanese clothes, sash, arm guards, pants

    • Add hood for Naoe's hood.

    • Yasuke prompt: 1boy, Yasuke, brown eyes, black hair, japanese armor

    • Add helmet to positive or negative prompt to add or remove Yasuke's helmet. You probably need to remove black hair tag when adding the helmet.

    • When showing both characters, the brown eyes and black hair tags can be skipped for one character.

    • It's recommended to use a tag to suggest a pose as the model can default to making the characters stand with arms straight.

    • For full prompt examples, look at metadata for the image uploads.

    Flux usage and notes:

    • Naoe prompt: A ACNaoe woman with dark brown, shoulder-length, and slightly messy hair wearing a dark blue ninja outfit

    • Yasuke prompt: A ACYasuke man wearing samurai armor

    • You can make the character prompts stronger by adding more relevant keywords by looking at the training captioning examples below. For instance, if you want to ensure Naoe's outfit is more accurate, I recommend mentioning arm guards, hood and red belt.

    • For full prompt examples, look at metadata for the image uploads.

    Full caption example for Naoe:

    • a highly detailed, CGI-rendered image of a ACNaoe woman with a rugged, adventurer-like appearance. She stands against a plain, dark background, which makes her the focal point. Her skin has a warm, golden undertone, suggesting a mix of ethnicities. Her hair is dark brown, shoulder-length, and slightly messy. She wears a dark blue, hooded tunic with a loose, flowing design, cinched at the waist with a wide, red leather belt adorned with intricate, golden buckles. Her outfit includes a pair of rugged, earth-toned pants that are tucked into knee-high, brown leather boots with visible laces. Her arms are adorned with a mix of leather and metal armor, featuring a combination of dark brown and metallic silver tones. Her hands are gloved, with the gloves appearing to be made of a durable, possibly leather material. Her accessories include a sword at her side, with a scabbard attached to her belt, and a quiver on her back, suggesting she is a skilled archer. The overall aesthetic is one of practical, battle-ready gear, with a focus on functionality and durability. The image exudes a sense of readiness for adventure and combat. The background is a solid, dark color, which helps to emphasize the subject's attire and accessories. The lighting is even, with no shadows, making the textures and details of her outfit and gear stand out. The style is reminiscent of fantasy adventure games or anime-inspired art.

    Full caption example for Yasuke:

    • a highly detailed, CGI-rendered image of a ACYasuke man, showcasing a realistic and intricate design. The warrior is depicted in a traditional Japanese samurai armor, with a robust build and a strong, rugged appearance. The armor is a mix of dark brown and gold tones, adorned with intricate patterns and designs, including a prominent golden circular emblem on the chest plate. The warrior's face is stern, with a short, dark beard and a few braids of hair, adding to the rugged and battle-hardened look. The arms are muscular, with visible scars and tattoos, suggesting a life of combat. The armor itself is a masterpiece of craftsmanship, with each piece carefully detailed to reflect the historical accuracy of samurai armor. The gloves are articulated, with visible fingers, and the boots are sturdy, with intricate lacing. The background is a dark, gradient gray, which helps to focus attention on the warrior. The overall style is realistic and detailed, with a focus on historical accuracy and cultural authenticity. The image is likely from a digital art or CGI animation, given the level of detail and realism. The texture and shading are highly realistic, with a focus on showcasing the intricate details of the armor and the warrior's physique. The atmosphere is one of intensity and readiness for battle. The image is a blend of historical and cultural accuracy with a modern CGI rendering style, making it a striking and immersive visual experience.

    Description

    FAQ

    Comments (19)

    Sanchez_May 17, 2024· 17 reactions
    CivitAI

    based.

    PangolinMasterMay 17, 2024
    CivitAI

    I pray they get the modern day stuff back on track with this next installment

    Nicozlukaz1990May 18, 2024· 1 reaction
    CivitAI

    I know it would be a lot to ask but could you also release it in a light version SD1.5 please... please... thank you... by the way this is epic

    WizardMcMerlin
    Author
    May 18, 2024

    Sorry, I only train Pony models for now.

    Nicozlukaz1990May 20, 2024

    @WizardMcDuck ok ...thank you

    Icey64Jun 7, 2024

    @Nicozlukaz1990 if you're willing to share your dataset, I could try making an SD1.5 version :)

    WizardMcMerlin
    Author
    Jun 7, 2024

    @Icey64 I uploaded the dataset I used. I imagine you'll need to re-do the captions for 1.5 since I went with "danbooru-style" tags, and you can probably improve the dataset by throwing out a lot of the low-quality images in the first folder. I didn't bother to do a lot of curation of dataset images when making the LoRA.

    Edit: I just realized I forgot to crop some of the images, so that's another way to improve the dataset.

    Nicozlukaz1990Jun 8, 2024

    @Icey64 @Icey64 I don't know what the data is, it could be that thing that is shaped like a lightning bolt, so I passed 100 ok

    Icey64Jun 10, 2024

    @WizardMcDuck thank you! Will take a look :)

    Nicozlukaz1990Jun 11, 2024

    @Icey64 Yes, since you offered to do naoe, well I don't know about the data you are saying, the only thing I could give is 100 rays for the one in the left corner of the site

    Icey64Jun 14, 2024· 1 reaction
    Icey64Jun 14, 2024

    @WizardMcDuck I actually used your dataset basically as is, just culled a few repeated images but the tags works perfectly :) 

    WizardMcMerlin
    Author
    Jun 14, 2024

    @Icey64 Good job!

    Nicozlukaz1990Jun 15, 2024

    @WizardMcDuck Yes, it turned out well but as I always say... you always have to try... no... and it's not to be shameless but... honestly you shouldn't also forget about the people who don't have as much to offer. Pony or XL versions

    WizardMcMerlin
    Author
    Sep 27, 2024

    @Icey64 You can let me know if you want me to upload the updated dataset in case you want to re-train your SD 1.5 variant of the model.

    Icey64Jun 7, 2024· 1 reaction
    CivitAI

    This Lora looks great, can't wait to try it! Potential dataset update with the upcoming Ubisoft Forward?

    WizardMcMerlin
    Author
    Jun 7, 2024

    I'm not sure if I will. I usually prefer making new LoRAs than update the ones I have. Unless there's a new, shiny base model to upgrade to.

    WizardMcMerlin
    Author
    Sep 27, 2024· 1 reaction

    Looks like I ended up updating the dataset, after all. This new version should be noticeably better.

    Icey64Sep 27, 2024

    @WizardMcDuck looks great, thanks for your efforts! Makes up for the delay

    LORA
    Pony

    Details

    Downloads
    187
    Platform
    CivitAI
    Platform Status
    Available
    Created
    5/17/2024
    Updated
    6/22/2026
    Deleted
    -
    Trigger Words:
    1boy, Yasuke, brown eyes, black hair, japanese armor
    1girl, Naoe, brown eyes, short black hair, japanese clothes, sash, arm guards, pants

    Files

    ACShadows_t2-000008-128.safetensors

    Mirrors

    ACShadows_t2.zip

    Mirrors

    CivitAI (1 mirrors)