CivArchive
    AnimaDrones-V2.3 - v1
    NSFW
    Preview 135134871
    Preview 135134884
    Preview 135134716
    Preview 135134848
    Preview 135134687
    Preview 135134721
    Preview 135134819
    Preview 135134997
    Preview 135134725
    Preview 135134773
    Preview 135134774
    Preview 135134788
    Preview 135134803
    Preview 135134902
    Preview 135134941
    Preview 135134955
    Preview 135134974
    Preview 135134988

    (Generation on CivitAI only works with the Lora version: https://civarchive.com/models/2737606/animadrones-v1-lora )

    The Only Full Murder Drones Finetune (V2.3 Update)

    (On CivitAI)

    V2.3 is rained on more than 6000 images, this Full Finetune achieves unparalleled generalization across everything that makes Disassembly Drones and Worker Drones destinct.

    I scraped literally every booru and image site and extracted and filtered every single one of the 300 000 frames of the ENTIRE show. I even finetuned a specialized CLIP model for the captions!

    Overall the distribution of the characters is wildly unbalanced, meaning: Very good at Uzi, V, J, Cyntessa, N, Doll and Lizzy; less good at Cyn and pretty bad at everything else. But with a specific enough prompt you can certainly bring every character from the Murder Drones universe on the canvas.

    I'm ngl this is a NSFW model first and foremost, so be warned!

    Please leave a like and a comment if you enjoy my models!

    v2.3

    A ground up retrain of the model on much more data, for longer and with multi caption.

    • MUCH better character consistency and prompt following

      • The model can generate MORE characters at once. Without mixups!

      • If mixups ocure, specify the hairstyle of the characters and that should give perfect accuracy.

    • MORE freedom with artstyle

      • Since the practically the entire show has been trained into this model you can switch two major artstyles!

        • Illustration

        • 3d

    • Best performance with a two stage workflow. You can get mine by downloading one of the example images.

    v2.3: ComfyUI + "res_2s" + sgm_uniform + CFG 4 + 30 steps

    Dataset Distribution

    Number of images ≠ Quality of character (at least not always)

    • UZIDOORMAN - 1960x

    • SD-N - 1897x

    • SD-V - 1590x

    • SD-J - 1140x

    • CYN - 546x

    • TESSA - 536x

    • DOLL - 271x

    • CYNTESSA - 242x

    • THAD - 172x

    • LIZZY - 164x

    • ALICE - 94x

    • NORI - 84x

    • KHAN - 75x

    • YEVA - 69x

    • SENTINEL - 64x

    • REBECCA - 22x

    • TESSA-ELIOT - 20x

    • DOLLS-DAD - 12x

    v1.9

    • Better character consistency and prompt following

      • The model can generate many characters at once. Without mixups!

      • If mixups ocure, specify the hairstyle of the characters and that should give perfect accuracy.

    • A little less freedom with the artstyle

    • Best performance with a two stage workflow. You can get mine by downloading one of the example images.

    v1.9: ComfyUI + "er_sde" or "res_2s" + sgm_uniform + CFG 4 + 30 steps

    v1

    • Pretty good prompt following: Except it likes to mix up the topping and bottoming characters.

      • Sometimes this can be fixed by just changing the order of them in the prompt

    • Clothes arerelatively accurate, but you can change them to whatever you like

    • Best performance with a two stage workflow. You can get mine by downloading one of the example images.

    • It can do 3D, 2D and pretty much every other artstile Anima it self can do

    v1: ComfyUI + "er_sde" + sgm_uniform + CFG 4 + 30 steps

    Prompting (for V2.3):

    • Works best if you structure your prompt very well, so that every character tag can be cleanly associated with certain actions. Take my showcase images as examples!

    • If a character isn't generated correctly, type more describing traits: SD-V -> SD-V with short white hair, robot

      • BUT be aware that longer captions will ultimately degrade the models output! I noticed that this is the case with Anima in general.

      • Anima sometimes gives the drones human skin. Fix that by typing: robot

        • (Tbh "robot" should almost always be included)

    • The Model sometimes forgets the tail. Luckily you can remind it by typing: tail

    • This time tag-based prompts are king, but natural language works too in some extend.

      • Qwen3 is very flexible

    • Sometimes the negative prompt can mess things up. Keep it minimal!

    • Always specify the amount of drones in an image!

      • Like: 2 robots, robot, 1girl, 2girls, 1boy, etc

    Description

    First full version. A bit undertrained.

    FAQ

    Checkpoint
    Anima
    by Chaosnet
    Download (Beta) View on CivitAI
    robots
    sex
    serial designation v
    serial designation n
    v
    uzi doorman
    murderdrones
    nsfw
    n
    j
    murder drones
    uzi
    md
    cyn
    serial designation j
    character

    Details

    Downloads
    164
    Platform
    CivitAI
    Platform Status
    Available
    Created
    6/28/2026
    Updated
    10/1/2026
    Deleted
    -
    Trigger Words:
    UZIDOORMAN
    SD-V
    SD-J
    SD-N
    CYN
    TESSA
    CYNTESSA
    DOLL
    DOLLS-DAD
    NORI
    KHAN
    ALICE
    YEVA
    LIZZY
    SENTINEL
    THAD
    DRONE

    Files

    qwen_image_vae.safetensors

    Size:
    242.05 MB
    SHA256:
    a70580f0213e67967ee9c95f05bb400e8fb08307e017a924bf3441223e023d1f

    Mirrors

    HuggingFace (314 mirrors)
    CivitAI (256 mirrors)
    TensorFiles (1 mirrors)
    ModelScope (3 mirrors)
    ModelScope CN (6 mirrors)

    animadronesV23_v1_txt.safetensors

    Size:
    1.11 GB
    SHA256:
    cd2a512003e2f9f3cd3c32a9c3573f820bb28c940f73c57b1ddaa983d9223eba

    Mirrors

    HuggingFace (133 mirrors)
    CivitAI (235 mirrors)
    ModelScope (1 mirrors)
    ModelScope CN (2 mirrors)

    animadronesV23_v1.safetensors

    Size:
    3.89 GB
    SHA256:
    65b667d9513c4e56eb5b881d72f660f17d8c8ec9ea2bf9a79a0a842819185166

    Mirrors

    CivitAI (1 mirrors)