CivArchive
    The Deep Abyss FLUX - V1
    NSFW
    Preview 39113774
    Preview 39113144
    Preview 39113044
    Preview 39113046
    Preview 39113067
    Preview 39113191
    Preview 39113070
    Preview 39113078
    Preview 39113734
    Preview 39113903
    Preview 39113927
    Preview 39115046
    Preview 39133904
    Preview 39133976
    Preview 39771994

    This is a very abstract creature lora, trained on a style board of mine that I have curated over a series of months. The more detailed and abstract your prompts, the better. Highly recommend using an LLM to help you think outside of the box on this one.

    Use the trigger word 4byss in front of everything for best results <3

    If you make any cool things with the LoRA, please share them in the gallery for others to see!

    Description

    FAQ

    Comments (15)

    machinedelusionsNov 9, 2024· 2 reactions
    CivitAI

    YEET

    makasakemix445Nov 11, 2024· 2 reactions
    CivitAI

    HOLEY SMOOKES

    jboogx_creative
    Author
    Nov 11, 2024

    Make something cooooooool with it!

    nkVasiNov 13, 2024· 2 reactions
    CivitAI

    Hi! Please tell me how do you make a model of such a small size?

    Is it possible to train a LORA of the same size (or even smaller) outside of CIVITAI?

    jboogx_creative
    Author
    Nov 13, 2024

    Hey there! Unsure, since i've only just started training LoRAs and i've only used the Civitai trainer! ALl of the settings and details for what I did are on our Civitai YT channel in a detailed video w/ @kftiger

    https://youtu.be/5C-6eoWe0yA

    DaemonratNov 13, 2024· 2 reactions

    Smaller DIM = smaller lora. DIM stands for network dimension and it's usually a multiple of 8. Typically people use a DIM of 64 or 128, but you can train with a DIM as low as 8.

    kftigerNov 13, 2024

    Thanks for the callout @jboogx_creative. It comes down to what you set your Network Dim (how much learning data is stored in the LORA) to. As FLUX is an incredibly powerful/knowledgeable model, you don't necessarily have to set it high, especially for concepts like this (I believe jboogx used the default setting of 2). If you're training elsewhere, so long as you have access to changing Network Dim, you could even set it at 1 if you wanted (but no guarantees how it would turn out!)

    jboogx_creative
    Author
    Nov 14, 2024· 2 reactions

    @kftiger 2 may have worked great for my LoRA's because of how small the actual dataset was. Good point to keep in mind. Most of my LoRA's used 17-35 images

    ModernMythosNov 17, 2024· 2 reactions
    CivitAI

    Hey Tyler! I really love this Lora but finding it a little difficult to maintain sharp details, have you any tips for keeping outputs crispy in comfyui ? : )

    jboogx_creative
    Author
    Nov 17, 2024

    It's only gotten the most crispy in the upscale. It also depends if you're mixing loras. I've had the most success with mixing one to two lora's but the upscale is crucial! Also, the more detailed the prompt, the better!

    coldwoodDec 2, 2024· 1 reaction
    CivitAI

    cool

    VoltesDec 25, 2024· 1 reaction
    CivitAI

    Time to make slumbering monsters from R'yleh

    jboogx_creative
    Author
    Dec 25, 2024· 1 reaction

    PLEASE DO! AND SHARE SOME!

    VoltesDec 26, 2024

    @jboogx_creative made some! they are not showing up to pose. Also. Why this one takes more buzz to generate? I am new at this.

    jboogx_creative
    Author
    Dec 31, 2024· 1 reaction

    @Voltes Probably because the lora is on Flux, and flux is a little more expensive :)

    LORA
    Flux.1 D

    Details

    Downloads
    849
    Platform
    CivitAI
    Platform Status
    Available
    Created
    11/8/2024
    Updated
    5/12/2026
    Deleted
    -
    Trigger Words:
    4byss

    Files

    The_Deep_Abyss-000021.safetensors

    Mirrors

    Available On (1 platform)

    Same model published on other platforms. May have additional downloads or version variants.