CivArchive
    Realmforge: Fantasy Anime - r3alm_fantasy_anime
    Preview 149530615
    Preview 149529662
    Preview 149530587
    Preview 149530750
    Preview 149530657
    Preview 149531268
    Preview 149531715
    Preview 149531754

    Realmforge - Fantasy Anime

    A style LoRA for MiniMax-H3-Image that produces heavily inked, hatched anime illustrations with a gritty seinen-manga look, inspired by the art of Vinland Saga. Strong linework, deep shadows and a single glowing accent colour against dark, moody scenes. It works across characters, creatures, environments and objects.

    Trigger

    r3alm

    Start your prompt with:

    r3alm, ...

    The image is a r3alm anime illustration of ...

    The image is a vertical r3alm anime illustration of ...

    The image is a horizontal r3alm anime illustration of ...

    - Strength: 0.8–1.0

    - Sampler / scheduler: lcm/er_sde/euler_ancestral / simple

    - Steps: 6

    - CFG: 1.0

    - Resolution: 3552 × 4736 (3:4 portrait), or anything around 16 MP+

    Prompting tips

    This LoRA was trained on long, plain-English captions in H3's native prompt style, so it responds best to the same:

    - Write 150–250 words of flowing prose, not tag lists.

    - Describe composition → subject → pose → setting → light, in that order.

    - Describe light as physical things in the scene: "violet flames rising from the skull light the ribs from below", "moonlight rims the fur".

    - Name the colours of glowing effects; the style loves one strong accent colour (red, violet, cyan, green, orange) against a dark scene.

    Training details

    40 images · auto-captioned in ZenTrainer with a custom system prompt · LoRA rank 16 · AdamW 2e-4 · 2500 steps · 768→1024 px · trained with [ZenTrainer by Nynxz] (ComfyUI custom node)

    Dataset source

    The training images are my own Midjourney generations.

    License / usage

    Free to use, and credit is appreciated but not required. You may not resell or redistribute this LoRA file itself, including behind a paywall or as part of a paid model pack.

    This LoRA is a modification of (a Model Derivative of) MiniMax-H3-Image, and its use is also governed by the MiniMax-H3-Image Community License Agreement: MiniMax-H3-Image is licensed under the MiniMax-H3-Image Community License Agreement, Copyright © 2026 MiniMax. All Rights Reserved. By downloading or using this LoRA you agree to the terms of that license, including its territorial limits, its use restrictions (Section V) and its Acceptable Use Policy (Exhibit A).

    Where my terms above and MiniMax's license differ, the stricter one applies. Read the license before use. Powered by MiniMax-H3-Image.

    Description

    FAQ

    LORA
    MiniMax H3 Image

    Details

    Downloads
    124
    Platform
    CivitAI
    Platform Status
    Available
    Created
    11/19/2026
    Updated
    11/20/2026
    Deleted
    -
    Trigger Words:
    r3alm

    Files

    r3alm.safetensors

    Mirrors

    CivitAI (1 mirrors)