CivArchive
    RadioIllustrated - v2.0
    NSFW
    Preview 5477000
    Preview 5476770
    Preview 5476779
    Preview 5476782
    Preview 5476784
    Preview 5476763
    Preview 5476777
    Preview 5476972
    Preview 5476954
    Preview 5476817
    Preview 5476859
    Preview 5476789
    Preview 5476816
    Preview 5476821
    Preview 5476827

    RadioIllustrated is intended to be a 2.5D-esque model that leans towards an illustrated aesthetic without needing very complex prompting. As shown in the example images, it is capable of more 2D/3D generations (YMMV). While I am happy with where it is at right now, I do plan on experimenting more with the style in the future.

    Note: Is slightly NSFW-prone on prompts that should otherwise be SFW, but that's highly prompt and seed dependent.

    Usage Guidelines

    (Note, these are updated as of 2.0)

    Sampler Settings

    Dimensions: 512 width, 768 height, but it should work with 512x512 and 768x512 as well since those are the most common dimensions that SD 1.5 models work with.

    Positive Prompt:

    "(masterpiece, highest quality, illustration)" to produce a more 2D style, "(masterpiece, highest quality, absurd res)" for a more 3D style. Adding keywords such as "cinematic lighting, thick outlines, comics style, flat colors" can also change the style.

    Negative Prompt:

    "(worst quality, low quality:1.4), poorly drawn, bad anatomy, wrong anatomy, blurry, open mouth" usually results in good images; alternatively, start with "(worst quality, low quality:1.4, normal quality)".

    Sampler Settings:

    DPM++ 2M Karras, 20-30 Steps, Clip Skip 2, CFG 5-9, Noise Seed Delta (ETA) 31337

    VAE:

    vae-vt-mse-840000-ema-pruned or kl-f8-anime2, both produce very similar results.

    Hires Fix and Up-scaling

    10 steps, 0.5 denoise strength

    I upscale by 2 with the help of the Tiled VAE plugin; if you have 8gb or less vram, you can use that or reduce the upscale factor.
    After generation, I send images to the process tab and upscale again by 2, but quality after hires should be decent enough.

    Upscale with RealESGRAN 4x+ Anime6B to emphasize the more illustrated aspects, or ESGRAN Superscale 8x to get more detailed images.

    ADetailer

    ADetailer is a plugin that automates the in-painting process for certain features. I use the "face_yolo8vn" model with it to fix faces; I'd recommend leaving the prompts blank. I do not touch any of the other settings, nor do I use ADetailer to touch up the hands, since it usually seems to make them wose

    Model Credits

    V1/V2:

    Description

    • Replaced Occidental Mix v2.0 with v3.0

    • Adjusted one of the underlying merges

    FAQ

    Checkpoint
    SD 1.5

    Details

    Downloads
    629
    Platform
    CivitAI
    Platform Status
    Deleted
    Created
    1/14/2024
    Updated
    4/22/2026
    Deleted
    9/22/2025

    Files

    Available On (1 platform)

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