CivArchive
    PC-98 Gal - v0.2
    NSFW
    Preview 134957591
    Preview 134957602
    Preview 134957582
    Preview 134957603
    Preview 134957581
    Preview 134957584
    Preview 134957587
    Preview 134957589
    Preview 134957596
    Preview 134957594
    Preview 134957599
    Preview 134957601

    PC-98 Galgame Style (Anima) v0.2

    Trained words: pc98gal_style

    Base model: Anima v1.0


    Description

    A LoRA trained to reproduce the aesthetic of NEC PC-98 galgame artwork from the early-to-mid 1990s. Captures the signature look of era-defining visual novel and eroge titles — flat cel shading, thin linework, muted limited color palettes, detailed retro anime eyes with characteristic highlights, atmospheric painted backgrounds, and the 16-color era palette feel.

    Trained on 87 images sourced from 17 authentic PC-98 titles including Words Worth, Nocturnal Illusion, Doukyusei 2, Yuno, Viper GTS, Magical Jam, Dokidoki Vacation, Thirsty, Sand Rock, Saint Dairy, Harenochi, Poison Needle, Ram, Tsuioku, and others, covering a wide range of character designs, poses, and settings.

    Training Details

    • Epochs: 10

    • Total steps: 2610

    • Final loss: 0.123

    • LoRA rank: 32 (alpha 32)

    • Resolution: 768×768

    • Training time: ~2h17m on Strix Halo (Radeon 8060S, ROCm 7.13)

    • LoRA strength: 0.7–1.0

    • Steps: 25–30

    • CFG: 4.0–5.0

    • Sampler: er-sde, scheduler: simple

    Usage

    Place pc98gal_style at the start of your prompt, followed by your subject description.

    Positive prompt base

    pc98gal_style, 1990s anime aesthetic, limited color palette, flat cel shading, thin linework, detailed retro eyes, dithering, muted colors, [your subject here]
    

    Negative prompt

    3d, realistic, photorealistic, modern anime style, thick outlines, gradient mesh shading, oversaturated, blurry, deformed, extra digits, watermark, text, signature
    

    Example prompts

    Character portrait (1:1, 768×768):

    pc98gal_style, young woman with long red hair, green eyes, sailor uniform, sailor collar, red ribbon, school rooftop background, cherry blossoms, soft smile, looking at viewer, upper body portrait
    

    Negative: modern, 3d, realistic, photograph, blurry, low quality, deformed

    Full body schoolgirl (2:3, 512×768):

    pc98gal_style, schoolgirl with short blue hair, twin tails, blue sailor dress, white socks, brown loafers, standing pose, hands clasped, classroom interior, wooden desks, chalkboard, soft afternoon light, full body
    

    Negative: modern, 3d, realistic, photograph, blurry, low quality

    Wide scene with multiple characters (16:9, 1024×576):

    pc98gal_style, two girls sitting on a bench under cherry blossom trees, petals falling, one with pink hair one with black hair, school uniforms, blue sky, clouds, school building in background, peaceful afternoon, wide shot
    

    Negative: modern, 3d, realistic, photograph, blurry, low quality

    Battle scene with magic effects:

    pc98gal_style, dynamic battle scene, young female mage with long teal hair in high ponytail, white and gold magical armor, wielding a glowing blue staff casting lightning spell, energy arcs and sparks around her, dark dramatic sky background, motion blur, action pose mid-cast, fantasy RPG battle scene
    

    Negative: modern, 3d render, realistic, photograph, blurry, low quality, deformed, text, watermark

    Mecha pilot in hangar:

    pc98gal_style, mecha anime hangar launch scene, female pilot with short silver hair in orange flight suit climbing into cockpit of giant humanoid battle mech, mech is 80 meter tall white and blue armored robot with glowing green visor and massive shoulder-mounted rail cannons, steam venting from hydraulic systems, hangar bay opening to reveal orange sunset sky, ground crew in yellow helmets waving her off, 80s and 90s mecha anime aesthetic, Macross inspired
    

    Negative: modern, 3d render, realistic, photograph, blurry, low quality, deformed, text, watermark, gunpla

    Tips

    • Trigger token must be the first word in your positive prompt

    • Higher LoRA strength (0.9-1.0) preserves the PC-98 aesthetic more strongly; lower (0.6-0.8) gives more freedom to your subject

    • The model works best with descriptive natural-language prompts rather than tag-style prompts

    • PC-98 linework is directional — do not use img2img with horizontal flips

    • For best results, pair with a sampler that respects the era’s flat cel shading (er-sde or similar)

    Notes

    This is a personal project to preserve and reimagine the distinctive aesthetic of NEC PC-98 era visual novels. The model captures both the technical limitations of the platform (16-color palette, dithering, scanlines) and the artistic sensibilities of the period. Still actively iterating — v0.3 will include more titles and refined training parameters.

    Description

    Training config

    • Base model: Anima base-v1.0 (DiT + Qwen3-0.6B + flow matching)

    • LoRA rank: 32, alpha 32

    • Optimizer: AdamW, lr 5e-5, weight_decay 0.1, betas [0.9, 0.99]

    • Scheduler: cosine_with_restarts, 100 warmup steps

    • Training: 87 images × 3 repeats × 10 epochs = 2610 steps, batch size 1

    • Precision: bf16 throughout, gradient checkpointing on

    • Resolution: 768×768

    • Caption dropout: 10%

    • No flip augmentation — PC-98 linework is directional

    • Duration: ~2h17m on Radeon 8060S (Strix Halo, gfx1151, ROCm 7.13.99)

    FAQ

    LORA
    Anima

    Details

    Downloads
    31
    Platform
    CivitAI
    Platform Status
    Available
    Created
    6/26/2026
    Updated
    6/27/2026
    Deleted
    -

    Files

    pc98gal_style-v02_anima.safetensors

    Mirrors