CivArchive
    UltraReal Fine-Tune Anima - v2.0 Anima_base
    Preview 1

    Description

    --base1

    To be completely honest, I trained this model on this specific caption format while I was pretty drunk, so it is what it is lol. I'm not even entirely sure how it works under the hood, and I didn't have enough time to thoroughly test everything out. Also, please don't come at me calling me retarded or saying "it doesn't work like that" - just chill, I'm just experimenting here. Hopefully, you guys can help me with that and test.


    Anyway, here is a mini-guide on how it's supposed to work (written by Gemini):

    To achieve maximum realism, analog grit, or authentic casual smartphone aesthetics, you should follow the exact prompt structure the model was trained on.

    📐 The Prompt Formula:

    [Prefix Tags] + [Natural Language Description] + [Suffix Tags & Score]

    1. Prefix Tags (The Setup): Set the camera type, lighting, safety, and shot style at the very beginning.
    2. Core Description (The Scene): Describe the subject, clothing, pose, and background using natural English sentences (avoid messy tag-soup).
    3. Suffix Tags (The Quality & Era): Close your prompt with the simulated year of the photo and the quality score.

    📋 KEYWORDS TO COPY-PASTE

    1. Camera & Tech Prefixes:

    • @smartphone_photo — Casual, modern mobile look with subtle computational processing.
    • @compact_digital_photo — Early 2000s "point-and-shoot" digicam vibe.
    • @film_photo — Authentic analog look with rich organic textures and grain.
    • @vhs_screencap — Retro video tape style with scanlines.
    • @dslr_photo — Clean, professional camera rendering.

    2. Lighting & Style Prefixes:

    • @available_light — Soft, natural indoor/outdoor daylight.
    • @direct_flash — Harsh, flat flash (perfect for late-night party vibes or digicam looks).
    • @candid_photo — Caught-on-camera, unposed, natural moments.
    • @posed_photo — Deliberate posing.
    • @mirror_reflection — Perfect for mirror selfies.
    • @underexposed / @overexposed — For dramatic low-light or high-contrast shots.

    3. Safety Blocks:

    • @sfw or @nsfw (choose depending on your target generation)

    4. Era & Quality Suffixes (Put at the very end!):

    • Years: @1995, @2000, @2005, @2010, @2020, @2025
    • Scores: score_5, score_6, score_7, score_8, score_9 (can be used in negative)

    💡 Tip: Use score_8 or score_9 for high definition and clean details. Use score_6 or score_7 combined with @smartphone_photo or @compact_digital_photo if you want a grittier, intentionally imperfect lo-fi look!

    📸 EXAMPLE PROMPTS

    • Modern Smartphone Selfie:
    @smartphone_photo @sfw @amateur_photo @candid_photo @available_light A close-up portrait selfie of a 20-year-old woman with neon green hair and heavy eyeliner. She is looking at the camera with a neutral expression. The background is a blurry minimalist bedroom. @2025 score_8
    
    • Retro Digicam Flash (Vibe from 2005):
    @compact_digital_photo @sfw @amateur_photo @posed_photo @direct_flash An overexposed snapshot of a young woman posing in a cluttered room at night. Harsh flash lighting, red-eye effect, visible digital noise, and washed-out colors. @2005 score_7
    
    • Analog Film Portrait (Cosplay):
    @film_photo @sfw @amateur_photo @candid_photo @available_light A medium shot of a young woman cosplaying Princess Zelda, sitting inside a

    P.S.: still WIP, i plan extend dataset and train more


    Details

    Downloads
    23
    Platform
    ShakkerAI
    Platform Status
    Available
    Created
    6/7/2026
    Updated
    6/7/2026
    Deleted
    -

    Files

    ultrarealFineTune_v20AnimaBase1.safetensors