∇ AnimeBoysNabla ∇

Creating models is a labor of love, but it takes a significant amount of time and compute power to get them just right. If you’re enjoying my models, consider fueling my next project with a coffee on Ko-fi ☕. Thank you for keeping this project going!

Introducing the NoobAI-based, versatile powerhouse of my anime boys model series. Perfect for creators who demand variety and precision in their husbando designs!

ComfyUI workflow

🚀 Inference Guide

• ⚠️ Important: This model uses Zero Terminal SNR with V-prediction. Please ensure you are using the correct settings during inference.
  • ComfyUI Users: Add the ModelSamplingDiscrete node into your workflow. Set sampling to v_prediction, zsnr to true.
  • Automatic1111 Users: Place the .yaml config file into the model folder. The .yaml file must have the exact same name as the model file, only with the .yaml extension instead of .safetensors. Set Noise schedule for sampling in settings to Zero Terminal SNR.
• Prompting: Always begin your prompt with a score tag (e.g. score_9). You can use any of these styles:
  • Tag soup: score_X, tag1, tag2, tag3, ...
  • Natural language: score_X, [your description here]
  • Mixed approach: score_X, [description], tag1, tag2, ...
  • Tip: If the score tags have too much influence on the style, try lowering the weight (e.g., (score_9:0.5)) or removing them entirely.
• Negative Prompt: Choose from one of these two presets depending on your needs:
  1. Light: score_1, lowres, artistic error, scan artifacts, jpeg artifacts, multiple views, too many watermarks, negative space, blank page
  2. Heavy: score_1, score_2, score_3, lowres, artistic error, film grain, scan artifacts, jpeg artifacts, chromatic aberration, dithering, halftone, screentones, multiple views, logo, too many watermarks, negative space, blank page
• VAE: Use the built-in VAE. This model uses KBlueLeaf/EQ-SDXL-VAE.
• CFG Scale: A CFG scale of 3 to 5 is recommended. For finer control, I suggest using dynamic thresholding.
  • Pro-tip: I use Half Cosine Up for both modes. Set separate_feature_channels to disable, scaling_startpoint to ZERO, and variability_measure to STD.
• Resolution: To get started, try these dimensions:
  • Portrait: 832 × 1216
  • Square: 1024 × 1024
  • Landscape: 1216 × 832
  • Some other supported sizes: 768×1344, 768×1280, 896×1152, 960×1088, 1344×768, 1280×768, 1152×896, 1088×960.

🧨 Diffusers Example Usage

import torch
from diffusers import DiffusionPipeline

pipe = DiffusionPipeline.from_pretrained(
    "Koolchh/AnimeBoysNabla",
    torch_dtype=torch.float16,
    use_safetensors=True,
    custom_pipeline="lpw_stable_diffusion_xl",
    variant="fp16"
)
pipe.to("cuda")

prompt = "score_9, 1boy, male focus, shirt, solo, looking at viewer, smile, black hair, brown eyes, short hair"
negative_prompt = "score_1, lowres, artistic error, scan artifacts, jpeg artifacts, multiple views, too many watermarks, negative space, blank page"

image = pipe(
    prompt=prompt, 
    negative_prompt=negative_prompt, 
    width=1024,
    height=1024,
    guidance_scale=5,
    num_inference_steps=28
).images[0]

🧪 Training Details

AnimeBoysNabla was fine-tuned from NoobAI V-Pred 1.0 using approximately 950k images. The knowledge cutoff is November 2025.

The following tags were used during training to help you steer the results toward your desired style.

Score tags

• Each image is tagged with score_X, where X is a range from 1 to 9.
  • score_9 represents the highest aesthetic quality based on my personal preferences.

Rating tags

Year tags

Use year YYYY (ranging from 2005 to 2025) to target specific era styles.

Training configurations

• Hardware: 4 × Nvidia A100 SXM 80GB
• Optimizer: AdamW 8-bit (Weight Decay: 0.1)
• Gradient Accumulation Steps: 8
• Effective Batch Size: 128 (4 × 8 × 4)
• Learning Rates:
  • U-Net: 2e-5
  • Text Encoders: 4e-6
• LR Schedule: Cosine with 1% minimal LR and 2,000 warmup steps
• Precision: BF16 Mixed Precision

🔄 Changes from AnimeBoysZeroXL

• Base Model: Updated to NoobAI V-Pred 1.0.
• VAE: Switched to KBlueLeaf/EQ-SDXL-VAE.
• Dataset Balancing: Reduced repeats for high-score images.
• Learning Rate: Lowered Text Encoder LR and migrated to a Cosine LR scheduler.
• Optimizer: Transitioned to AdamW 8-bit with 0.1 weight decay.
• Precision: Adopted BF16 mixed-precision training.
• Dropout: Increased full caption dropout to 10%.

License

AnimeBoysNabla is a derivative model of NoobAI V-Pred 1.0 by Laxhar Lab. Please read their license before using the model.