simpletunner-full-novaskin

This is a full rank finetune derived from stabilityai/stable-diffusion-3.5-medium.

No validation prompt was used during training.

None

Validation settings

• CFG: 5.0
• CFG Rescale: 0.0
• Steps: 20
• Sampler: FlowMatchEulerDiscreteScheduler
• Seed: 42
• Resolution: 64x64
• Skip-layer guidance:

Note: The validation settings are not necessarily the same as the training settings.

You can find some example images in the following gallery:

Prompt

unconditional (blank prompt)

Negative Prompt

blurry, cropped, ugly

Prompt

minecraft skin, police officer with dark blue uniform: crisp button‐up shirt tucked into matching pants; sturdy black boots covering lower legs; black tactical gloves on hands; silver badge on left chest; face with focused expression, blue eyes under a peaked cap; short brown hair peeking at nape; utility belt with radio pouches around waist

Negative Prompt

blurry, cropped, ugly

Prompt

minecraft skin, fantasy dragon‐scale armor in deep red and gold: overlapping scale plates on chest and shoulders; gauntleted gloves with claw tips on hands; armored greaves on legs; sturdy boots with scale fins; face framed by horned helm opening to reveal green reptilian eyes; dark flowing hair braided into a tail at back

Negative Prompt

blurry, cropped, ugly

Prompt

minecraft skin, cyberpunk streetwear: black bio‐leather jacket with neon blue circuit patterns, high‐collar around neck; fingerless gloves on hands; slim cargo pants with glowing seams and knee pads; tech‐bonded boots; face with glowing augmented ocular implants and angular cheek tattoos; buzzed undercut hair in electric purple

Negative Prompt

blurry, cropped, ugly

Prompt

minecraft skin, samurai warrior in red and black lacquered armor: layered cuirass over kimono sleeves; wrapped hand‐guards on forearms; hakama trousers tying at calf; straw‐soled sandals on feet; face calm with golden eyes beneath a half‐mask; jet‐black topknot hairstyle

Negative Prompt

blurry, cropped, ugly

Prompt

minecraft skin, elven forest scout: green hooded cloak draping over slender shoulders; leather bracers on forearms; fitted tunic and leggings dyed in moss and bark tones; soft boots for silent steps; delicate hands with finger‐woven gloves; face with sharp emerald eyes and freckled cheeks; long auburn hair braided with leaf ornaments

Negative Prompt

blurry, cropped, ugly

Prompt

minecraft skin, steampunk mechanic: worn brown leather apron over soot‐stained shirt; fingerless leather gloves with metal knuckle plates on hands; reinforced trousers with tool pockets; heavy leather boots with brass buckles; face smudged with oil, bright hazel eyes behind round goggles; tousled sandy hair

Negative Prompt

blurry, cropped, ugly

Prompt

minecraft skin, high‐tech spacesuit: white and blue pressurized suit with panel lines on torso; glove‐sealed sleeves and articulated gauntlets on hands; reinforced leggings with cable conduits to boots; helmet viewport revealing calm face with hazel eyes and chin strap; short dark hair neatly cut

Negative Prompt

blurry, cropped, ugly

Prompt

minecraft skin, desert robes: flowing sand‐colored tunic over loose pants; wrapped cloth around forearms and calves; sand‐proof gauntlets on hands; leather sandals on feet; face partially veiled with brown scarf exposing only bright amber eyes; sun‐bleached blonde hair tied back

Negative Prompt

blurry, cropped, ugly

Prompt

minecraft skin, shining plate armor: breastplate embossed with crest over padded gambeson; articulated gauntlets on hands; greaves and sabatons covering legs and feet; face visible through open helm, with steely gray eyes and a cropped brown beard; short cropped hair

Negative Prompt

blurry, cropped, ugly

Prompt

minecraft skin, arcane mage robes: deep violet robe embroidered with glowing runes on torso and sleeves; delicate fingerless silk gloves on hands; flowing skirt over fitted leggings; soft pointed boots; face with luminous violet eyes and pale skin; long silver hair cascading over shoulders

Negative Prompt

blurry, cropped, ugly

The text encoder was not trained. You may reuse the base model text encoder for inference.

Training settings

• Training epochs: 5
• Training steps: 60260
• Learning rate: 1e-06
  • Learning rate schedule: cosine
  • Warmup steps: 200
• Max grad value: 1.0
• Effective batch size: 12
  • Micro-batch size: 12
  • Gradient accumulation steps: 1
  • Number of GPUs: 1
• Gradient checkpointing: False
• Prediction type: flow_matching (extra parameters=['shift=3'])
• Optimizer: adamw_bf16
• Trainable parameter precision: Pure BF16
• Base model precision: no_change
• Caption dropout probability: 0.1%

Datasets

skins-64

• Repeats: 0
• Total number of images: 100000
• Total number of aspect buckets: 1
• Resolution: 0.004096 megapixels
• Cropped: False
• Crop style: None
• Crop aspect: None
• Used for regularisation data: No

skins-256

• Repeats: 1
• Total number of images: 20000
• Total number of aspect buckets: 1
• Resolution: 256 px
• Cropped: False
• Crop style: None
• Crop aspect: None
• Used for regularisation data: No

skins-heads

• Repeats: 3
• Total number of images: 5000
• Total number of aspect buckets: 1
• Resolution: 256 px
• Cropped: False
• Crop style: None
• Crop aspect: None
• Used for regularisation data: No

Inference

import torch
from diffusers import DiffusionPipeline

model_id = 'saviski/simpletunner-full-novaskin'
pipeline = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.bfloat16) # loading directly in bf16

prompt = "An astronaut is riding a horse through the jungles of Thailand."
negative_prompt = 'blurry, cropped, ugly'

pipeline.to('cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu') # the pipeline is already in its target precision level
model_output = pipeline(
    prompt=prompt,
    negative_prompt=negative_prompt,
    num_inference_steps=20,
    generator=torch.Generator(device='cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu').manual_seed(42),
    width=64,
    height=64,
    guidance_scale=5.0,
).images[0]

model_output.save("output.png", format="PNG")

Exponential Moving Average (EMA)

SimpleTuner generates a safetensors variant of the EMA weights and a pt file.

The safetensors file is intended to be used for inference, and the pt file is for continuing finetuning.

The EMA model may provide a more well-rounded result, but typically will feel undertrained compared to the full model as it is a running decayed average of the model weights.