sd35-training

This is a standard PEFT LoRA derived from stabilityai/stable-diffusion-3.5-large.

The main validation prompt used during training was:

j5155, This is a detailed 3D-printed model of a small, bright yellow boat gently resting on the shoreline of a tropical Caribbean beach. Viewed from a slightly angled side perspective, the boat showcases its rectangular cabin, compact smokestack, gracefully curved hull, and subtle square recess near the stern. Fine layer lines from the 3D printing process are visible, giving the model an authentic feel, complemented by crisp edges and realistic minor imperfections. Soft waves of crystal-clear turquoise water gently touch the boat, creating subtle ripples along the white sandy beach. Bright sunlight illuminates the boat's vivid yellow color, harmoniously contrasting against the serene tropical background featuring lush palm trees, distant islands, and a clear blue sky dotted with delicate clouds, evoking a relaxing, inviting atmosphere

Validation settings

CFG: 7.5
CFG Rescale: 0.0
Steps: 30
Sampler: None
Seed: 42
Resolution: 1024

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

Prompt
j5155, This is a detailed 3D-printed model of a small, bright yellow boat gently resting on the shoreline of a tropical Caribbean beach. Viewed from a slightly angled side perspective, the boat showcases its rectangular cabin, compact smokestack, gracefully curved hull, and subtle square recess near the stern. Fine layer lines from the 3D printing process are visible, giving the model an authentic feel, complemented by crisp edges and realistic minor imperfections. Soft waves of crystal-clear turquoise water gently touch the boat, creating subtle ripples along the white sandy beach. Bright sunlight illuminates the boat's vivid yellow color, harmoniously contrasting against the serene tropical background featuring lush palm trees, distant islands, and a clear blue sky dotted with delicate clouds, evoking a relaxing, inviting atmosphere

Negative Prompt
blurry, cropped

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

Training settings

Training epochs: 24
Training steps: 300
Learning rate: 0.00105
Max grad norm: 0.01
Effective batch size: 6
- Micro-batch size: 6
- Gradient accumulation steps: 1
- Number of GPUs: 1
Prediction type: flow-matching
Rescaled betas zero SNR: False
Optimizer: adamw_bf16
Precision: Pure BF16
Quantised: No
Xformers: Not used
LoRA Rank: 768
LoRA Alpha: 768.0
LoRA Dropout: 0.1
LoRA initialisation style: default

Datasets

benchy

Repeats: 5
Total number of images: 10
Total number of aspect buckets: 1
Resolution: 1.0 megapixels
Cropped: False
Crop style: None
Crop aspect: None
Used for regularisation data: No

Inference

import torch
from diffusers import DiffusionPipeline

model_id = 'stabilityai/stable-diffusion-3.5-large'
adapter_id = 'AngelZeur/benchy-sd3-dreambooth'
pipeline = DiffusionPipeline.from_pretrained(model_id)
pipeline.load_lora_weights(adapter_id)

prompt = "j5155, This is a detailed 3D-printed model of a small, bright yellow boat gently resting on the shoreline of a tropical Caribbean beach. Viewed from a slightly angled side perspective, the boat showcases its rectangular cabin, compact smokestack, gracefully curved hull, and subtle square recess near the stern. Fine layer lines from the 3D printing process are visible, giving the model an authentic feel, complemented by crisp edges and realistic minor imperfections. Soft waves of crystal-clear turquoise water gently touch the boat, creating subtle ripples along the white sandy beach. Bright sunlight illuminates the boat's vivid yellow color, harmoniously contrasting against the serene tropical background featuring lush palm trees, distant islands, and a clear blue sky dotted with delicate clouds, evoking a relaxing, inviting atmosphere"
negative_prompt = 'blurry, cropped'
pipeline.to('cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu')
image = pipeline(
    prompt=prompt,
    negative_prompt=negative_prompt,
    num_inference_steps=30,
    generator=torch.Generator(device='cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu').manual_seed(1641421826),
    width=1024,
    height=1024,
    guidance_scale=7.5,
).images[0]
image.save("output.png", format="PNG")