Impasto, Impressionism, Chiaroscuro - Painterly Styles

This LoRA is a study of styles suggestive or characteristic of a painting or of the art of painting.

It focuses on concepts of impressionism, impasto, and chiaroscuro (light-dark).

The dataset is mostly focused on modern works, with a selection of old masters and impressionists.

ℹ️ LoRA work best when applied to the base models on which they are trained. Please read the About This Version on the appropriate base models and workflow/training information.

ℹ️ For Anima Circlestone labs recommend "if going for a more realistic / painterly look, the beta57 scheduler (ComfyUI RES4LYF custom node pack) can help make better textures, since it puts more emphasis on low-noise timesteps."

Impressionism

Impressionist art shows life-like subjects painted in a broad quick style, with many short visible brushstrokes that are easily seen and colours that are often bright making up the artwork.

Works by Claude Monet, Van Gogh, Renoir, and Vermeer are included to learn this style.

Impasto

A painting technique where paint is laid on an area of the surface thickly, usually thick enough that the brush or painting-knife strokes are visible. Paint can also be mixed right on the canvas. When dry, impasto provides texture; the paint appears to be coming out of the canvas.

The technique is commonly achieved with oil or acrylic paint due to their thicker consistency.

Can be synonymous with impressionism, the difference being the depth of the paint.

Chiaroscuro

The use of strong contrasts between light and dark across an entire composition to achieve a sense of volume in modeling three-dimensional objects.

Works by Rembrandt Caravaggio, and Vermeer are included to learn this style.

Recommended tags:

# general painterly styles
painterly
brushstroke
impressionism
impasto
oil painting \(medium\)

# for light-dark
chiaroscuro
dark
candle
black theme

Recommended prompt structure:

Positive:

painterly, {{characters}}, {{copywrite}}, {{artists}},
{{tags}},
impasto, brushstroke, impressionism, oil painting \(medium\)

Negative:

error, bad anatomy, bad hands, ugly, distorted, lowres, watermark, signature, scanlines

Note on quality tags:

In most aesthetic tagging, low quality is often associated with aspects of painterly styles.

So to express these styles accurately, its recommended to use a minimal negative prompt without the tags worst quality, bad quality, low qualityor sketch.

Thanks to @epiTune for their article on this subject)

Description

[HIDREAM] LoRA

Trained with diffusion-pipe on HiDream-I1-Full
- 7,500 steps on a 4090
Previews generated with ComfyUI_examples/hidream/#hidream-dev-workflow
- Loading the LoRA with LoraLoaderModelOnly node and using the HiDream DEV fp8 model: hidream_i1_dev_fp8.safetensors

config.toml

# Dataset config file.
output_dir = '/mnt/d/hidream/training_output'
dataset = 'dataset-hidream.toml'

# Training settings
epochs = 50
micro_batch_size_per_gpu = 1
pipeline_stages = 1
gradient_accumulation_steps = 4
gradient_clipping = 1.0
warmup_steps = 100
blocks_to_swap = 20

# eval settings
eval_every_n_epochs = 5
eval_before_first_step = true
eval_micro_batch_size_per_gpu = 1
eval_gradient_accumulation_steps = 1

# misc settings
save_every_n_epochs = 10
checkpoint_every_n_minutes = 30
activation_checkpointing = true
partition_method = 'parameters'
save_dtype = 'bfloat16'
caching_batch_size = 1
steps_per_print = 1
video_clip_mode = 'single_beginning'

[model]
type = 'hidream'
diffusers_path = '../hidream-full'
llama3_path = '../llama-3.1'
llama3_4bit = true
dtype = 'bfloat16'
transformer_dtype = 'nf4'
max_llama3_sequence_length = 256
flux_shift = true

[adapter]
type = 'lora'
rank = 32
dtype = 'bfloat16'

[optimizer]
type = 'adamw_optimi'
lr = 5e-5
betas = [0.9, 0.99]
weight_decay = 0.02
eps = 1e-8

dataset.toml

# Resolution settings.
resolutions = [1024]

# Aspect ratio bucketing settings
enable_ar_bucket = true
min_ar = 0.5
max_ar = 2.0
num_ar_buckets = 7

[[directory]] # IMAGES
# Path to the directory containing images and their corresponding caption files.
path = '/mnt/d/huanvideo/training_data/images'
num_repeats = 5
resolutions = [1024]