Project Topic:
Training the Stable Diffusion model (via LoRA fine-tuning) on sunflower images to generate a series of artworks in various styles—from realism to impressionism.
Concept:
The project’s goal is to demonstrate how a neural network can master the pictorial language of nature. By training the model on images of sunflowers, we aimed to create a neural network that generates new paintings imbued with the warmth, light, and character of authentic painting—featuring soft brushstrokes, the texture of petals, and the feeling of sunlight. The outcome must prove that AI can do more than simply copy an object; it can express mood and style through it.
Source images for training:
Generation Results (Final Series):
Series Description: Each image was generated using one of the random prompts describing an artistic style:
«a sunflower field at sunset, painterly style, warm tones, oil painting texture» «a close-up of a sunflower in sunlight, brush strokes, golden palette, artistic composition».
Detailed Commentary and Visual Analysis:
Analysis of Results:
The model captures soft light and the texture of the brushstrokes. The colors are natural, dominated by yellow, gold, and ochre tones. The neural network reproduces not only the form of the flower but also the overall atmosphere of the painting. In some images, the influence of Impressionism and Post-Impressionism can be felt.
Stylistic Characteristics:
An ethereal atmosphere with soft transitions. The effect of brushstrokes, creating an illusion of painting. A balance between realism and artistic style.
Conceptual Alignment:
The project achieved its goal—the model learned to create paintings that convey the mood and visual language of sunflowers. The final series possesses a «living» quality to the depiction of light, rather than digital manipulation: every piece is unique yet tied together by a common warm palette and formal rhythm.
Jupyter / Colab Notebook
https://drive.google.com/drive/folders/1JK51lJPRx4DveSq_Jd0E-lJEYPbH_kL4?usp=sharing
Using GenAI
In the project, ChatGPT (GPT-5) was additionally used for:
- Drafting descriptions and prompts for training
- Generating caption pairs
- Writing the idea description and analyzing the final images
All visual results were created independently using a trained model.



































