Original size 2902x4096

Color Archive

PROTECT STATUS: not protected
Longread translated automatically
The project is taking part in the competition

Project Concept

My goal was not simply to generate individual images with flowers, but rather to establish a recognizable visual style that could be applied across various compositions. Flowers became the primary motif of the project because they allow for experimentation with form, color, texture, and object placement within the frame. Furthermore, the narrative itself can be changed—from a bouquet in a vase to a single branch or a botanical illustration—while maintaining the overall artistic direction.

Style

Visual cues: a soft palette, expressive organic forms, a sense of traditional painting, and the texture of aged paper.

Source Images and Dataset Formation

For this project, I used works from the open collection of The Metropolitan Museum of Art (The Met).

Source Images: Vincent van Gogh — «Bouquet of Flowers in a Vase» (1890) Margareta Haverman — «A Vase of Flowers» (1716)

The museum provides images of works designated as Public Domain and Open Access for free reuse in accordance with the terms of the program, Open Access and CC0. For this project, floral still lifes and botanical images available in the museum’s collection were selected.

Source Images: Odilon Redon — «Bouquet in a Chinese Vase» (1912–1914) Henri Fantin-Latour — «Roses in a Bowl» (1883)

Resulting Image Series

The final series features eight pieces, ranging from large floral compositions to classic still lifes and botanical illustrations. Despite variations in composition and color palette, the images are unified by a focus on the plant’s form—the expressiveness of the leaves and petals, the muted tones, and the effect of traditional fine art.

«White Magnolia» «Flowering Apple Tree»

«Peonies in a Glass Vase» «Fern on Old Paper»

«Chrysanthemums in a Frame» «Contrasting Dahlia Bouquet»

«Blue Iris» «Roses in a Painted Vase»

Detailed Results Commentary

The series exhibits several distinct visual directions:

Color Palette. Creamy, beige, olive, and muted green tones are frequently used. Brighter pink, red, yellow, and blue flowers stand out against a calm background.

Texture. Many pieces show soft color transitions and a painterly style. The botanical images incorporate the effect of aged paper, faded print, or archival illustration.

Composition. Some images are close-ups of a single plant, while others feature an isolated botanical specimen or a bouquet in a vase.

Relationship to Background. Neutral and light backgrounds accentuate the plant forms, whereas a blue background lends a sense of airiness and spring space.

Integration of Painting and Botanical Illustration. The series merges decorative still lifes with more documentary representations of plants, supporting the concept of a floral archive.

Visual Analysis of Images

  1. «White Magnolia»Description: A close-up of a magnolia branch featuring large white flowers and olive-green leaves.Features: A light, creamy palette, soft shadows, delicate branches, and the texture of aged paper. The composition appears gentle and serene.2. «Blossoming Apple Tree»Description: An apple branch with white blossoms set against a soft blue sky.Features: An airy composition, pastel tones, and a diagonal arrangement of the branch. The blue background evokes a feeling of spring and freshness.3. «Peonies in a Glass Vase»Description: A bouquet of pink peonies and small blue flowers housed in a clear glass vase.Features: A combination of pink, blue, and green, soft color transitions, and the translucency of the glass. The piece resembles a classic still life painting.4. «Fern on Old Paper»Description: A botanical illustration of a fern branch with numerous small fronds.Features: A monochrome gray-brown palette, high level of detail, and the effect of aged paper. The image recalls an antique scientific illustration.5. «Chrysanthemums in a Frame»Description: A bouquet of yellow and ochre chrysanthemums in a vase, depicted within a decorative frame.Features: Warm autumn hues, a symmetrical composition, and a muted background. The frame gives the impression of a painting from a museum collection.6. «Contrasting Dahlia Bouquet»Description: A composition of large red and yellow dahlias with dense green foliage.Features: Saturated colors, a striking contrast between the petals, a dense arrangement, and a textured background resembling old paper. The image is more dramatic and decorative.7. «Blue Iris»Description: A single blue-purple iris with long green leaves set against a light background.Features: A vertical composition, a contrast between blue and yellow, and minimal subject matter. The work resembles a botanical plate focused intently on the form of a single plant.8. «Roses in a Painted Vase»Description: A bouquet of soft pink roses in a light ceramic vase adorned with floral patterns.Features: A pastel color scheme, soft shading, decorative painting, and a neutral background. The composition merges floral painting with decorative art.

Technical Implementation

-Base Model: Stable Diffusion XL Base 1.0.
-Training Method: DreamBooth LoRA
-Image Resolution During Training: 384 × 384 pixels
-Training Steps: 120
-LoRA Adapter Rank: 4

To access the learned style, a specific trigger word, «floralarchive,» was employed. This was included in the text prompts during the final series generation. As a result of the training, a dedicated LoRA adapter was created, which can be connected to the base SDXL model and used to generate new images.

Generative Model Application Overview

  1. Stable Diffusion XL Base 1.0
    Link:https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0
    Purpose: a foundational generative model for creating images from text descriptions. It served as the base for training and subsequent generation of floral compositions.2. DreamBooth + LoRA
    Training Source Code: https://colab.research.google.com/github/shwars/ai-for-creatives/blob/main/6-DiffusionTraining/SDXL_DreamBooth_LoRA_Colab.ipynb
    Purpose: to adapt Stable Diffusion XL to a specific theme—vintage floral painting. The LoRA method allows for training supplementary model parameters without retraining all of the model’s weights. As a result, an adapter was created that connects to the base model when generating new images.
Color Archive
Project created at 24.10.2026