DayK0n/AIGI-styles

Midjourney Art Style Separation & Translation Dataset

This dataset consists of 192 high-quality images generated using Midjourney, carefully clustered into 12 distinct, style-separated categories. It includes 40 text prompt files corresponding to key images, detailing the exact prompts used to achieve each style's signature aesthetic.

The primary purpose of this dataset is to facilitate style translation and prompt engineering. It serves as a visual and textual training set for AI agents, generators, and LLMs to learn how to construct 50–200 line highly detailed, comprehensive, high-fidelity prompts for non-Midjourney text-to-image models (specifically Ideogram 4.0 or Stable Diffusion) to replicate Midjourney's visual style.

---

Dataset Structure

The dataset is organized as a folder-structured image collection. Each subfolder represents a distinct visual style cluster:

styles/
├── 3d_pixar_unreal/
├── chinese_ink_art/
├── cinematic_photography/
├── clean_vector_pop/
├── conceptual_neon_collage/
├── direct_flash_dusk/
├── flat_nordic_illustration/
├── grotesque_comic_sketch/
├── impasto_oil_watercolor/
├── lofi_anime_aesthetic/
├── surreal_glitch_dreamscapes/
└── vintage_gothic_lace/

File Counts by Style Cluster

Folder Name	Visual Vibe	Images	Prompts	Total Files
3d_pixar_unreal	3D clay-like smooth character rendering, cozy lighting.	1	1	2
chinese_ink_art	Traditional Chinese wash paintings, Song Dynasty scrolls.	5	1	6
cinematic_photography	Volumetric lighting, high fashion campaigns, photorealism.	41	4	45
clean_vector_pop	Sharp vector contours, minimalist fashion, pop-art blocks.	11	3	14
conceptual_neon_collage	Cyberpunk grids, thermal imaging, brutalist portals, poster collages.	18	5	23
direct_flash_dusk	CCD sensor grain, raw on-camera flash, suburban landscapes.	3	1	4
flat_nordic_illustration	Scandinavian folk art, geometric layouts, ditsy florals.	13	2	15
grotesque_comic_sketch	Absurd cartoon designs, pulp horror, rough sketches, coloring books.	14	3	17
impasto_oil_watercolor	Heavy palette knife oil textures, acrylics, loose watercolors.	21	4	25
lofi_anime_aesthetic	90s lofi anime stills, rainy city windows, soft pastel summer tones.	7	0	7
surreal_glitch_dreamscapes	SimAIty starry nights, dissolving glitter, magical dream corridors.	28	4	32
vintage_gothic_lace	Shabby lace repeating prints, Victorian Gothic dark velvets.	8	2	10
Total		192	30	222

---

Prompt Construction Guidelines (for Prompt Generators)

To translate these styles into non-Midjourney models, prompts should be expanded to 50–200 lines covering these 7 structural zones:

1. Core Subject & Pose: Full anatomy, posture, clothing weave, expression, and placement details.
2. Environment & Background: Physical scene details (catwalks, server grids, organic structures), depth layers, and negative space.
3. Lighting & Shadows: Volumetric shafts, directional lights (cyan monitor glow, amber rim), and contrast.
4. Color Palette: Dominant color harmony (analogous/triadic) and saturation.
5. Texture & Tactile Qualities: Ballistic nylon, cold-rolled steel, paper grain, paint ridges, or skin pores.
6. Camera & Render Setup: Lens parameters (e.g. 100mm f/2.2), camera body models, film grain simulation, or Octane/Unreal Engine render parameters.
7. Art Style Technicalities: Outlines (slender/bold), brush strokes (loose wash/heavy palette), and movements.

---

Licensing Information

This dataset is distributed under the Computational Use of Data Agreement (C-UDA-1.0) license.

• Restricted to Computational Use: The data is shared specifically for machine learning training, fine-tuning, computational analysis, and AI prompt engineering.
• Freedom of Results: Downstream models, generated prompts, and visual outputs (results) resulting from computational analysis of this dataset are not restricted by this license.
• No Warranty: The dataset is provided on an "as is" basis, without warranty of any kind.