Style DNA

Extract the visual DNA from any image. Generate new images -- or transfer styles to video -- that match that aesthetic.

Style DNA Extraction	Styled Image Generation
Image Style Transfer	Video Style Transfer

Video Style Transfer Examples

Birds — Pointillism Style

Woman — Van Gogh Style

How It Works

1. Analyze -- Upload any image. Llama 4 Maverick vision extracts a structured aesthetic profile covering six dimensions: color grading, lighting, texture, composition, contrast, and atmosphere.

2. Store -- Each profile is saved to your Style Library with a unique ID, source image reference, and auto-generated style tag.

3. Generate -- Select one or more styles from your library. The system composes a detailed prompt from the combined aesthetics, then FLUX.2 Pro generates a new image. Optionally provide a scene description or reference image.

4. Transfer (Image) -- Apply any saved style to a new image. The source content is preserved while the aesthetic is fully transformed.

5. Transfer (Video) -- Two modes:

Standard -- Transfers style to sampled frames, interpolates the gaps, applies temporal smoothing
Fast (Optical Flow) -- Transfers only a few keyframes via FLUX.2 Pro, then propagates the style to all frames using OpenCV Farneback optical flow. 3-5x faster with smoother temporal coherence.

The Aesthetic Profile

Every image is decomposed into six aesthetic dimensions:

Dimension	What It Captures	Example Values
Color Grading	Dominant colors (hex), palette type, temperature, saturation	`#FFFF00, #0097A7` / warm / vivid
Lighting	Direction, quality, contrast ratio, mood	frontal / diffused / high / energetic
Texture	Grain, surface quality, sharpness	none / smooth / sharp
Composition	Technique, depth layers, framing	dynamic posing / shallow / close-up
Contrast	Dynamic range, shadow depth, highlight character	compressed / lifted / clipped
Atmosphere	Mood, emotional tone, genre	heroic / confident / superhero

Full Example: "Comic Book Hero" Profile (JSON)

{
  "style_tag": "Comic Book Hero",
  "color_grading": {
    "dominant_colors": ["#FFFF00", "#0097A7", "#FF0000", "#000000"],
    "palette_type": "mixed",
    "color_temperature": "warm",
    "saturation": "vivid"
  },
  "lighting": {
    "direction": "frontal",
    "quality": "diffused",
    "contrast_ratio": "high",
    "mood": "energetic"
  },
  "texture": {
    "grain": "none",
    "surface_quality": "smooth digital illustration with clean lines",
    "sharpness": "sharp"
  },
  "composition": {
    "technique": "dynamic posing with exaggerated proportions",
    "depth_layers": "shallow",
    "framing": "close-up hero shot"
  },
  "contrast": {
    "dynamic_range": "compressed",
    "shadow_depth": "lifted",
    "highlight_character": "clipped with bold highlights"
  },
  "atmosphere": {
    "mood": "heroic and action-packed",
    "emotional_tone": "confident and powerful",
    "genre": "superhero comic book"
  }
}

Architecture

Image Analysis & Generation Pipeline

flowchart LR
    subgraph Input
        A[/"Upload Image"/]
    end

    subgraph Analysis ["Aesthetic Analysis"]
        B["Llama 4 Maverick\n(Vision)"]
        C[("Aesthetic\nProfile")]
    end

    subgraph Library ["Style Library"]
        D[(JSON +\nSQLite)]
    end

    subgraph Generation ["Image Generation"]
        E["Llama 4 Maverick\n(Prompt Composer)"]
        F["FLUX.2 Pro\n(Image Gen)"]
    end

    subgraph Output
        G[/"Generated Image"/]
    end

    A -->|base64| B
    B -->|structured JSON| C
    C -->|store| D
    D -->|select 1-5 styles| E
    E -->|prompt +\nreference images| F
    F --> G

    style B fill:#FF6B35,color:#fff
    style E fill:#FF6B35,color:#fff
    style F fill:#8B5CF6,color:#fff
    style C fill:#10B981,color:#fff
    style D fill:#3B82F6,color:#fff

Fast Video Transfer (Optical Flow)

flowchart TB
    subgraph Input
        V[/"Upload Video"/]
    end

    subgraph Extract ["Frame Extraction"]
        EX["FFmpeg\nExtract frames"]
        AU["Extract\naudio"]
    end

    subgraph Flow ["Optical Flow"]
        OF["OpenCV Farneback\npairwise flows"]
        KF["Select evenly-spaced\nkeyframes"]
    end

    subgraph Transfer ["Keyframe Transfer"]
        PR["Compose style\nprompt"]
        FX["FLUX.2 Pro\ntransfer keyframes\n(parallel workers)"]
    end

    subgraph Propagate ["Style Propagation"]
        WP["Warp styled keyframes\nvia accumulated flow"]
        BL["Bidirectional blend\nbetween segments"]
    end

    subgraph Reassemble
        RE["FFmpeg reassemble\n+ audio sync"]
        OUT[/"Styled Video"/]
    end

    V --> EX & AU
    EX --> OF & KF
    KF --> FX
    PR --> FX
    OF --> WP
    FX --> WP
    WP --> BL
    BL --> RE
    AU --> RE
    RE --> OUT

    style FX fill:#8B5CF6,color:#fff
    style PR fill:#FF6B35,color:#fff
    style OF fill:#3B82F6,color:#fff
    style WP fill:#10B981,color:#fff

Getting Started

Prerequisites

Python 3.12+
Together AI API key (free tier available)
FFmpeg (for video transfer features)
Flutter 3.10+ (for the frontend, optional)

Docker (Recommended)

git clone https://github.com/soulfir/style-dna.git
cd style-dna/aesthetic-style-builder
cp .env.example .env
# Edit .env and add your Together API key
docker compose up --build

The API will be available at http://localhost:8080.

Manual Setup

git clone https://github.com/soulfir/style-dna.git
cd style-dna/aesthetic-style-builder
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
# Edit .env and add your Together API key
uvicorn main:app --reload

The API will be available at http://localhost:8000.

Environment Variables

Variable	Required	Default	Description
`TOGETHER_API_KEY`	Yes	--	Your Together AI API key
`VISION_MODEL`	No	`meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8`	Vision model for analysis
`CHAT_MODEL`	No	`meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8`	Chat model for prompt composition
`IMAGE_MODEL`	No	`black-forest-labs/FLUX.2-pro`	Image generation model
`MAX_VIDEO_DURATION`	No	`60`	Max video duration in seconds

Note

The free Together AI tier includes sufficient credits to experiment with analysis and generation. Video transfer uses more credits due to multiple FLUX.2 Pro calls per keyframe.

API Reference

Method	Endpoint	Description
`POST`	`/analyze`	Extract aesthetic profile from an image
`POST`	`/create`	Generate image from 1-5 style profiles
`POST`	`/transfer`	Transfer a style to a single image
`POST`	`/transfer/video`	Video style transfer (frame-by-frame)
`POST`	`/transfer/video/fast`	Fast video transfer (optical flow)
`GET`	`/transfer/video/jobs`	List all video transfer jobs
`GET`	`/transfer/video/{job_id}`	Get video job status
`GET`	`/styles`	List saved style profiles
`GET`	`/styles/{id}`	Get a specific style profile
`DELETE`	`/styles/{id}`	Delete a style profile

Tip

Interactive API docs are available at /docs (Swagger UI) when the server is running.

POST /analyze -- Extract aesthetic DNA from an image