The Workspace

Understanding StyleFusion's three-column layout: references, IR viewer, and generation controls.

Layout Overview

The Workspace is StyleFusion’s main screen. Three columns, each with a specific job:

  • Left panel: your inputs (reference images, director’s note, saved characters)
  • Center panel: the Intermediate Representation and generated images
  • Right panel: compiled prompts, generation controls, and export

The top navbar shows your four page tabs (Workspace, Characters, History, Providers), six provider status dots, and the Settings gear.

Left Panel: References and Characters

Reference Images

You can upload up to 5 reference images. Each image card shows a thumbnail, a number badge, a weight percentage, and a close button.

Two controls per image:

Reference Type tells StyleFusion what role this image plays. Options: Subject, Style, Composition, Color, Lighting, Texture. A single image defaults to general extraction, but when blending multiple images, assigning types tells the system which visual qualities to pull from which source.

Weight slider (0.1 to 1.0) controls how much influence this image has in the blend. Higher weight means more of that image’s visual DNA carries through.

With one image, StyleFusion runs a straight extraction. With two or more, it runs a blend extraction that combines visual qualities according to your type assignments and weights.

Director’s Note

A free text field (500 character limit) where you can guide the extraction. Think of it as talking to the AI: “Focus on the neon lighting and ignore the background” or “I care about the color palette, not the composition.” The note shapes what the extraction emphasizes.

Active Models

Shows which extraction model and image generation model are currently selected, with “Change” links that take you to the Providers page.

Character Cards

Up to 5 saved characters appear here as quick-select cards. Click one to activate it; its DNA merges into the IR during extraction. More on this in the Characters doc.

Center Panel: IR Viewer and Output

Before Extraction

A placeholder prompting you to drop reference images to begin.

After Extraction: The IR Viewer

The IR (Intermediate Representation) is the structured analysis of your reference images. It’s the core of StyleFusion’s pipeline: everything between your inputs and the final prompt.

The header shows “Intermediate Representation” with a version badge and three action buttons: Pin (keeps this IR version), Copy JSON (copies the raw data), and Download (exports as .txt).

Below the header: which model performed the extraction and the currently selected arrangement.

Arrangement Dropdown: defaults to “Auto (Conductor),” which lets the AI pick the best arrangement for your content. You can override this with any available arrangement. Changing the arrangement triggers a recompilation of prompts.

Style Anchors: tag chips that define the core aesthetic direction. Add, remove, or reorder them to steer the output.

The Six IR Sections

Each section is collapsible and tagged with a priority level (P0 through P5):

  1. Visual DNA: the big one. Subject form, face details, hair, pose, attire, expression, colors (primary/secondary/accent with hex values), textures, structure, features, and confidence scores.
  2. Environment: setting, surfaces, props, atmosphere.
  3. Camera: shot type, lens characteristics, focus behavior.
  4. Lighting: key light, fill light, practical lights.
  5. Palette: color list, contrast level, mood associations.
  6. Rendering: medium, detail level.

All of these fields are editable. You can adjust anything before recompiling.

Negative Prompt: a separate section with red labeling. Tag chips for exclusions (things you want the AI to avoid). Add, remove, reorder.

Blend Section (only visible with multi-source extractions): shows per-reference breakdown including type, weight, notes, classification results with confidence percentages, reasoning, aesthetic tags, and dominant color swatches.

Character DNA Section (only visible when a character is active): shows the character name, anchor traits (immutable), and flex traits (variable), all as editable chips.

Generated Image

After generation, the image appears with hover controls: Zoom (opens a fullscreen lightbox), Download (PNG or JPG), Enhance (Bria Creative Upscale, available for images under 4MP), and Regenerate.

The model name displays in the top-left corner of the image card.

Enhanced Image

If you run Bria upscale, the enhanced version appears below the original with a purple “Enhanced” label and the same hover controls.

Right Panel: Prompts and Controls

Compile Button

The purple full-width button that generates prompts from your IR. Four prompt formats are created:

  • Image JSON: structured JSON for programmatic use
  • Description: natural language prose
  • Creative Slots: ordered paragraphs breaking down each visual slot
  • Compact: concise format

Each prompt card is collapsible with copy and download buttons. Download All Prompts bundles everything into a single .txt file.

Aspect Ratio Selector

A grid of 10 ratio buttons: 9:16, 2:3, 3:4, 4:5, 1:1, 5:4, 4:3, 3:2, 16:9, 21:9. Visual preview rectangles show the proportions.

The selector auto-detects the aspect ratio from your source image and filters options to show only ratios your current image model supports. A reset button restores the auto-detected ratio if you’ve overridden it.

Normal / Lame Mode

Normal (purple when active): full creative prompts with all the expressive language and detail.

Lame (amber when active): sanitized, simplified prompts for models or APIs with stricter content policies.

Image Model Dropdown

Grouped by provider. Models without a configured API key show “(no key)” and can’t be selected.

Generate Image Button

Disabled until you have an IR, compiled prompts, and a valid model selected. Shows “Generating…” with a spinner during processing.

Upscale Controls

Visible after generating an image. Default model is Topaz CGI. Scale factor buttons: 2x, 3x, 4x (toggle). Face Enhancement checkbox (on by default). Requires a fal.ai API key.

Export Panel

Visible when an IR exists. Copy All or Download .txt. A preview toggle shows the full export text before downloading. The export includes the generative description, all four prompt types, structured metadata, and full JSON.