The Director Console

The Director is the one place in the app where you assemble the prompt directly rather than describing a subject and letting the engine translate. The left rail lists categories grouped into sections — Subject, Styling, Scene, Camera, Grade. Click a category, pick from its grid, and the choice drops into the live prompt on the right. It reads and behaves identically in every mode; only the vocabulary changes.

The prompt updates live as you click. A small dot on a category means you have a selection in it, so you can see your coverage at a glance.

Work top-to-bottom through the groups — Subject first, Grade last. It mirrors how a real shoot is planned and keeps the prompt coherent.

The toggle at the top of the console switches between AUTOMOTIVE, FASHION, CHARACTER and MECHA. Each mode loads its own set of category grids. Your selections are kept separately per mode, so switching from Fashion to Character and back never wipes what you built.

The rail, search and grids all re-populate with the chosen mode's vocabulary. Automotive is unchanged from before — it is simply now one mode among four.

Set the mode before you start clicking. If a shot spans two worlds — say a character on a motorbike — build the primary subject in its own mode and describe the rest in the main input.

FASHION mode lays out the full editorial vocabulary as grids. SUBJECT covers casting — gender & presentation, model archetype, heritage, skin tone, age and stature. STYLING is the deepest group: garment type, silhouette, fabric, texture and print; hair, hair colour, makeup and nails; the fashion house; and the accessory families — jewellery, eyewear, headwear, footwear, bags, gloves & hosiery, other accessories and styling details. SCENE holds pose, expression, set and lighting; CAMERA holds body, lens and film stock; GRADE holds palette, mood, era and photographer style.

Selecting a Fashion House biases the palette grid toward that house's signature colours. Accessories and styling details are multi-select, so you can layer several at once.

Cast the model and lock the garment first, then style outward. Leave whole groups blank if you want the model to have latitude — an unset field is freedom, not an error.

CHARACTER mode covers concept-art and portrait design. SUBJECT holds archetype, race/species, class/role, plus the same inclusive casting axes as Fashion (gender, heritage, skin tone, age, build). STYLING covers face features, hair & colour, fantasy skin, eyes, wardrobe/armour and armour material, weapons, gear, powers/aura FX, wings/appendages and a creature companion. SCENE covers environment, lighting, pose/action, expression and mood; CAMERA covers shot type and composition; GRADE covers art style, genre, artist influence, render medium and palette.

Face features, weapons, gear, powers and appendages are multi-select — stack a scar plus war-paint plus a cybernetic eye, or dual-wield. The GRADE group is where the drawing style is decided, not the main input.

Describe who the character is in the grids and let ART STYLE + ARTIST carry the look. Keep the render style out of the subject fields.

MECHA mode covers concept industrial design. SUBJECT holds chassis class, faction/manufacturer, locomotion, silhouette, scale, tech era and narrative role. STYLING is the engineering layer: hull plating, greebles, material/CMF, weathering, paint & livery, weapon systems, propulsion, energy FX, cockpit, sensors/optics and limbs/manipulators. SCENE holds environment, lighting, atmosphere FX and pose/action; CAMERA holds shot type; GRADE holds render style, artist influence and palette.

Greebles, weapon systems, energy FX and atmosphere FX are multi-select — the detail-density controls of the mode. Scale and tech era steer the whole read of the machine.

Set chassis class, scale and tech era first — they anchor everything. Then layer greebles and wear to taste; a little weathering reads as 'used-future' far better than pristine surfaces.

Most categories are single-choice, but the ones marked '· MULTI' let you select many at once. In Fashion that's the accessory families and styling details; in Character it's face features, weapons, gear, powers and appendages; in Mecha it's greebles, weapon systems, energy FX and atmosphere FX. Click to add, click again to remove.

Every selected chip stays highlighted and all of them enter the prompt, comma-joined.

Multi-select is where density lives. Two or three layered choices add richness; ten turn the prompt to noise.

A few categories are linked. When a linked field is filtered, its grid shows the options that fit your earlier choice — for example hair and makeup respond to the gender presentation, and palette responds to the chosen fashion house. If nothing upstream is set, the full grid is shown.

Filtering is assistive, never restrictive — if a link would leave you with nothing sensible, the full list is shown instead.

Set the upstream field first (gender, house) to get a tighter, more coherent set of downstream options.

Fashion and Character both expose casting as explicit, deliberate axes — gender & presentation, a broad heritage spectrum, a full skin-tone range and age. They exist so you can cast intentionally and represent a wide range of people, rather than leaving it to a model's defaults.

These pair freely with any archetype, garment or beauty choice; nothing is gated.

Cast on purpose. Setting heritage and skin tone explicitly gives more consistent, more representative results than hoping the model chooses well.

When you have Knowledge Base tags active, the AUTO-SUGGEST button fills any empty categories with tag-aware picks, working through the grids in order so earlier choices influence later ones. Fields you've already set are left untouched.

It's a fast way to get a complete, coherent starting point you can then refine by hand.

Set the two or three choices you care about, then Auto-Suggest the rest and edit from there.

Randomize and Auto-Suggest don't draw blindly from every pool. A coherence layer reads what you've already committed to — mode, gender and presentation, archetype, era, domain — and narrows each pool to options that can physically and contextually coexist with it. Options that contradict your current selection are visibly dimmed in the grid rather than removed, so you can still see the full vocabulary and override deliberately.

Rolling the dice repeatedly gives varied but plausible results instead of contradictions you then have to clean up. Change a governing field — gender, mode, archetype — and the dimming re-evaluates immediately across every downstream group. Muted SEQUENCE groups are skipped by the roll entirely.

Set the two or three fields that define the shot — mode, casting, archetype — before you randomize. Coherence works from what you've committed, so the more you anchor first, the more usable every roll becomes.