Higgsfield Cinema

Higgsfield Cinema Studio 2.5

QUICK FACTS

Generated-checked block (scripts/build_index.py verifies anchors). Read the linked sections for full context — these lines are routing aids, not the rules themselves.

• Three Cinema Studio versions coexist (2.5 / 3.0 / 3.5) — user-selected, no auto-routing; always detect the version first →
• Hard 512-character prompt cap; 2.5 @ Element chips eat ~80–100 hidden chars each →
• Elements System: define @Characters/@Locations/@Props once, call everywhere →
• 3.5 main UI = three pills (Genre / Style / Camera), each Auto by default — override only with a creative reason →
• 7 confirmed 3.5 genres: General · Action · Horror · Comedy · Noir · Drama · Epic — never invent genre names →
• Style presets: 8 Color Palettes · 6 Lighting · 9 Camera Moveset Styles, or free-form Manual Style →
• Camera axes: 3 bodies · 5 lenses · focal 8/14/35/50/75mm · aperture f/1.4–f/4–f/11; two camera vocabularies coexist — vocabulary follows the selected model →
• Output enums: 7 aspect ratios incl. 21:9 · 480p/720p/1080p · duration 4–15s · Sound On/Off →
• Lead every delivered shot with the one-line settings strip (UI presets ≠ prompt text; never restate strip values in the body) →
• The UI shot counter caps internal cuts — "strictly N shots" must keep N within it (observed cap 4; API exposes no max — UI behavior, verify live) →
• Manual Style = saved ≤2,000-char block of project LAWS (grade, lighting law, texture, performance register); it replaces the preset axes →
• 480p drafts validate the prompt, NOT the take — no seed param; transfer a look via Hero Frame + start/end-frame pinning →
• Resolution matrix has two axes: shot physics × delivery context; 4K-finish pipelines master at model max res in std mode →
• Seedance fast mode cannot output 1080p — drafting in fast then "switching up" silently changes mode AND res →

Cinema Studio is Higgsfield's professional filmmaking environment — a full production workflow for multi-shot, character-consistent cinematic content. It's fundamentally different from single-clip generation: you're building sequences, not individual videos.

Cinema Studio vs Standard Generation

Use Cinema Studio when:

• You need 2+ shots that must feel like the same film
• Character geometry must be locked across cuts
• You want professional optical physics (lens flare, depth of field, sensor grain) — 2.5 only
• You're building a short film, branded content, or any sequence longer than one clip

Stick with standard generation when:

• Single clip is sufficient
• Speed matters more than consistency
• Exploring ideas before committing to a full sequence

⚠ Version Detection — Ask First

Before generating any Cinema Studio output, always ask the user:

Are you working in Cinema Studio 2.5 or Cinema Studio 3.0?

If the user has already stated their version (e.g., "I'm using 3.0" or "Cinema Studio 3.0"), remember it and don't ask again. But never assume — 2.5 and 3.0 have fundamentally different feature sets.

Why this matters:

• 2.5 has optical physics (camera body + lens stack), color grading, 3D Mode, grid generation
• 3.0 has none of those — outputting them wastes the user's time and causes confusion
• 3.0 has features 2.5 doesn't: native audio, Smart shot control, 21:9 ultrawide, 15s duration
• Speed Ramp options differ between versions
• Genre lists differ between versions

Once the version is known, use ONLY that version's output format and feature set. Never mix features from one version into output for the other.

The 10-Step Cinema Studio 2.5 Workflow

Cinema Studio 2.5 extends the pipeline in both directions: pre-production (Soul Cast + location prompt) before generation, and post-production (color grading) after.

 1. SCRIPT        → Write or paste your scene description / shot list
 2. SOUL CAST     → (New in 2.5) Generate AI actors from parameters or use saved Elements
 3. REFERENCE     → Upload character photo → create Reference Anchor (or use Soul Cast actor)
 4. ELEMENTS      → (Optional) Define @Characters, @Locations, @Props if needed
 5. OPTICAL STACK → Select camera body + lens + focal length + aperture (image mode)
 6. HERO FRAME    → Generate a key image that defines the visual tone
 7. COLOR GRADE   → (New in 2.5) Apply color grading to keyframes before video generation
 8. CAMERA CONFIG → Set Director Panel movement + Speed Ramp + Duration in UI
 9. SHOT MODE     → Choose Single Shot / Multi-Shot Auto / Multi-Shot Manual
10. GENERATE → EXPORT → Chain into timeline or export to editing

Elements System — Define Once, Call Everywhere

Elements are Cinema Studio's reusable asset library. Create a Character, Location, or Prop once and reference it with @ in any subsequent prompt.

Three element types:

Creation workflow:

1. Open Elements panel → New Element → choose type
2. Upload reference image(s)
3. Name the element (this becomes the @ tag)
4. Add description (appearance details, key features)
5. Save → now available across all shots in this project

Per-Character Emotion: In Multi-Shot mode, each character can have an emotion setting per scene. The available emotions are:

Set emotion per character per scene in the UI — this keeps expression changes out of the prompt field and lets the model handle the subtlety of facial animation.

@ tag rule — use exactly what the user provides, nothing more Only use @ tags for Elements the user has explicitly given you in this conversation. If they give you @Marcus but no location or prop tags, write the location and props as plain description. Never invent or assume @ tags for anything the user hasn't named. Each @ tag the user gives you = they have that Element set up. Silence = they don't.

No Elements provided — write everything as description:

A woman with dark hair and a red coat walks through a narrow downtown alley at night.
She carries a worn leather briefcase. She stops under a streetlight, turns to camera.

User provides @Sarah only — use it, describe the rest:

@Sarah walks through a narrow downtown alley at night.
She carries a worn leather briefcase. She stops under a streetlight, turns to camera.

User provides @Sarah, @DowntownAlley, @LeatherBriefcase — use all three:

@Sarah walks through @DowntownAlley, carrying @LeatherBriefcase.
She stops under a streetlight, turns to camera.

Key rule: Match exactly what the user gives you. No more, no less. Tags they give you = Element exists. Anything else = write it as description.

Element Library Surface — Source Tabs and Categories

The @ symbol opens a unified asset picker. The picker is organized as a two-dimensional structure: source tabs across the top (where the asset comes from), and element categories within the Elements tab (how project assets are grouped).

Source tabs (5):

Element categories within the Elements tab (6):

Build the element library before generation. For a project with recurring characters, locations, or props — a short film, a series of branded videos, a multi-shot sequence — establish each recurring asset as an Element first, then reference it via @ in every shot. This is the platform-native implementation of the Reference-Based Prompt Mode discipline already documented elsewhere: references with assigned roles, not references as inspiration. See ../higgsfield-seedance/SKILL.md for the Reference-Based Prompt Mode discipline, and the existing "@ tag rule — use exactly what the user provides" guidance earlier in this section for how to use @ tags inside the prompt itself.

Cross-shot continuity tip. For continuation work, the Image Generations and Video Generations source tabs let you reference prior outputs from the same project — useful when a Character Element doesn't capture a specific look or pose that emerged from a particular generation. Pull the exact frame back in via the Generations tab rather than re-prompting from scratch.

Soul Cast — AI Actor Generation

Soul Cast is Cinema Studio 2.5's character generation system — create AI actors from parameters instead of uploading photos. This is fundamentally different from Soul ID.

Soul Cast vs Soul ID

Soul Cast Parameter Categories (8 total)

Key Features

• Add up to 3 Soul Cast characters per keyframe — choose from saved Elements or generate on the spot
• "Save to elements" button to reuse a specific Soul Cast actor across projects
• Every actor auto-generates a backstory + character sheet (personality traits, motivation, fear, flaw, strength)
• Designed to eliminate the "plastic/waxy" AI look — excels at skin textures and emotions
• Powered by the Nano Banana 2 model under the hood

Soul Cast Workflow

1. Open Cinema Studio → Navigate to Soul Cast panel
2. Set Genre + Era + Budget to establish the visual world
3. Select Archetype + Identity + Physical Appearance
4. Add Details (imperfections) + Outfit
5. Generate → Review backstory + character sheet
6. Save to Elements → Now available as @CharacterName across all shots
7. Repeat for additional characters (up to 3 per keyframe)

Built-in Color Grading Suite

Cinema Studio 2.5 adds a post-production color grading suite applied to keyframes before video generation — enabling unified visual cohesion across all clips.

Controls

Workflow

1. Generate your keyframe image (Hero Frame or grid selection)
2. Click the keyframe → "Colorgrade" button
3. Adjust settings to taste
4. Apply — grade is baked into the keyframe before video generation
5. Repeat for each keyframe to maintain visual cohesion across the sequence

Tip: Grade your Hero Frame first, then match subsequent keyframes to it. This is the post-production equivalent of a DIT (Digital Imaging Technician) on set.

Optical Physics Engine

Cinema Studio's Image Mode gives you a full camera body + lens stack. These are the exact names from the UI (previous versions had wrong names — these are corrected).

Claude's job when building a Cinema Studio image prompt: Always recommend a specific optical stack — body + lens + focal length + aperture — with a one-line reason why. Never leave the optical stack blank or say "choose based on preference." Make a call.

Optical Stack Recommendations by Intent

Use the user's creative intent to select the stack. The most important signal is what the image needs to feel like — not just the genre.

Portrait / Character Focus

Scene / Environment / Establishing

Emotion / Tone-Driven

Commercial / Functional

How Claude Should Deliver the Output

Cinema Studio has two separate inputs:

• UI dropdowns — Camera body, Lens, Focal length, Aperture, Genre, Director Panel, Speed Ramp
• Prompt field — Scene description only. No camera/lens/aperture text goes here.

Claude must always output these as two clearly separated blocks:

Block 1 — UI Settings (select these in Higgsfield before pasting the prompt)

Camera:   Full-Frame Cine Digital
Lens:     Warm Cinema Prime
Focal:    50mm
Aperture: f/1.4
Genre:    Intimate
↳ Why: Flattest, most natural face rendering. Creamy background separation
       keeps all attention on the character without any lens distortion.

Block 2 — Prompt (paste this into the Cinema Studio prompt field)

[Scene description only — no camera, lens, aperture, or genre language here]

Rule: Never put camera body names, lens names, focal lengths, apertures, or genre names inside the prompt text. They live in the UI, not the prompt field.

Camera Body Reference

Lens Reference

Focal Length Reference

8mm · 14mm · 35mm · 50mm

• 8mm — Ultra-wide, immersive distortion. Action POV, environment, disorientation
• 14mm — Wide, environmental, scale. Establishing shots, landscape, architecture
• 35mm — Natural human field of view. Documentary, street, two-shots, candid
• 50mm — Classic portrait compression. Character close-ups, product, single subject

Aperture Reference

• f/1.4 — Shallow depth of field. Subject sharp, background creamy bokeh. Intimacy, focus, emotion
• f/4 — Balanced. Subject sharp, background slightly soft. Most versatile, natural look
• f/11 — Deep depth of field. Everything sharp front to back. Product, landscape, architecture

Director Panel — 18 Camera Movements

⚠ Use ONLY these exact movement names in Cinema Studio output. General cinematic terms like "Crane Down", "Whip Pan", "FPV Drone", "Crash Zoom" etc. (from vocab.md) are valid for standard video generation and image prompts outside Cinema Studio, but they are NOT Director Panel options. Inside Cinema Studio, map to the closest equivalent (e.g., Crane Down → Jib Down, Crane Up → Jib Up).

All movements available in Cinema Studio's Director Panel:

Speed Ramp

Controls temporal feel of the shot. Set per-scene in Cinema Studio.

⚠ Speed Ramp options differ between versions. Use ONLY the table for the version the user specified. Never output a 2.5 ramp name in 3.0 output or vice versa.

Cinema Studio 2.5 Speed Ramps (6 modes)

Custom curve: Blue line with draggable nodes. Pull up = slow down, pull down = speed up. Left = beginning of clip, right = end.

2.5-only values — NEVER use in 3.0 output: Linear, Slow Mo, Speed Up, Impact, Custom

Cinema Studio 3.0 Speed Ramps (7 modes)

3.0-only values — NEVER use in 2.5 output: Ramp Up, Flash In, Flash Out, Bullet Time, Hero Moment

Genre Selection

⚠ Genre lists differ between versions. Use ONLY the genres for the version the user specified.

Cinema Studio 2.5 Genres (8)

General · Action · Horror · Comedy · Western · Suspense · Intimate · Spectacle

Genre affects lighting defaults, color grading baseline, and motion style suggestions. You can always override with explicit prompt language — genre is a starting point.

2.5-only genres — NEVER use in 3.0 output: Western, Suspense, Intimate, Spectacle

Cinema Studio 3.0 Genres (7)

General · Action · Horror · Comedy · Noir · Drama · Epic

3.0-only genres — NEVER use in 2.5 output: Noir, Drama, Epic

Shot Modes

Single Shot

Standard generation — one prompt, one clip. Same as regular Higgsfield but with the optical stack and Director Panel applied.

Multi-Shot Auto

Describe the full sequence in one prompt. Cinema Studio breaks it into shots automatically. Good for: rough sequences, exploration, when you don't need per-shot control.

Example: "A woman walks into a bar, sits down, orders a drink. Bartender slides it over.
She takes a sip and stares at her reflection in the mirror behind the bottles."
→ Cinema Studio Auto generates: Establishing shot / Walk to seat / Order / Drink delivery / Reflection closeup

Multi-Shot Manual

Full per-scene control. Up to 6 scenes, each with its own:

• Prompt
• Camera movement (Director Panel)
• Speed Ramp
• Duration

⚠ 12-Second Total Runtime Cap: The combined duration of all scenes in a Multi-Shot sequence cannot exceed 12 seconds total. Plan your per-scene durations to stay within this limit (e.g., 6 scenes × 2s each, or 4 scenes × 3s, or 3 scenes × 4s).

Cost transparency: Multi-Shot Manual with 4 variations = 24 generations total (6 scenes × 4 variations). Plan credits accordingly.

Multi-Shot Manual workflow:

1. Scene 1: Establishing shot — wide, sets location
2. Scene 2: Character introduction — medium, shows protagonist
3. Scene 3: Action beat — closer, movement
4. Scene 4: Reaction — close-up, emotion
5. Scene 5: Consequence / turn — medium or wide
6. Scene 6: Resolution / button — varies

Morph-Cut and Smooth-Cut breathing room — 2-second still moments. Editors land morph cuts and smooth cuts (transition styles in DaVinci Resolve, Premiere, and similar editors that hide the cut by morphing similar frames) on stillness, not on motion. If every shot starts and ends mid-action, every cut between them has to happen mid-action too — and the model renders mid-action transitions poorly. Build cut-friendly material by prompting an explicit 2-second still-or-near-still moment at the start AND end of every shot. Examples of buildable stillness: a held look, a settled pose, a held pause before motion begins, a held breath after motion stops. The middle of the shot can be as kinetic as the script demands; the boundaries earn the cuts. This rule is adapted from the Mr. Core methodology and pairs with editor-side workflow in ../higgsfield-pipeline/SKILL.md § Stage 8 — Assembly.

Reference Anchor System

The Reference Anchor locks character geometry (facial structure, proportions, costume silhouette) across all shots in a sequence — preventing the character drift that happens in standard generation.

Setup:

1. Upload a clear, well-lit reference photo of your character
2. Cinema Studio generates an anchor from it
3. Every subsequent shot in the project references this anchor
4. Combine with Soul ID for full face + geometry lock

Best practices:

• Use a front-facing, neutral expression photo for the anchor
• Avoid glasses, hats, or accessories in the anchor image if they won't always be present
• One Reference Anchor per character — add multiple if you have a cast
• The anchor persists across the whole Cinema Studio project

Location Reference Sheets

Locations deserve the same asset-first treatment as characters. If a room, alley, vehicle interior, or exterior reappears across multiple shots, generate it once as a standalone asset and reuse it — don't re-describe it inside each scene prompt. Scene-by-scene re-description is the root cause of the "every shot reinterprets the environment" failure: walls drift, furniture rearranges, scale shifts, and the world stops feeling like a single place.

Plant visual anchors before you shoot the space

[FIELD — Higgsfield Studio, ADILIADA breakdown, 2026-08-14] A location sheet stops the room being reinvented; it does not by itself stop people moving around inside it. The cheap addition is to deliberately plant a few fixed, distinctive objects and then stage against them: the chair a character sits in, the window two of them talk by, the beaded curtain a body goes through. They give the model something nameable to hold position against across generations, and they give you something to write blocking against — "at the window, screen-left" is checkable in a way that "on the left side of the room" is not.

Choose anchors that are distinctive and immovable. A second matching chair is not an anchor, it is an ambiguity.

Tune the plates against each other before generating anything. Locations that will cut together get their colour, light and saturation matched at the asset stage, so the scenes already agree in character once shots start landing — otherwise every generation inherits a slightly different world and the mismatch surfaces only in the edit, where it is ../higgsfield-pipeline/SKILL.md § The Edit's unification problem to solve at colour-correction instead.

The Five-View Location Sheet

A location sheet captures the space from enough angles that the model has a complete spatial model to draw from. Generate these five views as one asset set:

Use Grid Generation (2×2 or 4×4) or 3D Mode on a hero wide to produce the alt angles from a single seed — this keeps the light and material identity consistent across views, which is the whole point.

What to Preserve on Reuse

When you bring the location back for a new shot, lock these four properties: architecture (walls, openings, scale), light quality (source direction, color temperature, hardness), color treatment (palette, grade), and key environmental details (signage, damage, specific props that make this place THIS place). Keep these out of the scene prompt — they live in the location asset. The scene prompt should describe only what changes: character action, camera behavior, atmospheric motion, time-of-day shift if any.

Failure Mode This Prevents

Without a location sheet, a six-scene sequence set in "a rain-soaked alley" will render six different alleys. Each scene prompt re-interprets "rain-soaked alley" from scratch. With a location sheet tagged as an @ Element or loaded as a reference, all six scenes share one alley — the puddles are in the same places, the neon sign says the same thing, the brick wall has the same pattern.

For the camera vocabulary that pairs with location reveals (Pan, Crane, Dolly Reveal), see ../../vocab.md.

Reference Sheet Types — Beyond Characters and Locations

Reference sheets aren't only for faces and rooms. The underlying principle is narrower than "character or location" — a reference sheet is any asset built once to lock a specific property of the production. Each sheet type fixes a different axis of the work: identity, architecture, motion behavior, wardrobe, palette, or product. If a property has to read consistently across multiple shots, give it its own sheet. If it only matters for one shot, write it inline and move on.

When a sheet pays for itself: if a property will appear in three or more shots, the sheet is cheaper than re-describing it every time and dramatically more consistent than relying on prompt text alone. Single-shot details — a one-time prop, a background extra, a one-off lighting cue — don't earn a sheet.

The character sheet (see ../higgsfield-soul/SKILL.md → Character Sheet Creation) and the Location Reference Sheet above already cover identity and architecture. The three sheet types below extend the same pattern to the rest of the production:

Motion / Camera Sheet

A short reference clip — 3–10 seconds is plenty — that captures a camera path or motion rhythm you want repeated across multiple shots. The sheet locks the movement signature, not the content of any single shot. Treat it as a project-wide style anchor pulled into a generation as an @Video reference, the way a DP would carry a "look" across an entire film.

Use cases: a music video where the same dolly-in arc punctuates every chorus; a fight sequence where a signature whip-pan recurs at every climax beat; a brand piece where every product shot starts on the same slow orbit. Distinct from a Kling 3.0 Motion Control reference clip — that drives a single generation's motion transfer. The Motion / Camera Sheet is reusable across the whole project.

Outfit / Material Sheet

Dedicated wardrobe reference. Locks fabric texture, color values, garment fit, fastenings, and how the material moves and folds. Generate it once with even lighting and multiple angles, then reuse it any time the costume needs to read identically across cuts.

Use cases: a hero costume that appears across 12 scenes; a specific jacket that must look the same in close-up and wide shot; a uniform that recurs across an ensemble cast. The outfit sheet is not the same as a character sheet — the character sheet locks identity (face, build, hair, distinguishing marks), while the outfit sheet locks what the character is wearing AND how that wardrobe behaves in motion. Build both when you have a hero costume on a hero character.

The piano test — wardrobe complexity has a generation cost. Every detail of wardrobe complexity — buttons, ties, jewelry, scarves, layered garments, intricate prints, fastening hardware — costs the model rendering budget. The piano test (adapted from the Mr. Core methodology) is the rule: if a wardrobe element is as visually demanding for the model to render as a piano in the frame, the model will spend its budget rendering that element instead of the action. Strip the wardrobe to the simplest silhouette that still reads as the character. A trench coat reads as a trench coat without the buttons rendering correctly. A uniform reads as a uniform without the rank insignia drifting across frames. Design the costume's signature — the silhouette and one identifying feature — and let the rest go simple. Hero costumes that must render under close-up scrutiny earn their complexity; background costumes do not.

Palette / Mood Sheet

A color and tonal anchor for the project. Lock the visual mood across an entire sequence: shadow density, key/fill ratio, saturation level, signature accent colors, grade direction. Use a single curated still or a small grid of grade references — the goal is to give every subsequent generation a consistent palette to reach for.

Use cases: a noir piece where every scene shares the same shadow density and warm-amber-on-cool-blue grade; a cyberpunk sequ

…(truncated)