Orelon logoOrelon
Tarifs

Photorealistic Cinematic Storyboards: An AI Workflow Guide

15 nov. 2026 · Par Orelon Team

Explorez les modèles vidéo IA

Parcourez quelques créations de la communauté pour trouver l’inspiration, puis ouvrez n’importe quel modèle pour continuer à créer dans Orelon.

Learn how to build photorealistic cinematic storyboards with AI: model selection, shot consistency, prompt structure, and a fast review workflow.

A photorealistic storyboard is not a mood board. It is a working document that a director, a cinematographer, and a client can all read the same way — where the light direction is legible, the lens feels plausible, the character in panel three is unmistakably the same person as in panel one, and the framing already tells you how the cut will breathe. AI image and video models have become good enough to produce frames at that standard, but only when they are used as part of a deliberate pipeline rather than a slot machine.

The practical question is no longer whether AI can render a convincing cinematic frame. It can. The question is how to get nine of them in a row that belong to the same film. That is a workflow problem: model selection, prompt architecture, continuity management, and a review loop that keeps a sequence coherent as it grows. This guide walks through that pipeline end to end, including a worked nine-panel example.

Start with what the board has to do, not which model is best

Every storyboard serves one of three masters: selling an idea, planning a shoot, or feeding a downstream motion pipeline. Those goals demand different outputs.

A pitch board needs atmosphere and emotional clarity. It can tolerate slight continuity drift if the mood lands. A production board needs spatial accuracy — where the camera stands, which way the actor turns, what is in the background of the reverse shot. A board intended as the first frame of a generated video clip needs something else again: clean edges, a subject that will not melt under motion, and a composition with room for movement.

Before touching a prompt, write a one-page board brief. Specify the format (16:9 or 2.39:1), the lens range you are implying, the color palette in three words, the light sources available in the scene, and the list of recurring elements that must survive every panel. That brief becomes your prompt template and your quality checklist. Teams that skip it end up with nine beautiful frames that look like nine different films.

The four qualities that separate photoreal from merely pretty

Photorealism in cinematic frames is a compound effect. It is rarely about resolution. It is about the presence of four signals that a viewer's eye checks unconsciously.

Skin, lens, and light

Skin must have subsurface behavior — a slight translucency at the edges of shadow, pores that catch a rim light, not a smooth plastic gradient. Lenses must misbehave: a 35mm has some barrel distortion at the frame edges, a long lens compresses the background into a soft wall, an anamorphic produces horizontal flares and oval bokeh. Light must have a source. If a face is lit, the frame should show you where the light came from — a window, a practical lamp, a bounce card just out of frame.

Continuity across panels

A board is read as a sequence, so drift is amplified. Two panels with slightly different jawlines read as two different actors. Three panels with a subtly shifting jacket color read as a continuity error the client will spot immediately. Continuity is the single biggest differentiator between a usable AI board and a demo reel.

Composition discipline

Generative tools love to center a subject and fill the frame. Cinematic framing often does the opposite: negative space on the side the character is looking toward, a foreground element breaking the bottom edge, headroom that leaves room for a title. Prompt for framing explicitly or you will get a portfolio of centered portraits.

Motion readiness

If a board is destined to become a clip, its composition must survive movement. Limbs crossing the body, hands near the face, and complex overlapping fabric are all high-risk zones. Boards that keep silhouettes clean animate far more reliably.

Matching tools to production stages

No single model wins at everything. The strongest pipelines assign different tools to different stages, then unify the visual language in post.

Stage one: look development

This is where you explore. You want speed, variety, and a tolerant prompt parser. Broad image models are ideal here — you are testing palettes, wardrobe, and lighting setups, not final frames. Generate twenty variations of a single hero frame, pick three, and write down the exact words that produced the ones you liked. That vocabulary becomes the backbone of your sequence prompts.

Stage two: sequence boards

Once the look is locked, move to a model with strong reference conditioning — image-to-image, character reference, or style reference support. This stage is about repetition with control. You are no longer exploring; you are reproducing a decision across nine framings. Tools with storyboard or sequence features, and dedicated image generators such as the Orelon AI image generator, are built for this kind of controlled iteration.

Stage three: motion tests

For animatics, hand off to a video model. Contemporary video generators such as Runway's Gen family or OpenAI's Sora line handle short shot-level motion well; Kling and similar models are worth testing for character-driven action. Each has a distinct motion signature, so test a single shot across two or three before committing a whole sequence. If you want a starting point for comparisons, the Orelon alternatives hub lays out the tradeoffs.

A useful rule: never change two variables between stages. Lock the look, then lock the framing, then lock the motion. Changing style and camera at the same time makes failures impossible to diagnose.

A prompt structure that holds up across a whole sequence

Freeform prompting produces one good frame and no reproducibility. A six-block structure produces a template you can reuse across every panel.

  1. Shot specification — format, lens, height, and movement intent. "Anamorphic 40mm, low angle, slow push in."
  2. Subject block — age, build, wardrobe with specific materials, posture, eyeline direction.
  3. Action block — the single verb of the frame. One action, not three.
  4. Environment block — location, time of day, weather, depth cues in the background.
  5. Lighting block — key source, direction, quality (hard or soft), and the color of any practical light.
  6. Texture block — film stock feel, grain, halation, contrast curve, and any imperfections you want.

Keep the order fixed. When something breaks, you can identify which block caused it. When something works, you can lift that block into the next panel.

Lighting language that models understand

Vague words like "cinematic lighting" produce a generic look. Specific constructions work better: "single practical sodium lamp camera left, hard shadow falling across the far wall, cool ambient spill from a doorway behind." Describe the physical source, its quality, and what it does to the nearest surface. That is what a cinematographer would say on set, and it maps surprisingly well to model behavior.

Negative constraints

Be explicit about what you do not want: no text overlays, no watermark, no extra fingers, no centered composition, no wide-angle face distortion. Keep the negative list short and stable. A long, shifting negative list creates unpredictable results because each new term competes with the rest.

Consistency: the hard part

There are three continuity problems, and each has a different remedy.

Character lock

The reliable method is reference-first. Generate one clean, evenly lit character frame — full face, neutral expression, wardrobe visible — and reuse it as an image reference for every subsequent panel. Pair the reference with a short fixed description block so the model receives the identity twice: once visually, once textually. If your tool supports character or identity references, use them for the whole sequence rather than only the close-ups.

Location lock

Build a location plate: one wide frame of the space with no actors, showing the key light, the exits, and the dominant textures. Every panel set in that space references the plate. This alone eliminates the most common board failure, which is a room that quietly redecorates itself between shots.

Wardrobe and props

Describe materials, not colors alone. "Cropped raw-hem denim jacket, faded indigo, brass buttons" survives better than "blue jacket." Name the number of visible props and where they sit in the frame. If a prop matters to the story, give it a fixed position relative to the subject's body.

A practical habit: maintain a continuity sheet in a text file with the exact character block, location block, and wardrobe block. Paste from it. Never retype from memory.

Worked example: a nine-panel street dance sequence

Imagine a short film opening: a lone dancer warming up in an empty underground parking level at dawn, building to a full routine.

The board brief might read: 2.39:1, 35mm and 85mm, cool concrete tones with one warm practical, hard top light from service fixtures, recurring elements being the dancer's cropped denim jacket, a chalk mark on the pillar, and a rolling shutter door in the background.

Panel 1 — establishing wide. 24mm, static, dancer small in frame at the far end, pillars receding, shutters closed. Purpose: geography.

Panel 2 — detail. 85mm, tight on shoes and chalk mark, shallow depth. Purpose: texture and hint of ritual.

Panel 3 — medium profile. 50mm, dancer stretching, head down, warm practical behind the shoulder. Purpose: introduce the character's face.

Panel 4 — low angle. 35mm from floor level, first step of the routine, silhouette clean against a lit pillar. Purpose: energy.

Panel 5 — over-the-shoulder. 50mm, foreground shoulder soft, mid-ground movement sharp. Purpose: rhythm change.

Panel 6 — close-up. 85mm, sweat and breath visible, eyes off-camera. Purpose: emotional beat.

Panel 7 — wide re-establish. Same angle as panel 1 but the dancer now center frame and the shutter door half-open, letting in a shaft of daylight. Purpose: progression.

Panel 8 — tracking implication. 35mm, motion blur at the edges, subject sharp, background streaking. Purpose: acceleration.

Panel 9 — final wide. Back to panel 1's camera position, dancer still, dust in the light shaft, chalk mark worn. Purpose: resolution.

Note what makes this sequence work: panels 1, 7, and 9 share a camera position, so the audience feels time passing in a fixed space. Panels 2 and 6 are the only tight shots, so they carry weight. Nothing is centered for its own sake.

When prompting, panels 1, 7, and 9 use the identical location block — only the action block and the state of the shutter door change. That is how you get continuity without retouching.

Once the stills are approved, feed the strongest frames into a video model as first-frame conditioning. Eight to twelve seconds per shot is usually enough to judge whether the motion reads. A good entry point for that step is the Orelon AI video generator, and ready-made starting structures live in the templates library.

Review workflow: how to run a fast approval loop

A board sequence should be reviewed in three passes, not one.

Pass one — continuity. Spread all panels on one screen at thumbnail size. Any panel that reads as a different film gets regenerated first, regardless of how beautiful it is individually.

Pass two — story. Read the panels in order without captions. If the action is unclear without text, the framing is wrong, not the caption.

Pass three — craft. Now look at full size for lens behavior, skin, light logic, and edge artifacts.

Keep a decision log: panel number, prompt version, what changed, why. When a client asks for "the earlier one," you will have it. Version your prompts the way you version cuts.

Common mistakes that quietly ruin a board

  • Over-prompting. Twelve stylistic adjectives dilute each other. Three precise ones land.
  • Changing the seed and the prompt together. You lose the ability to attribute the result.
  • Ignoring lens logic. Every panel at 50mm feels like television, not cinema.
  • Centered subjects in every frame. Vary weight in the frame deliberately.
  • Skipping the location plate. This is the cheapest fix with the largest payoff.
  • Chasing photorealism before narrative clarity. A slightly soft frame that tells the story beats a flawless frame that does not.
  • Never testing motion early. A frame that will not animate is a dead end; test one shot before committing to nine.

FAQ

Do I need a video model to make a cinematic board? No. Stills alone can carry a pitch or a shot plan. Video becomes valuable when you need to prove pacing, camera movement, or performance timing.

How many panels should a board have? As many as the sequence needs and no more. Nine to twelve is a practical range for a short scene; a commercial may need four. Every panel should earn its place with a distinct beat.

Why do my characters keep changing between panels? Almost always because identity is being described only in text. Add an image reference and repeat the exact same wardrobe block verbatim. Consistency is a reference problem before it is a prompt problem.

Can AI boards replace hand-drawn storyboards? For photoreal pitches and client-facing presentations, often yes. For complex action choreography where you need to specify exact spatial relationships, a hybrid approach — rough hand sketches for blocking, AI frames for atmosphere — usually communicates faster.

How do I keep a consistent look across different tools? Define the look once as a written style block plus one approved reference frame, then carry both into every tool. Match final grading in post rather than fighting for identical output from each model.

What resolution should I work at? Board at whatever resolution lets you judge the frame quickly, then re-render only the approved panels at full size. Iterating at maximum resolution is the most common hidden time sink.

Where to take this next

Photorealistic cinematic boards reward pipelines, not prompts alone. Lock the brief, lock the look, lock the references, and only then let the sequence grow. The teams that move fastest are not using the most exotic model — they are using a consistent one, with a continuity sheet open beside them and a review loop that catches drift before it costs a re-render.

When you are ready to build the next sequence, start with a single hero frame in the Orelon AI image generator, push it into motion with the AI video generator, and pull proven prompt structures from the Orelon prompt library. More workflow breakdowns and technical guides are on the Orelon blog.