How to Use Shot Lab: A Step-by-Step Guide to the Cinematography Workbench

What Shot Lab actually is

Shot Lab is the cinematography workbench inside Studio, and the fastest way to describe it is this: it is the room you walk into before you generate a video, not after. Most people open a text-to-video box, type a paragraph, get something mediocre, and then try to guess which words were wrong. Shot Lab removes the guessing by giving you the vocabulary, the structural templates, and a score before you spend a single generation.

It has two working parts on one screen. The left side is your prompt editor with a live character counter. The right side is an assist panel with four tabs: the Shot Dictionary (81 cinematography terms), Shot Remix (6 complete production templates), the Template Library (16 templates across 8 categories, each with optional director style packs), and the Prompt Wizard (a 5-stage guided builder). Above the editor sits a prompt scorer that grades what you have written out of 100 and names the ingredients you are missing.

Four tabs sounds like a lot, but they are not four ways of doing the same thing — they are four levels of control, and the right one depends on how much you already know. If you have nothing but a vague idea, start in the Wizard. If you know the shot you want and just need a proven skeleton, take one from the Library. If you want to understand why the good prompts work, read the Dictionary. And if you want to study a real, full-length production prompt end to end, go to Shot Remix.

The whole point is that you finish in Shot Lab with a prompt that is already structured, already scored, and already portable. From there, one button pushes it into the video generator with the text intact.

Before you start: what you need

You do not need an account to browse. Shot Lab lives at /studio/shot-lab and the page is public, so you can open the dictionary, pick a template, run the scorer, and copy the result without signing in. The only step that needs a session is the last one — pressing Generate hands you to the video generator, and that generator requires you to be signed in and to have credits available. So you can do all of the learning and all of the writing for free, and only pay when you actually render.

Two practical notes. The Remix tab and the Wizard tab both save drafts automatically to your browser, so you can fill in a character description, close the tab, and come back later to find your character, scene, and event fields restored. And the workbench is desktop-shaped: it is a two-column layout with a fixed-width assist panel, so it is usable on a tablet but genuinely designed for a laptop or larger screen.

Step 1 — Open the workbench and orient yourself

Go to /studio/shot-lab. You land on a two-column layout. On the left is a large prompt textarea with a character counter underneath, currently at 0 / 8,000. On the right is the assist panel, which opens on the Shot Dictionary tab by default. Two shortcut buttons at the top of the page take you to the video generator and to Storyboard Studio if you would rather start from a different tool.

That 8,000-character limit is worth internalizing early, because it is the same limit the video generator enforces. Anything you build here will fit there. It is also much more room than you think — most of the built-in templates land between 3,000 and 8,000 characters once assembled, and only one of them is long enough to be trimmed.

One thing that is easy to miss: the character counter also shows a small "CINEMATIC MODE" badge once your prompt passes 1,000 characters. It is cosmetic, not a setting — nothing changes except the label. Do not go hunting for a toggle.

Step 2 — Learn the vocabulary in the Shot Dictionary

The Shot Dictionary tab holds 81 terms across nine categories: Shot Size (10), Camera Move (19), Camera Angle (9), Optics (17), Lighting (12), Composition (7), Timing (4), Style / Grade (2), and Sound (1). Each term shows its English name, its Chinese name, a one-line definition, and a usage count drawn from a real production corpus. Sorting is by usage count by default, which is a genuinely useful signal — the terms at the top are the ones that show up most often in prompts that worked. Handheld sits at the very top, followed by SFX Only, Close-Up, and Silhouette.

The search box matches against the English term, the Chinese term, the definition, the pro-tip note, and the term ID, so you can search in either language. The category chips below it let you narrow to just one family, which is how most people actually work: "I know I need a camera move, show me the moves."

Click any term card and a detail panel slides over the list. It gives you the fuller definition, an optional pro-tip (marked with a lightning bolt when the term has a real-world caveat attached — 17 of the 81 terms have one), and a snippet showing the term used inside an actual prompt from the corpus. That snippet is the most valuable part of the whole dictionary, because it shows you the syntax, not just the word. Reading "contre-jour backlight, practicals only" teaches you more than reading a definition of contre-jour ever will.

When a term is the one you want, press the insert button at the bottom of the detail panel. The term is appended to your prompt in the editor on the left and a toast confirms it. If you are only exploring, just close the panel — nothing is modified.

One caveat about the dictionary in non-Chinese locales: the definitions are authored in Chinese, and the English UI substitutes the corpus snippet's first line as the card description instead. That works well as a working example but it is not a translated definition. If you read Chinese, switch the interface language to 中文 for the full definitions and all 17 pro-tips; in English the pro-tips are hidden entirely.

Step 3 — Start from a template instead of a blank page

The second tab, Shot Remix, is where most people should actually begin. It ships with six complete, production-grade shot templates, each one a fully written prompt with named structural blocks rather than a loose description. They are not suggestions — they are working prompts that have been used to generate real footage.

The six templates cover the situations that come up most often. The Emotional Dialogue Shot is a 4-second single-camera, single-line setup built around an anchor phrase, dialogue lock, and a 60:30:10 color grade. The Armor-Shatter Action Shot is a beat-by-beat breakdown of a first clash between two armored fighters, with a critical positioning block and a shot-by-shot camera plan. The FPV One-Take is a 15-second uncut drone flight through seven painted locations, chasing three glowing artifacts. The Chinese Voiceover Narrative template mixes English camera terminology with Chinese storytelling, locking one line of dialogue per close-up. The Environment Establishing Shot is a 14-second, two-shot slow dolly fly-through for setting up a space. And the Atmosphere Narrative template is a detailed single-shot lighting and continuity setup for an interior scene.

Click a template chip to select it. The description underneath tells you what structural problem that template solves, and which parts you are expected to swap out.

Step 4 — Fill in your character, scene, and event

Once a template is selected, Simple mode shows three fields. Character Description is where you write your subject in obsessive detail: age, height, hair, skin, distinguishing marks, and every garment. Scene Description is the space itself — layout, light sources, color temperature, weather, atmosphere. Shot Event is a single line describing what actually happens in this shot, like "she recognizes the killer and slowly sets down her coffee cup."

Here is the part that matters more than anything else in this tutorial: verbatim repetition prevents identity drift. If you describe your character's mole, her mole stays put across the shot. If you describe it once and then refer to her loosely, the model reinvents her face. This is exactly why the templates front-load a REFERENCES block with explicit character and scene entries, and why the placeholder text in the character field nags you about "every garment — verbatim repetition prevents face drift."

The event field is optional but punches above its weight. It is added to the assembled prompt as its own labelled YOUR SHOT EVENT block, sitting just above the template body. That is what turns a static composition into a moment with a beginning and an end.

If you want to see the machinery, switch to Advanced mode with the toggle in the header. It shows the fully assembled prompt in a read-only panel: your character and scene references first, then any template reference blocks, then the template body with all the negative constraints still in place. Nothing is hidden, which is useful for learning how the blocks fit together.

Step 5 — Apply the template and edit like a director

Press "Apply to Prompt" and the assembled text drops into the editor on the left. This is the moment Shot Lab stops being a template library and becomes your workspace. From here you are editing a real prompt, and the templates are designed to be edited rather than used verbatim.

One thing to watch for: the longest remix template is the Armor-Shatter Action Shot, and it assembles to roughly 14,400 characters — well past the 8,000-character field. When that happens Shot Lab trims the tail and tells you how many characters it dropped, rather than silently discarding them. Be aware of what you lose in that case, because it is a lot. What survives is the reference blocks and the scene description; what falls outside the limit is the entire critical-positioning block, the full beat-by-beat timing list, the horde-density note, the dialogue block, the key rules, and the closing format line. Roughly 6,400 characters — about 44% of the template — is gone. If you want the whole thing, use the Copy button in the Remix tab rather than Apply, and paste it somewhere with more room.

A few editing habits that pay off. Keep the lock and negative-constraint blocks intact — the Emotional Dialogue template, for example, carries an explicit HARD LOCKS list that pins the character count in frame, the camera side, and the identity details, and that is exactly the kind of block preventing clones, drift, and CGI-looking output. It is the least obvious and most valuable part of any template. Adjust the specifics, not the skeleton: swap the lens, change the light source, rewrite the dialogue line. And if you are working in Chinese, note that the templates deliberately keep camera and technical terms in English while letting narrative and dialogue stay in Chinese, because that hybrid consistently produces more reliable camera behavior.

Step 6 — Read the prompt score

Above the editor is a suggestion panel that grades your prompt out of 100 and shows a progress bar. It is checking for six things: lighting vocabulary, camera movement, camera angle, shot size, quality specification, and lens or film style. Lighting and camera movement are worth 20 points each, angle and shot size 15 each, quality and style 10 each, with a small bonus for length.

When something is missing, it does not just tell you — it gives you a one-click fix. Missing lighting offers "contre-jour backlight, practicals only." Missing camera movement offers "handheld 35mm, slow dolly-in." Missing angle offers "low angle." Missing shot size offers "medium close-up." Click and the phrase is appended to your prompt. These are deliberately concrete examples rather than generic advice, so you learn the pattern while you fix the score.

The panel also flags softer problems: prompts under 200 characters get an "add more detail" tip, prompts without a quality token get told to specify 8K or 4K, long prompts without any style language get nudged toward a lens or film stock, and anything over 6,000 characters gets a trim warning. When your score reaches 90 or above, the panel disappears entirely — that is your signal that the prompt is in good shape.

One quirk worth knowing: the scorer reads the text, not the template. A template that carries lighting language in prose but never uses one of the scorer's lighting keywords will still be marked as missing lighting, so a strong template can sit at 65 until you add an explicit phrase. Treat the score as a checklist, not a verdict on quality.

Step 7 — Hand the prompt to the video generator

When the score looks good, press "Generate video with this prompt." Your text is passed to the video generator and the browser navigates you there, with the prompt already filled in. The handoff is one-shot by design: it is consumed on arrival and cleared immediately, so refreshing the generator page gives you a clean form rather than re-filling the old prompt.

If you would rather work outside the app, "Copy prompt" puts the whole thing on your clipboard with a confirmation toast. That is the path for pasting into Storyboard Studio, a notes file, or a different model entirely. Because the templates are plain structured text with no proprietary syntax, they are fully portable.

Step 8 — The Template Library, when you want a proven skeleton

The Template Library is the third tab, and it sits between Shot Remix and the Wizard in terms of control. Where Shot Remix gives you six long, opinionated production prompts, the Library gives you 16 shorter templates spread evenly across eight categories — Cinematic, Short Drama, Action, UGC / Social, Ads & Commercial, Anime, FPV / Fly-through, and Environment — with two templates each.

Each template has placeholder slots rather than free-form fields: most carry character, scene, and event, while the ad, FPV, and environment templates drop the character slot entirely because there is nobody in frame. Fill a slot and it is substituted into the body; leave it empty and a bracketed guidance string is used instead, so the prompt is always coherent even when you have not finished writing.

What makes the Library different is the Director Style Pack row. There are 10 packs — IMAX 70mm, Wong Kar-wai 90s, Deakins / Villeneuve, Lubezki natural light, Anamorphic 2.39:1, Claymation 12fps, Hand-painted oils, Found footage CCTV, VHS 1996, and FPV drone — and any number of them can be toggled on. Selected packs append their suffix text to the end of the prompt, which is a fast way to change the look without touching the structure. Some packs also carry scorer keywords, so switching one on can visibly move your score.

The Library is also the safest place to experiment with length. Every one of the 16 templates, with all 10 style packs enabled at once, still lands under 7,100 characters — comfortably inside the 8,000 limit. Nothing in this tab will ever be trimmed.

Step 9 — The Prompt Wizard, when you are starting from nothing

The Wizard is the fourth tab and the one to use when you have an idea but no vocabulary. It is a five-stage builder that assembles a prompt top to bottom: Stage 1 sets the core theme with style tags and a target model, Stage 2 takes your character and scene, Stage 3 adds atmosphere and quality, Stage 4 sets the camera rules, and Stage 5 takes your storyboard beats.

Stages 1, 3, and 4 are chip pickers rather than blank text boxes. Stage 1 gives you eight style presets like "neo-noir rain city" or "documentary realism" and four target models — seedance, kling, sora, and runway — and picking a model reveals a "Model Gotcha" note about that model's particular behaviour. Stage 3 offers six atmosphere presets and a free-text notes field. Stage 4 offers eight camera presets and a single-shot versus multi-shot toggle, which changes the sentence the builder writes into the CAMERA RULES section.

Stage 5 is where the Wizard earns its place. You add beats, and each one is labelled with a five-second block — [00-05s], [05-10s], and so on — so you are forced to think about what happens when, not just what happens. That is the same discipline the timecoded production templates use, handed to you as a form.

The assembled output is a clean sectioned prompt: CORE THEME, CHARACTER, SCENE, ATMOSPHERE & QUALITY, CAMERA RULES, and STORYBOARD, separated by rules. The Wizard also saves its draft automatically, so a half-finished build survives a page reload.

A complete worked example

Here is the whole flow end to end, using the Environment Establishing Shot template in Shot Remix. Select the template. In the character field, write nothing — this is an empty scene with no subject, which is fine for this template. In the scene field, write something specific: "abandoned subway platform at 3am, single flickering fluorescent tube overhead, wet concrete floor with standing water, tiled walls with missing sections, cold 4500K overcast spill from a stairwell entrance." Leave the event line blank.

Switch to Advanced mode to confirm the assembled prompt reads correctly, then apply it. The prompt is about 5,200 characters and the scorer will put it at 65 — the template has plenty of camera language but no explicit lighting keyword, and no shot-size keyword either, so the panel will flag both. Click the lighting fix and the shot-size fix and the score jumps to 100, at which point the panel hides itself.

Then press Generate. The important thing to notice is what you did not have to do. You did not have to remember that "slow continuous straight-line camera dolly fly-through" is the phrase that produces a stable establishing move, or that you need negative constraints against shake, roll, pan, and zoom to keep it clean. The template already knew that. Your job was only the part that requires taste: the scene description, and two clicks on the score panel.

Common mistakes worth avoiding

The first is treating the dictionary as a glossary to read rather than a toolbox to click. The terms only help once they are in your prompt — reading 81 definitions and generating a vague paragraph teaches you very little. Open the dictionary while you write, insert as you go.

The second is deleting the constraint blocks because they look like clutter. Those dense negative lists are the difference between a shot that looks filmed and one that looks rendered.

The third is describing a character once and then referring to them loosely for the rest of the prompt. Consistency in AI video comes from literal repetition, not from elegant writing. The templates repeat their character blocks on purpose.

The fourth is ignoring the score and generating anyway. A 45 is not a stylistic choice; it is a prompt that is missing four of the six things that control framing. Fix the score first, then spend the credits.

The fifth is assuming the score is the whole story. It is a keyword checklist, and it can be satisfied by a prompt that is still badly written. Passing 90 means you have named the six ingredients, not that the shot is good.

Where this fits in the bigger workflow

Shot Lab handles one shot extremely well. When you are ready to build a sequence rather than a single clip, Storyboard Studio takes over. The division of labor is deliberate: learn and perfect the single shot in Shot Lab, then scale to a full sequence in Storyboard Studio. Note that the two do not share state automatically — the Generate button hands your prompt to the video generator, not to the storyboard, so if you want a Shot Lab prompt inside a multi-scene board, use Copy prompt and paste it in.

That is the real argument for starting here. A single well-structured shot teaches you the grammar of camera language, and that grammar is what makes everything you generate afterward faster to fix and easier to direct.

Quick reference

The workbench lives at /studio/shot-lab. The dictionary has 81 terms in 9 categories, sorted by real-world usage. Shot Remix has 6 templates covering dialogue, action, FPV, Chinese voiceover, environment, and atmosphere. The Template Library has 16 templates in 8 categories with 10 director style packs. The Wizard has 5 stages and saves drafts. The scorer checks lighting, camera, angle, shot size, quality, and style, and hides itself at 90+. The prompt limit is 8,000 characters, matching the video generator. Drafts in the Remix and Wizard tabs save automatically to your browser, and the handoff to the generator is single-use.