Why Dutch Prompting Is a Distinct Skill
Most generative video models learn the world through English-language captions, shot descriptions, and screenplay fragments. A Dutch prompt therefore travels through a quiet translation layer before it ever becomes pixels. Sometimes that layer is harmless. More often it flattens detail: faces soften, framing drifts toward generic mid-shots, and motion loses the specific direction you imagined.
The productive response is not to abandon Dutch and write in English. It is to write Dutch that is structurally unambiguous — short clauses, concrete nouns, explicit camera instructions, and a vocabulary you reuse consistently across an entire project. Consistency is what makes a model treat your description as a stable style rather than a new idea every time.
This guide is a practical, repeatable system for writing Dutch video prompts that produce controlled, cinematic output. It covers prompt architecture, Dutch-specific language pitfalls, camera vocabulary, character consistency, motion direction, a full workflow, and a troubleshooting map for when a render refuses to cooperate.
The Six-Block Anatomy of a Dutch Video Prompt
Long prompts are not automatically better. Models respond well to structure: predictable slots filled with specific content. A reliable Dutch video prompt has six blocks, usually written as separate sentences or lines rather than one flowing paragraph.
1. Onderwerp en setting (subject and setting). Who or what, where, at what time of day, in what weather. Keep it to one sentence: “Een oudere visser in een gele regenjas staat op een houten pier bij zonsopkomst, lichte mist boven het water.”
2. Actie (action). One primary action per shot. “Hij trekt langzaam een net binnen.” If you need two actions, split them into two shots and describe each separately.
3. Camera (shot framing and movement). Format, angle, lens feel, movement. “Halftotaal, lichte kikvorshoek, 35mm-look, langzame rij-opname naar voren.”
4. Licht en sfeer (light and mood). Direction and quality of light, plus emotional register. “Zacht tegenlicht, warme gloed, rustige en melancholische sfeer.”
5. Stijl en textuur (style and texture). Film stock, grain, color grade, realism level. “Documentaire-realisme, fijne korrel, licht gedempte kleuren.”
6. Uitsluitingen (negatives). What should not appear. “Geen tekst in beeld, geen watermerk, geen vervormde handen, geen extra ledematen.”
Splitting the prompt this way makes debugging possible. When a render fails, you can point at exactly which block is weak rather than rewriting everything.
Order matters more than length
Put the subject first, camera second, and style last. Models weight early tokens more heavily, so the thing you care about most — the person, the object, the action — should never be buried at the end behind stylistic adjectives.
Keep one idea per shot
A shot is a unit of time and attention. If your description contains a cut, a location change, or a new character entering, it is two shots. Writing them as one prompt is the single most common cause of chaotic output.
Dutch Language Craft: Word Order, Compounds, and Motion Verbs
Dutch is a beautiful, compact language and a surprisingly difficult prompting language. Four quirks cause most quality problems.
Verb-final subordinate clauses
Dutch pushes verbs to the end of subordinate clauses: “de man die op de pier staat te wachten.” Models handle this unevenly because the action arrives late in the sentence. When motion is the point, use main clauses with the verb early. Instead of “De vrouw die langzaam naar de deur loopt, kijkt achterom,” write two clean main clauses: “Een vrouw loopt langzaam naar de deur. Ze kijkt achterom.” The model sees the movement sooner and animates it more clearly.
Compound nouns and separable verbs
Dutch builds meaning by stacking: “sleutelgatverlichting,” “tegenlichtopname,” “bovenaanzicht.” Compounds are efficient but models sometimes parse them as noise. Write them out when precision matters: “verlichting die door het sleutelgat valt” beats a single invented compound. The same applies to separable verbs — “de camera zoomt langzaam in” is far clearer than “de camera inzoomt.”
Anglicisms versus native vocabulary
Some English film terms are so embedded in Dutch production language that they work fine in prompts: close-up, pan, tracking shot. Others create ambiguity because the Dutch word already means something else. Where a native term is precise, use it; where the English term is the industry standard, keep it and add a Dutch clarifier. “Een tracking shot (rij-opname) langs een marktkraam” removes all doubt.
Adjective density
Dutch tolerates stacked adjectives, but video models do not reward them. Three well-chosen descriptors — “versleten, donkerblauw, waterdicht” — outperform ten vague ones. Every adjective you add is a small amount of attention taken away from the subject.
Concrete over abstract
“Een mooie, sfeervolle scène” tells the model nothing. “Een natte straat met spiegelingen van oranje straatlantaarns” tells it everything. Replace judgments with observable details: texture, color, distance, weather, material.
Camera, Lens, and Light Vocabulary in Dutch
Building a personal glossary is the fastest quality upgrade available. Write it once, reuse it forever, and your output becomes visually consistent across dozens of clips.
| Dutch term | English equivalent | Typical prompt use |
|---|---|---|
| Groothoekopname | wide shot | establishing the environment |
| Halftotaal | medium shot | dialogue and body language |
| Close-up | close-up | emotion and facial detail |
| Kikvorsperspectief | low angle | power, heroism, scale |
| Vogelperspectief | high angle | vulnerability, geography |
| Over-de-schouderopname | over-the-shoulder | conversation framing |
| Langzame rij-opname | slow dolly in | building tension |
| Pannen | pan | revealing a space |
| Statische opname | static shot | calm, observational tone |
| Ondiepe scherptediepte | shallow depth of field | isolating a subject |
| Zachte lichtval | soft lighting | intimate scenes |
| Tegenlicht | backlight | silhouettes, atmosphere |
| Gouden uur | golden hour | warm, cinematic mood |
| Blauwe uur | blue hour | cool, quiet, urban mood |
| Neonverlichting | neon lighting | night city, genre looks |
| Bewegingsonscherpte | motion blur | speed and kinetic energy |
Two practical notes. First, combination beats accumulation: “halftotaal met ondiepe scherptediepte” reads as a single intention, while “halftotaal, groothoek, close-up” reads as confusion. Second, lighting direction is the strongest mood lever you have. Say where the light comes from before you describe how it feels.
Character and Style Consistency Across a Shot List
Consistency is where hobby prompts and production prompts diverge. A single beautiful clip is easy; ten clips that look like the same film is engineering.
Write a character bible in Dutch — and freeze it
Create one block of text describing each recurring character and never paraphrase it. If your lead is “een vrouw van in de dertig met kort rood haar, groene ogen, een litteken boven de linkerwenkbrauw en een versleten leren jack,” copy that exact string into every prompt. Rewriting it from memory is how faces drift between shots.
Lock style with a reusable suffix
Build a style suffix you append to every prompt: film stock, grain, contrast, color palette, aspect ratio, and realism level. Something like “35mm-film look, fijne korrel, natuurlijk contrast, gedempte teal-en-oranje kleuren, 16:9.” Because the suffix is identical across shots, the model treats it as part of the scene rather than a new style request.
Use reference images when the tool supports them
Image-to-video and multi-image workflows are dramatically better at identity preservation than text alone. A clean, evenly lit reference photo of your character, plus a locked description string, gives you two independent anchors. If results still drift, add the character’s most distinctive feature again in the action block — it acts as a mid-prompt reminder.
Separate wardrobe as its own variable
Describe the character’s body and face in the character bible, and describe clothing in the shot block. That way you can change a jacket between scenes without unlinking the face, which is exactly the kind of change that normally breaks continuity.
Directing Motion and Time in Dutch
Motion is the hardest thing to prompt and the easiest to get wrong, because a video model has to decide not just what moves but how fast, in which direction, and with what kind of weight.
Name the direction, not just the verb
“Hij loopt” is ambiguous. “Hij loopt van links naar rechts door het kader” is not. Directional phrasing — van links naar rechts, naar voren, van de camera af, de trap op — removes guesswork and stabilizes the composition.
Specify speed with adverbs you trust
Dutch has a rich set of pacing adverbs: langzaam, rustig, vloeiend, abrupt, haastig, traag, gelijkmatig. Choose one per shot and commit to it. “Langzaam en gelijkmatig” produces a very different clip than “snel en schokkerig,” and models do respond to that distinction.
Control time with explicit pacing language
If you want slow motion, say so literally: “vertraagd beeld, alsof het wordt afgespeeld op een kwart snelheid.” For timelapse: “tijdsverloop, wolken trekken snel over het landschap.” Vague words like “cinematisch” do not communicate temporal pacing at all.
Give the camera a reason to move
Camera motion works best when motivated. “Camera volgt de hardloper” is coherent; “camera draait rond terwijl niets gebeurt” is noise. If you cannot justify the movement with an on-screen subject, use a static shot instead — it will look more professional.
A Repeatable Workflow From Brief to Render
This sequence keeps projects organized and makes failures cheap to fix.
Step 1: Write a shot list in Dutch. One line per shot, in plain language. No camera terms yet. Example: “Visser trekt net binnen op de pier bij ochtendmist.”
Step 2: Expand each line into the six blocks. Subject, action, camera, light, style, negatives. Keep each block to one sentence.
Step 3: Build your shared assets. A character bible string, a style suffix, and a negative list. Save them in a text file or notes app so every prompt pulls from the same source.
Step 4: Generate short test clips. Render three to five seconds per shot at lower resolution. You are validating structure, not quality.
Step 5: Read the failure, not the clip. If the face drifts, the character string changed. If the framing is wrong, the camera block is weak. If motion is chaotic, the action block has two actions. Diagnose by block.
Step 6: Fix one variable at a time. Change exactly one thing per re-render. If you change framing, lighting, and motion simultaneously and the result improves, you have learned nothing reusable.
Step 7: Freeze the prompt when it works. Copy the winning prompt into a library tagged by shot type — establishing, dialogue, action, insert. Consistency compounds across a project once you stop rewriting from scratch.
Step 8: Assemble and check continuity. Watch the clips in sequence at low volume. Differences in color temperature and grain between shots are what make AI footage feel stitched together, so correct those before polishing anything else.
Common Mistakes and Their Fixes
Mixing multiple shots in one prompt. Fix: one action, one camera setup, one prompt. Split everything else.
Writing in Dutch without checking keyword alignment. Fix: for key visual terms, add a short parenthetical clarifier in the language your tool handles best. This is not cheating; it is reducing ambiguity.
Overloading with adjectives. Fix: cap yourself at three descriptors per noun and delete any adjective you cannot point to in the finished frame.
Forgetting negatives. Without a negative block, models invent subtitles, watermarks, extra fingers, and background text. Write the exclusions once and append them every time.
Changing style mid-project. Fix: treat the style suffix as a contract. If you need a new look, make it a deliberate decision and re-render the affected shots.
Describing emotion instead of behavior. “Verdrietig” is abstract. “Ze kijkt omlaag, haar schouders zakken, ze ademt langzaam uit” gives the model something to animate.
Ignoring aspect ratio and delivery format. Fix: state 16:9, 9:16, or 1:1 in every prompt. Vertical social edits crop differently, and discovering that after rendering horizontal footage wastes time.
Troubleshooting and Tool Selection Criteria
When a render fails repeatedly, work through this order: simplify the prompt to the subject and one action; then re-add camera; then light; then style. The failure point tells you which block is overloaded.
If faces morph, shorten the prompt and strengthen the character string rather than adding more adjectives. If motion looks like a slideshow, reduce the number of objects in frame and name a single directional movement. If output looks plasticky, add texture language — korrel, huidstructuur, stof, reflectie — and reduce the realism-word, because “hyperrealistisch” often pushes models toward a synthetic finish.
For tool selection, judge candidates on five criteria that actually affect Dutch-language work:
- Prompt length tolerance. How much detail survives before the model starts ignoring the middle of your prompt.
- Reference image support. Single-image, multi-image, or none — this decides whether consistency is achievable at all.
- Motion control granularity. Can you ask for a specific camera move and get it, or only a general sense of movement?
- Aspect ratio and duration options. Match these to your distribution channel before you build a library.
- Iteration speed. Fast low-resolution previews matter far more than final-render quality for the first twenty attempts.
Pick one primary tool, learn its Dutch quirks deeply, and keep a second as a fallback for shots your main tool handles badly.
FAQ
Should I write AI video prompts in Dutch or English?
Write in Dutch if your team works in Dutch and your tool handles it well, because review and iteration become faster and less error-prone. If a specific shot keeps failing, translate just that prompt into English, compare results, and keep whichever version is more reliable. Many professionals write Dutch prompts with English technical tags for camera and lens terms.
How long should a Dutch video prompt be?
Between 60 and 120 words is a practical range for a single shot. Under 40 words usually leaves too much to chance; over 150 words, models start dropping mid-prompt instructions.
Why do my characters change appearance between shots?
Almost always because the character description was paraphrased instead of copied. Freeze one exact string, use reference images if available, and repeat the character’s most distinctive feature inside the action block.
Does Dutch word order confuse video models?
It can. Subordinate clauses that place the verb at the end delay the action, which is where models lose clarity. Rewrite those as two short main clauses with the verb near the front.
How do I get a specific camera movement?
Name the movement and the direction, keep the rest of the prompt simple, and give the movement a subject to follow. Complex camera requests fail most often in visually busy scenes.
Can I reuse one prompt for vertical and horizontal video?
Reuse the description blocks, but rewrite the framing block. Vertical formats need closer framing and more headroom awareness than 16:9, so treat them as separate shot designs.
A Short Checklist Before You Render
Before hitting generate, confirm six things: one clear subject, one action with a named direction, an explicit camera setup, a lighting direction, a frozen style suffix, and a negative list. Then test at low resolution, fix one variable per attempt, and archive the prompt the moment it works.
Dutch prompting rewards patience and discipline more than vocabulary. The writers who get consistently cinematic output are not the ones with the biggest word lists — they are the ones who built a small, precise glossary, locked their character descriptions, and stopped changing five things at once. Start with one scene, one character, and one camera move. Build the library from there, and the quality gap between your first attempt and your twentieth will be obvious.
