Video generation models don't read your mind — they read your sentences. The difference between a blurry, random clip and a deliberate shot is almost always in how you phrase the prompt. In this advanced guide we'll use the /video-prompting skill in FilmMovie Studio to learn the craft of language: a concrete subject, action through visual verbs, one camera move, named lighting and a specific film style. We'll also compare two prompt variants side by side on the canvas and refine the winner.
What you'll need
- An open project in FilmMovie Studio with a video model configured (kie.ai or deAPI).
- Basic familiarity with the canvas and chat (see the "Your first project" guide).
- A concrete idea for one shot — one subject and one main action, not a whole scene.
Step by step
- 1
Invoke the /video-prompting skill
In chat type "/video-prompting" and pick it from the list, or just describe the clip ("a drone over a snowy forest at dawn"). The skill acts as a video prompt engineer: it builds one production-ready prompt from your words and saves it to the canvas as a document card.
Film Movie StudioRenderLast Tuesday At LumièReNew projectSkills/video-prompting/short-drama/image-creator/audio-studio/app-builder/video-prompting Cinematic live-action photography shot through a heavy nostalgic soft-glow filter. Dreamy, ethereal lighting featuring hazy golden light leaks and a shallow…
SkillsModelsType / to switch skills anytime
Tip: Describe one shot, not a whole story. One subject + one main action = one clean clip. - 2
Name the subject concretely
Open the sentence with a concrete subject: "a woman in a red silk dress", not "a beautiful woman". The model invents anything you leave out — so state the detail you care about (material, color, age, setting) in the text.
Film Movie StudioRenderLast Tuesday At LumièReNew projectFiles
frame_1.png
frame_2.png
frame_3.png
frame_4.png
frame_5.png60%promptframe_01.png
Generating… 42%Remixframe_02.png
Generating… 68%Remixframe_03.png
Generating… 55%Remixframe_04.png
Generating… 80%RemixoutputTip: Swap judgement adjectives for visible detail: instead of "a stunning shot" write "golden light glides across the bottle's glass". - 3
Write the action with visual verbs
Describe motion in the order start → development → end and use temporal markers "first… then… finally…". Choose only visible verbs (blooms, cracks, levitates, dissolves), never cognition verbs (realizes, decides) — the camera can't film those.
Film Movie StudioRenderLast Tuesday At LumièReNew projectSkills/video-prompting/short-drama/image-creator/audio-studio/app-builder/video-prompting Cinematic live-action photography shot through a heavy nostalgic soft-glow filter. Dreamy, ethereal lighting featuring hazy golden light leaks and a shallow…
SkillsModelsType / to switch skills anytime
- 4
Add one camera move and a lens
Append exactly one camera move plus lens language: "slow dolly in", "orbiting medium shot", "static close-up, 85mm, shallow depth of field". One clip = one move; stacking moves makes the model wander.
Film Movie StudioRenderLast Tuesday At LumièReNew projectFiles
frame_1.png
frame_2.png
frame_3.png60%promptframe_01.png
Remixframe_02.png
Remixframe_03.png
Generating… 55%RemixoutputTip: Move vocabulary: push-in = rising intensity, pull-back = context reveal, orbit = show volume, tracking = energy and movement. - 5
Add lighting and style, avoid AI tells
Add a named light source ("golden hour backlight", "soft window light") and a specific film style ("16mm grain, muted teal-orange grade"). Don't use words like "8K", "4K", "hyper-realistic" or "masterpiece" — they trigger the generic AI look.
Film Movie StudioRenderLast Tuesday At LumièReNew projectFiles
frame_1.png
frame_2.png
frame_3.png
frame_4.png60%promptframe_01.png
Generating… 42%Remixframe_02.png
Generating… 68%Remixframe_03.png
Generating… 55%Remixframe_04.png
Generating… 80%Remixoutput - 6
Choose T2V or I2V with a reference image
To control appearance, drag an existing image card onto the chat or attach a file — the skill switches to image-to-video (I2V) and spends the prompt on motion, camera and atmosphere, plus one sentence locking the look to the reference. With no image it's text-to-video (T2V) and the prompt must describe the whole appearance.
Film Movie StudioRenderLast Tuesday At LumièReNew projectFiles
frame_1.png
frame_2.png
frame_3.png
frame_4.png
frame_5.pngDrag an image here, or click to upload
reference.png
Tip: For I2V pick a frame "just before the action" (about to pour, mid-breath) so the motion has somewhere to go. - 7
Generate and compare variants on the canvas
After the document is saved, confirm "Generate now". Run the prompt, then change it slightly (e.g. dolly in → orbit) and generate a second version. Lay both video cards side by side on the canvas and compare which holds the subject and pacing better.
Film Movie StudioRenderCupcake'S First FlightNew projectFiles
frame_1.png
frame_2.png
frame_3.png
frame_4.pngclip_5.mp4
frame_6.png60%promptframe_01.png
Remixframe_02.png
Remixframe_03.png
Remixframe_04.png
Generating… 80%RemixoutputTip: Change only one thing at a time (camera, or lighting, or pacing) — otherwise you won't know what caused the difference. - 8
Refine the winner for length and sound
Pick the better prompt and edit it via "remix": trim the number of actions to fit the duration (~1 visual moment per 1.5s; a 6s clip holds 3–4 moments) and, if sound matters, add one sentence of diegetic audio. Then add the card to the timeline.
Film Movie StudioRenderLast Tuesday At LumièReNew projectSkills/video-prompting/short-drama/image-creator/audio-studio/app-builder/video-prompting Cinematic live-action photography shot through a heavy nostalgic soft-glow filter. Dreamy, ethereal lighting featuring hazy golden light leaks and a shallow…
SkillsModelsType / to switch skills anytime
What you get
You're no longer asking the model for a "nice video" — you're directing it. You can build a prompt in the order subject → action → camera → lighting → style, choose visual verbs, use one camera move per clip, and control pacing by fitting actions to the duration. And when unsure, you stand two variants side by side on the canvas, change one thing at a time, and let the result decide.
Pro tips
- Phrase the camera move concretely: "slow dolly in, 35mm, shallow depth of field" beats "cinematic shot".
- Control pacing by action count: ~1 visual moment per 1.5s. A 6s clip = one main action + one reaction; more and the model drops or rushes beats.
- Always ask for the desired state, not the absence: "empty, deserted street", not "no people"; "sharp focus", not "no blur".
Try it yourself
FilmMovie Studio is a desktop app that walks you from idea to finished export. Download it and start creating.
Get FilmMovie Studio