Model note: prompting is now model-specific, and vendor guides differ. These patterns use the structure most current guides share. Adapt them to the model you're using, and always read that model's own guide. Phrasing that works well on one model may be ignored by another. Test, and log what works.
1. The base recipe #
Most current guides agree on roughly this order:
[SHOT SIZE + ANGLE] of [SUBJECT], [ONE ACTION], in [SPECIFIC ENVIRONMENT + TIME],
[CAMERA MOVE + SPEED], [LENS / DEPTH OF FIELD], [LIGHTING: source, direction, quality],
[AUDIO, if the model generates sound], [STYLE / TEXTURE — last and brief]
Example
Medium close-up, eye level, of a woman in her 60s with silver braids and a rust wool coat, slowly unfolding a letter, in a cramped kitchen at dawn. Static camera. 50mm lens, shallow depth of field. Soft warm window light from screen left, cool fill from the hallway. Quiet room tone, the paper rustling. 35mm film grain, muted palette.
The five rules #
- One shot, one action, one camera move. Stacking actions or moves is the most common cause of warping and morphing.
- Be physical, not emotional. "Slow dolly in, 85mm, f/1.8" beats "cinematic and dramatic."
- Specific environment over mood words. "Rain-wet neon alley in Osaka at 2 a.m." beats "moody city."
- Style goes last. Style words at the front tend to override everything after them.
- References beat adjectives. Where the tool accepts images, a reference frame does more than a paragraph of description.
2. Shot sizes #
| You want | Prompt phrasing | Notes |
|---|---|---|
| Scale, isolation | "extreme wide shot," "tiny figure in a vast…" | Add the subject's position: "small in the lower third" |
| Geography | "wide shot showing the full room," "full body in frame" | Good establishing shot and image-to-video base |
| Body language | "full shot, head to toe" | |
| Group, westerns | "cowboy shot, framed from mid-thigh up" | |
| Conversation | "medium shot, waist up" | |
| Dialogue emotion | "medium close-up, chest up" | The workhorse for dialogue |
| Emotion | "close-up of her face" | Faces large in frame reduce identity drift |
| Intensity, detail | "extreme close-up of his eyes," "macro shot of…" | |
| Information | "insert shot of a hand turning a key" | Make separately and cut in |
Two-person framings: "over-the-shoulder shot from behind [A] looking at [B]," "two-shot, both in profile, facing each other."
POV: "first-person point of view, looking down at their own hands holding…"
3. Angles #
| You want | Prompt phrasing | Effect |
|---|---|---|
| Neutral | "eye-level shot" | Default; state it if the model keeps drifting high or low |
| Power, threat | "low-angle shot looking up at…" | |
| Vulnerability | "high-angle shot looking down at…" | |
| Pattern, overview | "top-down overhead shot," "bird's-eye view" | |
| Unease | "Dutch angle, camera tilted 20 degrees" | Use sparingly |
| Monumental | "worm's-eye view from ground level" | |
| Height precision | "camera at knee height," "hip-level shot" | Useful for kids, animals, action |
4. Camera movement (video) #
| Move | Prompt phrasing | Tips |
|---|---|---|
| Static | "static locked-off camera," "the camera does not move" | Saying it explicitly stops unwanted drift |
| Push-in | "slow dolly in toward her face" | Add speed: slow, steady, quick |
| Pull-out / reveal | "slow dolly out revealing the empty stadium" | |
| Pan | "slow pan left to right across the skyline" | Name the direction |
| Tilt | "tilt up from his boots to his face" | |
| Truck / track | "camera tracks sideways alongside her as she walks" | |
| Follow | "camera follows behind him down the corridor" | |
| Lead | "camera leads in front of her, moving backward" | |
| Orbit | "camera slowly orbits 90 degrees around the statue" | Limit the arc to reduce warping |
| Crane | "crane up from street level to rooftops" | |
| Handheld | "handheld camera, subtle natural shake" | "Subtle" prevents seasick results |
| Gimbal | "smooth gimbal shot gliding forward" | |
| Drone | "aerial drone shot flying low over the forest" | |
| FPV | "fast FPV drone dive between buildings" | High warp risk |
| Whip pan | "fast whip pan to the right" | Usually better done in the edit |
| Rack focus | "rack focus from the glass in the foreground to the man in the background" | Model support varies; Resolve CineFocus can do it in post |
| Dolly zoom | "dolly zoom, background stretches while subject stays the same size" | Hard for most models; consider post |
Use the tool's camera controls when it has them. Native camera-control features, camera LoRAs or camera nodes are more reliable than words alone.
5. Lenses & depth of field #
| You want | Prompt phrasing |
|---|---|
| Expansive, slight distortion | "shot on a 16mm wide-angle lens" / "24mm lens" |
| Natural perspective | "35mm lens" / "50mm lens" |
| Flattering portrait, compression | "85mm portrait lens, compressed background" |
| Isolation, stacked depth | "200mm telephoto lens, heat haze, compressed perspective" |
| Blurred background | "shallow depth of field, creamy bokeh, f/1.8" |
| Everything sharp | "deep focus, everything sharp from foreground to background" |
| Widescreen character | "anamorphic lens, oval bokeh, horizontal lens flares" |
| Miniature look | "tilt-shift, miniature effect" |
| Macro | "macro lens, extreme detail" |
| Vintage | "vintage lens, soft edges, slight vignette" |
Watch out: generated footage often comes with the wrong depth of field whatever the prompt says. Budget for post-focus fixes.
6. Lighting #
Classic setups #
| Setup | Prompt phrasing |
|---|---|
| Three-point | "key light from the left, soft fill, rim light separating her from the background" |
| Rembrandt | "Rembrandt lighting, small triangle of light on the shadowed cheek" |
| Butterfly | "butterfly lighting from above, small shadow under the nose" |
| Split | "split lighting, half the face in shadow" |
| Silhouette | "backlit silhouette against a bright window" |
| Low-key / noir | "low-key lighting, deep shadows, single hard source, venetian blind shadows" |
| High-key | "bright high-key lighting, soft and even, minimal shadows" |
| Chiaroscuro | "chiaroscuro, dramatic light and dark contrast like a Caravaggio painting" |
Sources & times #
| You want | Prompt phrasing |
|---|---|
| Motivated practical | "lit only by the glow of a laptop screen" / "a single bare bulb overhead" |
| Window light | "soft daylight from a large window, screen left" |
| Golden hour | "golden hour, low warm sun behind, long shadows" |
| Blue hour | "blue hour twilight, cool ambient light, warm windows glowing" |
| Night exterior | "night, sodium-vapor streetlights, wet reflective pavement" |
| Neon | "magenta and cyan neon signs reflecting on wet asphalt" |
| Firelight | "flickering warm firelight on faces" |
| Volumetric | "haze, visible light beams through the dusty air" |
| Overcast | "flat overcast light, soft shadows, muted colors" |
Keep light consistent across a scene #
Paste the same lighting line into every prompt in the scene:
"…warm key light from a window screen-left, cool blue fill from the hallway screen-right, 5600K daylight."
7. Composition #
| You want | Prompt phrasing |
|---|---|
| Rule of thirds | "subject positioned on the left third, looking toward the right" |
| Symmetry | "perfectly symmetrical, centered composition" |
| Negative space | "small figure in the bottom right, vast empty sky above" |
| Frame within frame | "framed through a doorway," "seen through the car window" |
| Leading lines | "railway tracks leading to the figure on the horizon" |
| Depth layers | "out-of-focus foliage in the foreground, subject in the midground, mountains behind" |
| Lead room | "facing right with open space in front of her" |
| Short-sided unease | "pressed against the right edge of frame, looking off-screen right" |
8. Color, texture & film look #
| You want | Prompt phrasing |
|---|---|
| Film texture | "35mm film grain, slight halation on highlights" |
| Specific stock feel | "Kodak Portra–like warm skin tones" / "cool Fuji greens" (results vary by model) |
| Muted | "desaturated, muted earth tones" |
| Teal and orange | "teal shadows, warm orange skin tones" |
| Monochrome | "black and white, high contrast" |
| Bleach bypass | "bleach bypass look, low saturation, crushed contrast" |
| Pastel | "soft pastel palette" |
| Palette lock | "color palette: deep navy, rust orange, cream" |
Tip: for a consistent film, set the look in the grade rather than the prompt. Prompts get you into the right area; the grade unifies everything.
9. Performance & emotion #
Models handle visible behavior better than abstract emotion. Describe what the face and body do:
| Instead of | Try |
|---|---|
| "sad" | "eyes glistening, jaw tight, she looks down and swallows" |
| "angry" | "nostrils flared, he slams his palm flat on the table" |
| "nervous" | "fidgeting with her ring, glancing at the door" |
| "happy" | "a slow smile breaks across his face, eyes crinkling" |
| "thinking" | "she pauses, eyes unfocused, lips slightly parted" |
Subtext tip: generate a neutral reaction and let the edit do the work (the Kuleshov effect).
10. Audio direction (models with native sound) #
| Element | Prompt phrasing |
|---|---|
| Ambience | "ambient sound: distant traffic, a dog barking far away" |
| Specific SFX | "sound of boots crunching on gravel" |
| Dialogue | She says, quietly and hesitantly: "I didn't think you'd come." |
| Silence | "no music, only room tone" |
| Music | "soft solo piano underscore" (or leave music out and score in post) |
For voice models with inline tags, put delivery direction next to the line, e.g. [whispering], [laughs softly]. Check which tags your voice model supports. Some newer models disable older SSML tags.
11. Coverage pack: one scene, many shots #
Templates for generating matching coverage of one dialogue scene. Keep the character blocks, location block and lighting line identical in every prompt.
[LOCATION BLOCK] = "a cramped 1970s kitchen, yellow tile, rain on the window, dawn"
[LIGHT BLOCK] = "warm key light from the window screen-left, cool fill from the hallway"
[A] = "<Mara canonical description>"
[B] = "<Teo canonical description>"
[STYLE] = "35mm film grain, muted palette"
1. Master: Wide shot, eye level, [A] at the table screen-left, [B] in the doorway screen-right, [LOCATION]. Static. 28mm. [LIGHT]. [STYLE]
2. OTS on B: Over-the-shoulder shot from behind [A]'s left shoulder looking at [B] in the doorway. Static. 50mm, shallow DoF. [LIGHT]. [STYLE]
3. OTS on A: Over-the-shoulder shot from behind [B]'s right shoulder looking at [A] at the table. Static. 50mm, shallow DoF. [LIGHT]. [STYLE]
4. CU A: Close-up of [A], looking screen-right toward the doorway, [reaction]. Static. 85mm. [LIGHT]. [STYLE]
5. CU B: Close-up of [B], looking screen-left toward the table, [reaction]. Static. 85mm. [LIGHT]. [STYLE]
6. Insert: Extreme close-up of the letter on the table, [LIGHT]. [STYLE]
Note the eyelines: A looks screen-right and B looks screen-left in every shot. That keeps you on one side of the 180° line.
Workflow tip: generate the master as a still first, then derive the other angles with multi-reference or camera-angle tools, and animate each with image-to-video.
12. Troubleshooting prompts #
| Problem | Try |
|---|---|
| Camera moves when you wanted static | Add "static locked-off camera, no camera movement" |
| Wrong shot size | Put the shot size first, and add "framed from chest up" |
| Too "cinematic," generic | Remove "cinematic," "epic," "masterpiece"; add specific physical detail |
| Characters cross and morph | Stage them apart; one moves, one is still |
| Warping during the move | Simplify to one move, slow it down, shorten the duration |
| Style overrides content | Move style words to the end, reduce them, or use a style reference |
| Light changes between shots | Paste the identical light block; state the direction |
| Ignores part of the prompt | Shorten. Cut adjectives first, keep nouns and verbs |
| Text is garbled | Remove text from the prompt; add it in post |
13. Build your own library #
Keep every prompt that works, with:
- Model + version (prompts don't transfer reliably between models)
- Seed and settings
- The output (thumbnail)
- What it's good for (e.g. "reliable slow push-in on faces, H3 I2V")
Share your best patterns on AiFC as workflow cards. Tagged with shot terms from the glossary, they become searchable for everyone.