| name | shot-prompt-builder |
| description | Convert a scene plan into stronger image or video generation prompts by using explicit shot language, movement, lighting, depth, and subject texture instead of generic style prefixes. |
Shot Prompt Builder
Use When
- AI-generated scenes are part of the short.
- Existing prompts are too vague, repetitive, or style-prefix heavy.
- The scene plan already knows roughly what each beat should show, but
the prompts need to become generation-ready.
Core Rule
- Do not prepend the same aesthetic prefix to every scene and call that
prompt design.
- Each prompt should encode why this shot is different:
framing, movement, subject, texture, and lighting.
Inputs
- scene description
- optional shot language:
shot_size, camera_movement, lens_mm, depth_of_field,
lighting_key, color_temperature
- optional
texture_keywords
- optional high-level style context for the whole project
- optional provider fields:
duration, aspect_ratio, negative_prompt, start_image,
end_image, reference_elements, motion_strength, seed,
provider, model, and rights_notes
Output
- one generation prompt per visual scene
- optional hero-scene markers for the scenes that deserve extra effort
- optional provider-ready fields for image-to-video, text-to-video,
multi-shot, or reference-driven providers
Prompt Construction Order
- Camera:
lens and depth of field
- Framing and movement:
wide, close-up, tracking, dolly, whip pan, etc.
- Subject:
the concrete thing happening in the scene
- Texture:
tactile descriptors, environment details, material cues
- Lighting and color:
high-key, neon, golden hour, tungsten, cold office, etc.
- Style:
only as a light final nudge, not the whole prompt
For video providers, also specify:
- first frame or start image when image-to-video is safer than
text-to-video
- end image or last-frame seed when continuity matters
- reference elements for character, object, style, or environment
- duration, aspect ratio, motion strength, and negative prompt
- provider/model and rights notes for any reference media
Good Prompt Behavior
- prompts from adjacent scenes should not read like minor variations of
the same sentence
- hero moments get stronger contrast, framing, or movement language
- the wording should be specific enough that an image model has a clear
visual target
- prompts should reflect the beat's editorial job, not just its topic
Bad Prompt Behavior
cinematic beautiful high quality futuristic style
- repeating the same style clause on every scene
- only naming the subject with no shot intent
- scenes with no texture clues or no change in camera language
Pair With
Aggregated From
calesthio/OpenMontage shot_prompt_builder.py
rushindrasinha/youtube-shorts-pipeline niche-aware draft prompts
Validation Checklist
- Each prompt encodes framing and subject, not just vibe.
- Adjacent prompts are visually distinct.
- Hero scenes are marked and read as higher-impact shots.
- Style language supports the prompt instead of replacing it.