How to prompt GPT Image 2 for storyboards and multi-panel scenes
The exact structure for 3-, 4-, and 6-panel storyboards with consistent character and lighting.
The structure
A working storyboard prompt has four parts:
- Grid — canvas ratio + panel layout + panel ratio
- Character anchor — one paragraph repeated in every panel
- Beats — one short sentence per panel describing action + camera
- Shared style — lighting, palette, medium, applied to every panel
Working template
4-panel storyboard, 2×2 grid, 16:9 landscape canvas, each panel 16:9, thin black border between panels. Character: a man in his 40s, weathered face, grey beanie, navy parka — same person in every panel. Panel 1: wide shot, character standing alone on a snowy ridge at dawn, looking right. Establishing. Panel 2: medium shot, character glances over shoulder, breath visible. Tension. Panel 3: close-up on character's eyes, narrow, focused. Decision beat. Panel 4: wide shot from behind, character walking toward camera-distant lodge lights. Resolution. Shared style: cinematic 35mm film, cold blue-grey palette, soft directional dawn light, painterly storyboard wash. Consistent character across all four panels.
Why each part matters
Grid
Three reliable formats:
- 3-panel vertical — 3:4 canvas, panels stacked
- 4-panel 2×2 grid — 16:9 canvas, four equal panels
- 6-panel comic style — 2:3 canvas, 3 rows × 2 columns
State both the *canvas* ratio and the *panel* ratio. Without panel ratio cues, the model squishes panels into whatever shape is left over.
Character anchor
Three to four specific physical features, exact same phrase every panel. The model's consistency is good when you give it a sharp anchor; it drifts when descriptions are loose.
- ❌ "the man" / "the same man" — drifts to a different face every panel
- ✅ "a man in his 40s, weathered face, grey beanie, navy parka" — stays consistent
Beats
One sentence per panel. Include the shot type (wide, medium, close-up) and the action (looking, walking, deciding). Skip dialogue — it garbles.
Shared style
The single line that makes the storyboard feel like one piece, not four images glued together. Lighting + palette + medium. Same phrase, every panel.
Skip the rewrite
Depikt's generator recognizes storyboard intent and applies this structure automatically — including the character anchor and shared-style line.
Generate yours
Generate polished prompts in seconds.
Paste a rough idea. Get back a structured prompt that ships.
More in How-to
How to prompt GPT Image 2 for logos and brand marks
GPT Image 2 can spit out genuinely usable logo directions if you brief it like a designer. Here's the four-block structure: concept, mark type, type system, finish — plus copy-paste templates.
How-toHow to prompt GPT Image 2 for infographics and diagrams
GPT Image 2 can render genuinely usable infographics — but only if you treat the prompt like a brief, not a vibe. Here's the five-block structure: format, hierarchy, data, type system, and finish.