01
Build the frame
Shape subject, setting, composition, light, texture, and exact on-image text.
Explore the workflow Model-aware creative direction
Cinematic prompts. Crafted with intention.
Turn rough ideas into polished image and video prompts with clear direction for subject, setting, lighting, camera, movement, and mood.
Four distinct model strategies Natural language + structured JSON
Scene
A quiet ritual in morning light
A beauty founder prepares a product beside a tall hotel window, calm and self-possessed.
9:16GPT Image 2Best Format
CAMERAMedium close-up · 50mm
LIGHTWindow, camera right
Finished prompt
Create one vertical editorial image of an adult beauty founder standing beside a sheer-curtained hotel window…
IDEA TO IMAGEREFERENCE ANALYSISIMAGE TO VIDEOCHARACTER CONTINUITYPROMPT REPAIR
A more intentional workflow
From a flicker of an idea to production-ready direction.
Enough structure for precision. Enough restraint to keep the process approachable.
02
Direct the motion
Turn a still or written idea into a timed Seedance direction with camera, sound, and continuity.
Explore the workflow03
Hold the identity
Create reusable character anchors and give every reference a precise production role.
Explore the workflowBefore / after
A brief idea becomes visible direction.
CineVellum replaces generic adjectives with choices a generation model can actually see.
“A luxury skincare creator in a hotel.”
Vague setting · no camera · no visible lightingCreate one vertical editorial image of an adult beauty creator at a limestone vanity beside a floor-to-ceiling hotel window. Frame a medium close-up at eye level with a 50mm perspective…
Single composition · exact finish · natural realismModel-aware by design
One intent. Four distinct directions.
CineVellum recompiles your creative intent for the way each supported model responds best.
ModelBest forDirection style
GPT Image 2Editorial images, edits, exact textNatural, direct, composition-first
Nano Banana ProMulti-reference, identity, productsRole-aware, preservation-led
Nano Banana 2Rapid variations and social conceptsConcise, spatial, instruction-forward
Seedance 2.0 via HiggsfieldText or image to videoMotion, timing, continuity, audio
01 / CHARACTER
Keep the person recognizable.
Build an editable character anchor for facial structure, skin tone, hair, proportions, wardrobe, and distinguishing details.
Character consistency
Identity is a direction, not a checkbox.
Reuse what matters across scenes while still changing the setting, action, framing, or format with intention.
- Reference roles with clear priority
- Preserve and borrow rules
- Stable character anchors
- Scene-by-scene recompilation
02 / REFERENCES
PRIMARY IDENTITYWARDROBEENVIRONMENTLIGHTING
Every reference has a job.
Assign identity, product, wardrobe, environment, composition, lighting, motion, or final-frame roles.
Prompt Library
Your strongest direction gets better with time.
Save, search, favorite, archive, duplicate, adapt, and restore every meaningful version without overwriting your history.
Explore the Library01
Hotel Window RitualGPT Image 2
9:1602
Midnight Product RevealSeedance 2.0
16:903
Tailored Street PortraitNano Banana Pro
4:5Questions, answered
Built for creators, not prompt engineers.
Does CineVellum generate images or videos?
CineVellum creates copy-ready, model-optimized direction. Direct provider generation is only added when a documented integration is available.
Why not use the same prompt for every model?
Each model responds differently to structure, preservation rules, reference roles, text, and motion. CineVellum uses a dedicated strategy for each supported model.
Can I use reference images?
Yes. References remain private, receive explicit roles, and are compiled into the final direction without trusting text embedded inside them.
Is JSON always better?
No. Best Format chooses natural language, JSON, or both based on the actual workflow and intended use.
The next frame is yours
Start creating