Learn how to separate scene, subject, action, camera, lighting, audio and constraints into reusable fields with a practical workflow, ready-to-copy.
Key takeaways
- Start with a clear creative objective: separate scene, subject, action, camera, lighting, audio and constraints into reusable fields.
- Lock important identity details before changing style or environment.
- Direct composition, light, material and motion separately.
- Refine one variable at a time and review the output at full size.
- Create the clean visual first; add final typography during design.
JSON prompts for AI video
In this guide, JSON prompts for AI video means using generative tools as part of a controlled creative workflow rather than treating the first generation as finished work. The process combines a clear brief, source references, structured prompting, visual review and deliberate refinement.
The goal is not simply to make something eye-catching. It is to make an asset that is accurate, usable in the intended format and consistent with the brand. For advanced creators, teams, and automation workflows, that distinction saves time because fewer attractive-but-unusable results reach the final selection.
Why this workflow matters
JSON does not magically improve a model, but it can make complex direction easier to review, reuse and hand off across a team. A reliable system protects the details viewers notice first while still leaving room for creative exploration.
It also improves collaboration. A social media manager can explain the desired outcome, a designer can control the visual system, and an editor can understand which parts are fixed. The prompt becomes a shared production brief instead of a private experiment.
Step-by-step workflow
1. Define the scene and duration
Keep this stage simple and measurable. Write down what must remain unchanged and what may vary before you generate anything.
2. Create a locked subject object
Use precise visual language. Name position, scale, material, direction or duration instead of relying on broad adjectives.
3. Write ordered action beats
Create one controlled test. If the test fails, correct the smallest possible variable rather than rewriting the whole direction.
4. Separate camera from subject action
Compare the output with the source or brief at full size. Attractive lighting should never hide an accuracy problem.
5. Add lighting and audio only when supported
Save the approved wording and output as a reusable reference. Consistency becomes easier when the team can see what passed review.
6. Centralise negative constraints
Prepare the final asset for its real placement. Check the crop, safe area, resolution and whether later typography has enough breathing room.
Ready-to-copy prompt
{
"duration_seconds": 8,
"format": "9:16",
"scene": "warm modern kitchen at night",
"subject_lock": "exact uploaded noodle pack; unchanged logo, colours and proportions",
"action": ["pack lands softly on counter", "camera reveals steaming bowl", "chopsticks lift noodles"],
"camera": "smooth table-height dolly forward; no cut",
"lighting": "warm practicals with soft cool rim light",
"constraints": ["realistic food texture", "no morphing", "no extra text", "no extra hands"]
}
How to customise it: Replace the audience, setting, palette, action and aspect ratio, but keep the identity lock and the exclusions that protect your most important details.
How to improve the first result
Use a controlled refinement loop:
- Accuracy pass: Fix identity, proportions, anatomy, label, ingredients or continuity.
- Composition pass: Adjust subject size, crop, balance and negative space.
- Lighting pass: Correct direction, softness, reflections and colour temperature.
- Texture pass: Remove waxy, plastic or over-smoothed surfaces.
- Delivery pass: Export the correct ratio and test it in the final layout.
Change one category per round. If you ask for a new angle, new wardrobe, new background, new light and new action together, you will not know which instruction caused the next error.
Common mistakes
- Adding unsupported technical fields. Correct it with a specific positive instruction and regenerate only the affected area when possible.
- Writing contradictory values in separate keys. Correct it with a specific positive instruction and regenerate only the affected area when possible.
- Turning every adjective into a field. Correct it with a specific positive instruction and regenerate only the affected area when possible.
- Assuming structure replaces clear creative direction. Correct it with a specific positive instruction and regenerate only the affected area when possible.
Featured image direction
Concept: A cinematic production dashboard concept with modular scene, subject, camera and lighting blocks surrounding a video frame; no readable text.
Alt text: A cinematic production dashboard concept with modular scene, subject, camera and lighting blocks surrounding a video frame; no readable text
Final thoughts
The strongest way to separate scene, subject, action, camera, lighting, audio and constraints into reusable fields is to combine clear creative judgment with a repeatable production system. Start simple, protect the non-negotiable details, and refine with purpose. AI can make production faster, but the quality still comes from the brief, the review and the decisions you make between generations.
Frequently asked questions
Can beginners use this workflow?
Yes. Start with one clear source image or idea, follow the steps in order, and keep the first test simple. The workflow is designed for advanced creators, teams, and automation workflows.
How many variations should I generate?
Generate enough to compare direction, not dozens without a plan. A practical first round is four variations, followed by one or two controlled refinements of the strongest result.
What should I check before publishing?
Check subject or product accuracy, anatomy, lighting, unwanted text, edge quality, aspect ratio, brand fit and whether the image or video delivers the promised message at mobile size.
Can I reuse the prompt for another brand?
Reuse the structure, but replace the identity, audience, palette, product, setting and exclusions. A reusable framework is valuable; copied creative direction is not.
Should I add text inside the AI generation?
Usually create a clean visual plate and add final typography in Canva, Photoshop or your design system. This gives you better spelling, hierarchy, localisation and crop control.
Related reading
READY FOR THE NEXT FRAME?
Turn the brief into
a visual system.
Explore prompt structures, visual directions, and practical creative workflows built for useful output.