A prompt is easier to revise when the subject, action and camera movement are separate fields. Here is a reproducible offline workflow for planning a reviewable scene before deciding whether to generate media.
This is Free AI Video’s independent, MIT-licensed prompt kit: a JavaScript library and CLI, not a model, model SDK or generation API. It does not call private website APIs, read reference images, upload files or require an account. It validates its own planning format, not whether a model can follow your instructions.
Install an exact version
Use Node.js 20 or later:
npm install --global freeaivideo-prompt-kit@0.1.1
freeaivideo-prompt-kit init --mode text-to-video > my-brief.json
Choose a new output filename: shell redirection overwrites an existing file. Source and tests are available in the independent repository.
Separate the creative intention into fields
{
"version": 1,
"mode": "text-to-video",
"subject": "A handmade ceramic cup on a wooden table",
"action": "Steam rises gently while the cup stays still",
"camera": "Single shot with a slow push-in",
"lighting": "Soft window light from the left",
"avoid": "Added lettering and logos"
}
The cup stays still, the steam moves, and the camera pushes in. Keeping these separate makes ambiguity easier to spot. For version B, change only the camera to “Locked camera, single shot” and compare the plans before generating anything. These are creative intentions, not generated results or quality guarantees.
Validate and export
freeaivideo-prompt-kit validate my-brief.json
freeaivideo-prompt-kit render my-brief.json --format prompt > my-prompt.txt
freeaivideo-prompt-kit render my-brief.json --format markdown > my-plan.md
A valid brief returns JSON with valid: true and exit code 0. Invalid input returns 1; command-usage mistakes return 2. Missing action text in a video plan produces a warning. Passing validation does not establish provider compatibility.
The four planning modes are text-to-video, image-to-video, text-to-image and image-to-image. A brief needs version: 1, a supported mode and a nonempty subject. Image-to modes also need a reference reminder:
{
"version": 1,
"mode": "image-to-video",
"subject": "The cup in my authorized reference",
"action": "Steam rises gently",
"reference": "my-authorized-cup-photo.png"
}
The reference string is never opened or uploaded. Checking image rights and uploading to an online service are separate manual actions.
Reuse the JavaScript API
Install locally with npm install freeaivideo-prompt-kit@0.1.1, then use an ES module:
import { createExample, validateBrief, toMarkdown } from "freeaivideo-prompt-kit";
const brief = createExample("image-to-image");
brief.reference = "my-authorized-cup-photo.png";
const result = validateBrief(brief);
if (result.valid) console.log(toMarkdown(result.brief));
The library does not read files or access the network. The CLI reads only the brief file or stdin you explicitly name.
Limits and troubleshooting
Text fields are limited to 500 UTF-16 code units, the assembled prompt to 2,000, and CLI input to 64 KiB. Unknown fields are rejected. Model parameters such as duration, resolution, seed and provider credentials are outside this schema. avoid is ordinary prompt text, not an automatically configured negative-prompt parameter.
If parsing fails, check double quotes and commas; JSON has no comments. If a reference reminder is missing, add a string instead of assuming the validator will inspect a photograph. Keep secrets, private links and personal data out of published briefs.
The tool, examples and tutorials are free under MIT. The Free AI Video website is separate: generation requires sign-in, credits and working providers. A standing free signup allowance and all public generation paths have not been confirmed. This exercise needs no purchase or generation task.
Top comments (0)