AI Tools Alat telah diperbarui 2 hari yang lalu

JSON Prompt Generator

Build a validated video prompt as JSON in a guided 5-step wizard. Pick a starting template, choose a model (Veo, Sora, Kling, Runway Gen-4, Luma Ray 3, or Hailuo), describe the scene, tune the cinematography with clickable chips, and copy a ready-to-paste JSON object. The schema and model limits are fully configurable, and a built-in linter checks any pasted JSON for you.

Selalu Gratis Mudah digunakan Hasil seketika Pribadi dan aman

Gunakan JSON Prompt Generator

How this works — this builder is deterministic: it assembles a prompt from your fields, it does not render or generate video. Copy the JSON into a video model UI (Veo, Sora, Kling, Runway, Luma, Hailuo) or any prompt-to-video tool. Drafts stay in your browser (localStorage) — nothing is uploaded.
Build a video prompt
Select a model
  1. 1 Mulai
  2. 2 AI model
  3. 3 Scene
  4. 4 Gaya
  5. 5 Review

Pick a starting point, or start blank and describe your own shot. Everything can be changed later.

Pick the model that will render your video. Limits below follow the selected model.

Who or what the camera is on. Concrete nouns beat abstractions.
What the subject is doing, with a clear arc the model can render.
Where the action happens - time, place, grounding details.
Spoken line, ambient bed, or SFX. Veo and Sora generate audio; other models ignore it.
Your selection
Prompt JSON
Incomplete
Validate JSON
Lint

Paste a JSON prompt to lint it against the schema for the selected model.

Schema · 2026-05
Reference
Key Jenis Diperlukan Options / notes
subject string * Who or what the camera is on. Concrete nouns beat abstractions.
action string * What the subject is doing, with a clear arc the model can render.
environment string opsional Where the action happens - time, place, grounding details.
camera enum opsional locked-off static shot slow push-in slow pull-back tracking shot slow pan slow tilt up dolly in dolly zoom vertigo effect smooth orbit around the subject crane up handheld shot aerial drone shot whip pan slow zoom in rack focus
lens enum opsional portrait 85mm lens 50mm lens 35mm lens 24mm wide-angle lens 100mm macro lens anamorphic lens 200mm telephoto lens
lighting enum opsional golden hour light soft diffused light hard sunlight rim light low-key lighting high-key lighting volumetric light neon light warm practical light overcast light backlit silhouette
grade enum opsional teal and orange grade cinematic grade natural grade desaturated grade warm grade cool grade noir black and white grade vintage film grade pastel grade
mood enum opsional cinematic energetic serene dreamy moody playful epic intimate nostalgic surreal
audio string opsional Spoken line, ambient bed, or SFX. Veo and Sora generate audio; other models ignore it.
duration number *
aspectRatio enum * 16:9 9:16 1:1 4:3 3:4 4:5 21:9
Model limits
Model Audio Durasi Aspect ratios Note
Kling AI Tidak 1–15s (5 / 10 / 15 seconds) 16:9, 9:16, 1:1, 4:3, 3:4, 21:9 Kling 3.0: kinetic motion, strong physics, per-character lip-sync (Omni). Honors camera and motion; audio ignored.
Veo Ya 1–8s (4, 6, or 8 seconds) 16:9, 9:16 Veo 3.1: native synchronized audio and lip-sync. Strongest prompt fidelity on the Google pipeline (Gemini / Vertex AI).
Sora Ya 1–60s (up to 60 seconds) 16:9, 9:16, 1:1, 4:5 Sora 2: best-in-class prompt fidelity and native audio. Consumer app closed Apr 2026; API sunsets Sep 24, 2026.
Runway Gen-4 Tidak 5–10s (5 or 10 seconds) 16:9, 9:16, 1:1, 4:3, 3:4, 21:9 Gen-4 / Gen-4.5: character references and strong camera control. Audio added in post.
Luma Ray 3 Tidak 5–10s (5 or 10 seconds) 16:9, 9:16, 1:1, 4:3, 3:4, 21:9 Dream Machine Ray 3: HDR/EXR output, seamless loops, character references. No native audio.
Hailuo (MiniMax) Tidak 5–10s (5 or 10 seconds) 16:9, 9:16, 1:1 Fast and low-cost, excels at high-motion action. No native audio.

Tautan Alat

Telah disalin ke papan klip.

Bagikan

Facebook Twitter WhatsApp Pinterest LinkedIn Telegram Reddit Tumblr VK Baris

Bantu Perbaiki Alat Ini

Laporan Permohonan Komentar

Baca Isi

Pelajari cara kerja alat ini dan kapan harus menggunakannya.

How to Use This Tool

Follow the 5-step wizard and copy the generated JSON into Veo, Sora, Kling, Runway Gen-4, Luma Ray 3, Hailuo, or any prompt-to-video tool. The builder is deterministic: it assembles your inputs into JSON and does not render video.

1. Start

Pick a starting template (Cinematic, Product / ad, Action, or Dreamy) or start blank. Templates pre-fill the style choices so you can jump straight to describing your shot.

2. Select a model

Choose Veo, Sora, Kling, Runway Gen-4, Luma Ray 3, or Hailuo. The wizard adapts instantly: the duration range, allowed aspect ratios, and audio field visibility all follow the selected model. Veo and Sora generate native audio; the other models ignore the audio field but honor camera and motion.

3. Describe the scene

Fill the subject (who or what the camera is on) and the action (what it is doing, with a clear arc). Add an environment for time and place. These are the core of the prompt.

4. Style the shot

Tap chips to pick camera movement, lens, lighting, color grade, and mood, then set the duration. Every option value is validated against the schema and filtered to what the selected model supports.

5. Review and copy

The output updates live and only includes keys with values, in schema order: subject, action, environment, camera, lens, lighting, grade, mood, audio, duration, aspectRatio. Copy it to your clipboard or download it as a .json file.

Validating JSON

Paste any JSON prompt into the Validate JSON section and pick a model. The linter reports missing required keys, unknown keys, invalid enum values, out-of-range durations, and aspect ratios the selected model does not support.

Features

  • Guided 5-step wizard with a progress bar
  • Starting templates plus an example shot
  • Clickable chip selectors instead of plain dropdowns
  • Configurable schema, templates, and model limits from a single config file
  • Per-model duration clamping and aspect ratio filtering
  • Built-in JSON linter with per-model checks
  • Bundled schema reference table
  • Draft auto-save in the browser (localStorage)
  • Copy JSON, download .json, and Ctrl/Cmd+Enter to copy

Pertanyaan yang Sering Diajukan

Jawaban singkat atas pertanyaan-pertanyaan yang diajukan pengguna sebelum mempercayai hasilnya.

The output uses these keys in order: subject, action, environment, camera, lens, lighting, grade, mood, audio (Veo/Sora only), duration, and aspectRatio. Required keys are subject, action, duration, and aspectRatio.

No. It is a deterministic builder that arranges your inputs into JSON. Paste the output into a video model UI such as Veo, Sora, Kling, Runway, Luma, or Hailuo to generate the actual video.

Only Veo and Sora generate native audio, so the audio field is hidden for Kling, Runway Gen-4, Luma Ray 3, and Hailuo. The linter still warns if you paste JSON that includes an audio key while linting against those models.

Veo: 1–8s at 16:9 or 9:16. Sora: 1–60s at 16:9, 9:16, 1:1, or 4:5. Kling: 1–15s at six ratios. Runway Gen-4 and Luma Ray 3: 5–10s at six ratios. Hailuo: 5–10s at 16:9, 9:16, or 1:1. The schema table shows the exact limits.

No. Everything runs in your browser. Your draft is saved only in your browser's local storage and is never transmitted anywhere.

Yes. Models, templates, durations, aspect ratios, and enum options are defined in config/json-prompt-generator.php. The tool reads everything from that single config file.

Komentar (0)

Untuk diskusi pengguna, kasus-kasus khusus, dan tips lanjutan.
Tinggalkan komentar
Komentar Anda akan ditampilkan secara publik setelah dikirim.
Belum ada komentar. Jadilah yang pertama memberikan komentar!
Beritahu Kami!

Laporkan masalah terkait alat ini

Jelaskan masalah yang Anda alami agar kami dapat menyelidikinya dan meningkatkan alat ini.

Apa yang paling membantu
Sertakan data masukan, hasil yang diharapkan, hasil yang sebenarnya, browser/perangkat, serta tangkapan layar jika diperlukan.
Tombol tangkapan layar akan mengambil gambar halaman di peramban dan melampirkan gambar yang dihasilkan ke formulir ini.
Tangkapan layar satu halaman penuh
Dibuat di peramban Anda menggunakan html2canvas.
Belum ada tangkapan layar yang diambil.
Alat Laporan