Opus Video Prompt Refiner
Rewrite a rough idea into a focused, provider-ready video prompt with a negative prompt and controlled creative variations.

STARTING PROMPT
Reusable prompt
The original English prompt is kept unchanged so you can reproduce the Opus Video Prompt Refiner example.
You are a meticulous video-prompt editor. Rewrite the rough idea below into one provider-ready prompt without losing the creator's intent. Rough idea: [paste a sentence, sketch, or reference description] Work in this order: 1. infer the subject, action, setting, time, mood, and intended audience; mark unknowns as [CHOICE] instead of guessing 2. choose one camera plan and one physically plausible motion path 3. describe lighting, materials, texture, color, depth of field, and motion cadence 4. preserve identity and object continuity across the whole shot 5. provide the final prompt in a compact paragraph, then a short negative prompt 6. list three optional variations that change only one creative variable each Do not add readable text, brand marks, unsupported claims, or a second storyline. Keep the final prompt specific enough to reproduce and short enough for a video model input.
What came back when we ran it
We ran the template once on 2026-10-27 through POST /v1/chat/completions with claude-opus-5-5 at default settings, using the rough idea "a red fox walking through a snowy city street at night". The reply took 31 seconds and 2,114 output tokens.
- It followed the 6 steps in order under headings, starting with a table of the inferred brief.
- It marked 5 unknowns as [CHOICE] (mood, audience, style, duration and aspect ratio, city character) and still picked a default for each, such as 6 seconds at 16:9.
- It chose one camera plan, a low tracking dolly at the fox's eye height with a 50 mm lens, and listed continuity locks such as paw prints that stay put and snow that builds on the fur.
- It ended with one final prompt paragraph, a negative prompt and 3 variations that each change a single variable: camera, snowfall and mood.
Make it yours
- 01
Answer the choices up front
Put the mood, audience, style, length and frame into the rough idea, for example "wry mood, for festival programmers, vertical", so the refiner has fewer [CHOICE] marks to fill with defaults. In our run it marked 5 unknowns and picked a default for each.
- 02
Constrain the camera
Add a camera instruction to the rough idea, for example "handheld at ground level, no dolly", if a low tracking dolly is not the plan. The run chose that move without being asked, and the camera is the first variable a variation changes.
- 03
Ask for one variation axis
Name the single variable to explore, for example "3 variations that change only the snowfall", so the set stays comparable. The run defaulted to camera, snowfall and mood, which suits a pitch but makes a like-for-like comparison harder.
Check before you rely on it
- Thinking cannot be turned off on claude-opus-5-5: a request that disables it, or enables it with a manual token budget, is rejected. Lower output_config.effort when the wait matters, then compare the result with a default run.
- The reply is long: 2,114 output tokens and 31 seconds for one short idea. If you only need the final prompt and the negative prompt, ask the refiner to return step 5 alone.
Run it through the API
Built from the template's model id, prompt and settings. Set OMNIAKEY_API_KEY in your environment first.
curl --fail-with-body https://api.omniakey.com/v1/chat/completions \ -H "Authorization: Bearer $OMNIAKEY_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "claude-opus-5-5", "messages": [ { "role": "user", "content": "You are a meticulous video-prompt editor. Rewrite the rough idea below into one provider-ready prompt without losing the creator'\''s intent.\n\nRough idea: [paste a sentence, sketch, or reference description]\n\nWork in this order:\n1. infer the subject, action, setting, time, mood, and intended audience; mark unknowns as [CHOICE] instead of guessing\n2. choose one camera plan and one physically plausible motion path\n3. describe lighting, materials, texture, color, depth of field, and motion cadence\n4. preserve identity and object continuity across the whole shot\n5. provide the final prompt in a compact paragraph, then a short negative prompt\n6. list three optional variations that change only one creative variable each\n\nDo not add readable text, brand marks, unsupported claims, or a second storyline. Keep the final prompt specific enough to reproduce and short enough for a video model input." } ] }'
Questions
Why does the refiner invent a duration and ratio?
Because the rough idea leaves them open, and the template tells it to mark unknowns as [CHOICE] while still returning a usable prompt. In our run it marked 5 unknowns and picked defaults such as 6 seconds at 16:9. Answer those marks to control them.
Can I run this refiner through the API?
Yes. Send POST /v1/chat/completions with model claude-opus-5-5 and the template plus your rough idea as the user message. The reply arrives as text with headings, a final prompt, a negative prompt and 3 variations. Input and output tokens are billed at the rates on the Claude Opus 5.5 model page.
Can I turn thinking off to save waiting?
No. Thinking cannot be turned off on claude-opus-5-5; a request that disables it, or enables it with a manual token budget, is rejected. The run took 31 seconds with default settings, and output_config.effort is the setting to lower when the wait matters.
More text templates
- GPT-6.1 Creative Brief JSONTurn a campaign request into a strict, editor-friendly JSON brief with timed scenes, continuity, audio, and open questions.
- Gemini Research-to-Video ScriptTurn source notes into a fact-checked 90-second explainer with timed narration, visual beats, and a claim-to-source ledger.
- Opus Video StoryboardTurn a rough brief into a timed four-scene video plan with camera language, continuity constraints, paste-ready prompts, and a global negative prompt.
ONE ACCOUNT · ONE API
Create with OmniaKey
Start with the prompt and settings above, then adapt the idea to your project.