Free preview — AI Video & Shortform Production SOP Pack (v1.2)

no signup · full page below · download as markdown

This preview shows the operating style of the full pack without giving away the production system. You get the full table of contents, a one-paragraph summary of the pre-production script gate with one worked example check, one complete prompt from the cookbook, and a look at what the paid pack adds. This preview is roughly 8–12% of the pack's total content — enough to judge the voice and rigor, not enough to run a production on.

Preview scope: a small slice. The full pack includes the complete SOP (including a worked example that runs one still image through every ffmpeg stage to a shipped clip), setup checklist, full 12-prompt cookbook, QA receipt template, rights/provenance checks, known pitfalls, a runnable ffmpeg toolkit, a ComfyUI image-to-video workflow, and the QA receipts from a real render built with the worked example's workflow.

Full pack table of contents

README

Overview · What you get · Why this instead of a free guide or ChatGPT · Buyer-facing files · Start here · Glossary · Who it's for · What it's not · Prerequisites · Core promise · License notes

SOP

• 0. Operating principles
• 1. Production lanes (cinematic I2V intro · 9:16 narrated short · licensed-footage short · talking-character clip)
• 2. Pre-production script gate
• 3. Source image and keyframe rules
• 4. Image-to-video shot recipe
• 5. Assembly grammar
• 6. Audio SOP
• 7. Subtitle SOP
• 8. QA gate
• 9. Delivery receipt template
• 9.5. Worked example: still image → shipped vertical short (real ffmpeg command lines)
• 10. Known pitfalls

Setup Checklist

Workstation · CLI tools · ComfyUI / I2V lane · image generation · audio & TTS · fonts & subtitles · rights & provenance · QA readiness · first smoke test

Prompt Cookbook (12 templates)

concept · script rewrite gate · 9:16 hero still · multi-shot consistency · I2V positive · I2V negative · cinematic intro shot list · narration · subtitle chunking · licensed-footage search · QA review · delivery receipt

ffmpeg toolkit (runnable)

loudnorm 2-pass · Ken Burns vertical · blurred-bg 9:16 fit · frame-accurate cut · black/silence QA · subtitle burn-in · drawtext hook/caption

comfyui/

minimal API-format image-to-video workflow (Wan2.2 I2V + LightX2V 4-step) + setup README

examples/

QA receipts from a real render built with the SOP's worked example workflow — ffprobe_receipt.json, loudnorm_receipt.json, qa_black_silence.txt.

Sample: the pre-production script gate, and one of its six checks

Before rendering anything, a script has to clear a six-check gate — first sentence works without context, conclusion-first structure, an explicit emotional target, a repetition scan, an ending with a point, and an AI-slop blacklist. It ships only if all six pass. Here's the gate's easiest check as an example of the bar the other five hold to:

Example check — First sentence works without context

The first line should make the topic or tension clear within the first few seconds. If a viewer needs a long setup before they know why the video matters, the opening is not ready.

Sample: one complete cookbook prompt

9:16 hero still prompt

A high-quality 9:16 vertical image for a shortform video about [topic].
Subject: [main subject].
Composition: centered readable subject, strong silhouette, no clutter, safe empty space for subtitles in the middle band.
Style: [documentary / cinematic / scientific macro / illustrated editorial / premium 3D].
Lighting: [lighting direction and mood].
Color palette: [palette].
Camera: [lens/framing].
No text, no watermark, no logos, no UI, no distorted anatomy, no extra limbs.

The other 11 cookbook prompts — script rewrite gate, multi-shot consistency, I2V positive/ negative, cinematic intro shot list, narration, subtitle chunking, licensed-footage search, QA review, and delivery receipt — are in the paid pack.

What the paid pack does that a free blog post won't

• The full six-check script gate plus a rewrite-gate prompt that returns PASS/FAIL + blockers + a revised script, so the gate runs in one paste instead of by memory.
• Runnable ffmpeg scripts, not descriptions. Five were checked on ffmpeg 8.1 — e.g. a normalized clip measured back at −13.95 LUFS, and the QA script correctly flagged an injected black + silent gap.
• A worked example (SOP §9.5) that chains those scripts end-to-end on one still image, with the real command lines, plus the QA receipts (examples/) from a real render built with that workflow — proof the pipeline was actually exercised, not just described.
• The pitfalls that cost real time: the zoompan d=1 one-frame-freeze, keyframe-only cuts landing on the wrong frame, subtitles buried under the app UI, and black/silent tails that pass a casual watch but fail a QA scan.
• A ComfyUI I2V workflow with the exact model files, download links, VRAM guidance, and a low-spec (SVD) fallback — so setup is a checklist, not a scavenger hunt.

Who this is not for

If you only want model weights, a one-click finished result, or sample media to repost, this is not that product. It's a production standard plus the commands, the graph, and the receipts proving the commands were run.