Turning a Reviewed Shot List into Multishot Video Prompts with a VL Model and a Field Template
kim_deadja4951 · reddit · 2026-09-29
A practical workflow for going from a reviewed shot list to multishot video generation prompts:
- Ling-3.0-flash-VL (via OpenRouter) reads 19 frames sampled every 2 seconds from a 38-second cut, outputting JSON with time ranges, shot sizes, subject/action, setting/lighting, camera movement and transitions.
- After manually reviewing shot order and boundaries against the reference playback, the JSON fields feed a structured template: a Sequence line (total duration + shot count), a carry-through block for subject identity, setting, color/lighting and props consistency, and per-shot Framing/Action/Camera/Transition fields.
- Revisions are field-level: swap Shot 2's action while keeping its duration and framing, or change its camera treatment — producing an inspectable prompt diff before the next generation.
More from coding & agent
- shadcn explains why AI coding tools use horizontal tabs: nav is vertical, the working set is horizontal — xkonjin · 2026-09-29
- "A brief history of bunnies": one-shot short film via Opus 5.5 + Runway MCP — tlakomy · 2026-09-29
- Base44 launches Base Code: non-engineers describe changes in plain English, shipped as GitHub PRs — heyshrutimishra · 2026-09-29
- Dev builds content strategy agent with Hindsight memory layer that learns from past content performance — vikramsaiandra · 2026-09-29
- Agent spins up 322 Hugging Face Jobs in 90 minutes for about $4 — victormustar · 2026-09-29
- Dev builds local Windows voice assistant with 133 tools and hard risk gates the LLM can't bypass — Safe-Cucumber-9316 · 2026-09-29