Higgsfield Pipeline

What the source does

• FilesystemNone observedNo reads or writes in the source
• NetworkNone observedNo outbound calls in the source
• ShellNone observedNo commands run in the source
• DependenciesPinnedNothing floats
• SecretsNone foundNo keys in source

Five audit layers

Gates applied: no behavioural pass.

Audit 0.4.1 · source sha 9e3a2f685c72 · 31 Oct 2026Report an issue or request a re-scan

• PublisherOSideMedia GitHub userOwner of the repository this skill is read from
• Source folderOSideMedia/higgsfield-ai-prompt-skill/skills/higgsfield-pipelineThe exact folder we audited
• LicenceMITAs the repository states it
• Commit pinned for this verdict9e3a2f685c72A new audit runs when the source changes
• Repository statusNot archivedLast push 6 days ago
• Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

Its description: Use when the user wants to create a complete multi-shot video, asks how to chain Higgsfield tools together, wants to build a short film or branded content series, asks "what's the full workflow", needs to connect…

---
name: higgsfield-pipeline
description: >
  Use when the user wants to create a complete multi-shot video, asks how to chain
  Higgsfield tools together, wants to build a short film or branded content series,
  asks "what's the full workflow", needs to connect Popcorn → image → video → Recast
  → audio → assembly, or wants to understand how the platform works as a production
  system rather than isolated tools.
user-invocable: true
metadata:
  tags: [higgsfield, pipeline, workflow, chain, production, multi-shot, short-film, popcorn, recast]
  version: 3.6.2
  updated: 2026-09-26
  parent: higgsfield
---

# Higgsfield Production Pipeline

## QUICK FACTS
*Generated-checked block (scripts/build_index.py verifies anchors). Read the linked sections for full context — these lines are routing aids, not the rules themselves.*
- 8-stage Master Chain: Popcorn → Seedream/Soul → Animate → Recast → Lipsync → Vibe Motion → Upscale → Assemble; most good short-form uses 3–5 stages [→](#the-master-production-chain)
- Lock 9 project fields before touching any tool; "what must stay consistent" is the load-bearing one [→](#step-01--start-with-the-project-not-the-prompt)
- One job per scene — six scene purposes; a good scene prompt answers six questions [→](#step-06--give-every-scene-one-job)
- 7 reusable prompt-module types: character identity, camera, lighting, style, motion, negative prompt, continuity [→](#step-07--use-prompt-modules)
- 80% rule: keep what worked, fix only the mistake; every diagnosed failure becomes a new negative rule [→](#step-0809--fix-failures--protect-what-worked)
- Build in 8 passes: Concept → Project script → Scene breakdown → Shot list → Image prompts → Video prompts → Review → Fix [→](#step-10--build-the-project-in-passes)
- Use the EXACT same character description (copy-paste) in every Popcorn prompt — continuity without Soul ID [→](#stage-1--storyboard-with-popcorn)
- Seedream edits the image, not the video — always edit the Hero Frame before animating, never after [→](#stage-2--image-editing-with-seedream)
- Model by scene type: Seedance 2.0 / Minimax Hailuo 2.3 for stunts/epic ("one continuous shot, no cuts"), Kling 2.6 portraits, Seedance quiet interiors [→](#stage-3--animate-by-scene-type)
- Recast swaps identity while preserving motion, camera, and lighting; the "prompt" is the reference image you upload [→](#stage-4--recast-character-swap)
- Audio routing: existing video + speech → Lipsync Studio; new content with audio → Kling 3.0; talking head → Seedance 1.5 Pro (start image + native audio) or Kling Avatars 2.0 (not in the API catalog, 2026-09-26 — verify in the live UI) [→](#stage-5--lipsync--audio)
- Higgsfield has no native timeline editor — assemble in DaVinci Resolve / Premiere / CapCut [→](#stage-8--assembly)
- Pipeline E hard rules: 15-second cap per scene, one generation per style, feed the previous scene's video as continuity reference [→](#stage-5--seedance-20-with-keyframe--previous-video)
- Soul Cinema keyframes: deliberately short 5–15 word prompts with enhancer ON — long prompts starve the enhancer [→](#stage-1--soul-cinema-keyframe-style-first-enhancer-on)
- Never describe character age in Seedance prompts; >15s per scene degrades prompt adherence — split the scene [→](#pipeline-e-pitfalls)
- No extend button: attach the accepted clip as a video reference + open with "The scene continues." — and match the source's resolution AND duration [→](#the-extend-a-clip-workflow)
- Cap seamless extension chains at 2 (hard ceiling 3); re-anchor from ORIGINAL canonical refs, break chains with B-roll [→](#chain-management--depth-caps-and-re-anchoring)
- An attached source clip carries the state — prompt only the delta; motion vectors, camera-move phase, audio phase stay in prose when handing off from a still frame [→](#source-carries-state-rule)
- End extension prompts on a camera-angle change so the join reads as coverage; plan transitions ahead (last-channel-on-TV trick) [→](#clean-join-planning)
- Draw a top-down schema when 2+ characters, a key prop placement, or complex camera geometry — prompt in absolute terms ("A 2m from B") [→](#spatial-blocking--top-down-schema-for-multi-character-scenes)
- Never animate a "good enough" image; if the character looks wrong in the Hero Frame, Recast is the fix — not the animation prompt [→](#pipeline-pitfalls)
- [FIELD] The edit is a loop with a declared exit: assembly → rough cut → **generation supervision** (re-generate broken shots here, and only here) → fine cut → picture lock, after which there are no new generations; colour's first job is unifying each generation's baked-in grade [→](#the-edit--five-stages-to-picture-lock)


## The Core Insight

Every Higgsfield tool is strong individually. The real power is when you chain them.
A professional result almost always involves at least 3 tools in sequence.
This skill documents the key chains and how to prompt for each stage.

---

## Building Complete AI Projects — The 10-Step Methodology

The Core Insight above answers *why* you chain tools. The next
question is *how to plan the project* before any tool is touched.
Most people skip planning and start prompting straight from a
creative impulse — a generation here, a clip there, hoping the
pieces will connect. They don't. A project that's planned at the
script and bible level lands consistently; a project assembled
prompt-by-prompt drifts on character, style, and continuity within
three or four generations.

This section documents a 10-step methodology for building complete
AI projects — from idea through finished sequence — that sits
upstream of the Master Production Chain below. Use these 10 Steps
as a planning discipline; use Pipelines A-E to execute the
tool-level chain once your project plan is locked. The closing §
Simple Workflow gives the imperative-action version of the same 10
Steps if you want the execution recipe before the principles.

### Step 01 — Start With the Project, Not the Prompt

Before writing a si

Excerpt, first 6,000 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

1. 2026-10-31
   ✓SAFE B 89 / 100

   first audit · source 9e3a2f685c72

Questions about Higgsfield Pipeline

• github.com/OSideMedia/higgsfield-ai-prompt-skill/tree/main/skill — the audited source at commit 9e3a2f685c72 and every finding aboveObserved · 31 Oct 2026

Spotted something wrong? Report a correction · Methodology