# Higgsfield Storyboard Generation

> End-to-end storyboard-to-image production for Higgsfield. Use when the user gives a storyboard, conte, shot list, Figma board, screenshot, or cut list and wants Codex to create cut-by-cut prompts, run logged-in Higgsfield browser generation, choose quality-first models such as GPT Image 2 or Nano Banana Pro, generate two or more variants per cut, download 4K outputs, build contact sheets, select the best frame per cut, update logs, and package final deliverables.

- Skill: `godhonggi-cmd/higgsfield-storyboard-generation` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add godhonggi-cmd/higgsfield-storyboard-generation`
- Raw SKILL.md: https://api.skillmd.com/api/skills/godhonggi-cmd/higgsfield-storyboard-generation/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Design & Media
- Author: godhonggi-cmd (https://skillmd.com/u/godhonggi-cmd)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/godhonggi-cmd/higgsfield-storyboard-generation

---


# Higgsfield Storyboard Generation

## Overview

Run a controlled Higgsfield storyboard image batch from raw conte to delivery: normalize cuts, write prompts, set generation options, queue variants in the browser, download finished images, create contact sheets, choose selected frames, and return a ZIP plus logs.

Never ask for or store passwords, session cookies, recovery codes, or payment details. Ask the user to complete login, 2FA, captcha, and payment confirmations directly in the browser. Do not bypass site protections. Treat each Generate click as credit-spending: proceed only within the user's explicit scope.

## Default Decisions

- Quality-first still frames: use `GPT Image 2`, `High`, `4K`, requested aspect ratio, and 2 variants per cut unless the user chooses otherwise.
- Fashion/cinematic look with no exact text requirement: consider `Higgsfield Soul Cinema` or `Higgsfield Soul 2.0`.
- Faster or lower-cost broad exploration: consider `Nano Banana Pro`, `Nano Banana 2`, `Auto`, or other Unlimited models when visible.
- For presentation storyboards, prefer consistency and readable composition over novelty.

## Workflow

1. Normalize the storyboard.
   - Extract one row per cut with: cut id, story beat, subject, action, location, wardrobe/props, camera/framing, lens/motion, mood, lighting, style references, aspect ratio, and constraints.
   - Default to 2 variants per cut when the user asks for "two each" or does not specify.
   - Preserve character, product, wardrobe, location, color, and art-direction anchors across prompts.

2. Prepare the batch workspace.
   - Use `scripts/prepare_storyboard_batch.py` for CSV/TSV/JSON storyboards or when a reusable folder is useful.
   - Create outputs under the task `outputs/` folder unless the user requests another destination.
   - Use `generation_queue.csv`, `selection_log.csv`, `cuts/<cut_id>/variants/`, `selected/`, and `contact_sheets/`.

3. Write prompts.
   - Produce a prompt table before generation for non-trivial batches.
   - Use one prompt per cut when Higgsfield quantity is set to 2; do not manually submit duplicate A/B prompts unless the user wants two different creative directions.
   - Add an avoid clause when no negative prompt field is available: `Avoid deformed hands, extra fingers, duplicate limbs, warped faces, unreadable text, distorted logos, inconsistent wardrobe...`

4. Run Higgsfield in the browser.
   - Before browser work, read `references/higgsfield-browser-automation.md`.
   - Ask the user to log in directly. Pause for login, 2FA, captcha, payment, or unexpected credit confirmation.
   - Set model, quality, resolution, aspect ratio, and quantity from the run scope. If the cost indicator materially changes after a setting change, report it before spending beyond the user's latest instruction.
   - Queue each cut, wait for completion, and build `download_manifest.json` from the generated image URLs.

5. Download and organize.
   - Run `scripts/finalize_higgsfield_batch.ps1 -BatchRoot <batch-root>` after `download_manifest.json` exists.
   - If downloads fail because of network sandboxing, rerun the same command with approval.
   - Verify image dimensions and keep original variants.

6. Select best frames.
   - Review `contact_sheets/generated_contact_sheet.png` and local images.
   - Read `references/quality-rubric.md` when the batch is large, visual continuity matters, or selection is contested.
   - Update `selection_log.csv`, copy winners to `selected/`, and make `contact_sheets/selected_contact_sheet.png`.

7. Deliver.
   - Return selected images, contact sheets, selection log, and ZIP.
   - Use predictable names: `cut_001_A.png`, `cut_001_B.png`, `cut_001_selected.png`.
   - Mark a cut `needs_regen` if no variant is acceptable.

## Scripts

Create a batch:

```bash
python scripts/prepare_storyboard_batch.py --storyboard storyboard.csv --out outputs/higgsfield-batch --project-name client-film --variants 2
python scripts/prepare_storyboard_batch.py --template outputs/storyboard_template.csv
```

Finalize generated files after browser URL extraction:

```powershell
powershell -ExecutionPolicy Bypass -File scripts/finalize_higgsfield_batch.ps1 -BatchRoot outputs/higgsfield-batch/client-film
```