20 Videos in One Evening. Built on Competitor Data.

Higgsfield × Meta × Claude

Claude analyzes what's taking off for your competitors, writes a brief for 20 videos, Higgsfield generates them, Meta feeds back the metrics, and Claude optimizes the next batch. The full loop, with no filming and no videographers.

The fastest route: no terminal required

Higgsfield inside Claude in 4 steps.

Don't want to mess with API keys and .env files? Hook up Higgsfield as a custom connector right in the Claude interface. You write the script in a regular chat, and the video lands right there too.

01 — Open Connectors In the Claude desktop app or on claude.ai: Settings → Connectors. This is where you plug Claude into external services. → Settings → Connectors

02 — Add a custom connector Click "Add custom connector". Name it Higgsfield. Paste the server URL below. Hit Connect and sign in to your Higgsfield account. → https://mcp.higgsfield.ai/mcp

03 — Write your script in the chat In a normal conversation, describe the video: hook, topic, style, length. You can ask for a whole series at once, and Claude will split it into separate generations. → "Make 3 cinematic 15-second Reels about AI agents"

04 — Claude delivers the videos Claude calls Higgsfield, waits for the render, and drops the finished MP4s right into the chat. No tab-switching. → Rendering a 15-sec clip ≈ 60-90 seconds

When to use this route vs. the full pipeline The connector in Claude is for hands-on work and one-off generations: testing an idea, picking a style, making 1-3 videos. If you need a batch of 20+ videos with competitor analysis, metrics, and auto-publishing, keep reading for the Claude Code + Higgsfield API part of this guide.

What the reel shows and how it works

The reel shows the finished pipeline. Here's how to build it.

The reel shows the end result: Claude Code + a Higgsfield connector + a Meta connector, one task, 20 finished videos. It genuinely works, but it takes a one-time setup of three components. Here's exactly how.

01 — Higgsfield connector In Higgsfield: Settings → API Keys → create a key. In Claude Code, add it to .env: HIGGSFIELD_API_KEY=hf_... The /smm skill picks up the key and talks to the API directly. higgsfield.ai → Settings → API

02 — Meta connector You need an Instagram Business account and a Meta Developer App with the instagram_content_publish permission. Drop the token into .env: INSTAGRAM_PAGE_ACCESS_TOKEN=... From there, Claude Code publishes Reels through that token. developers.facebook.com → My Apps

03 — One command Once set up: type the task to Claude Code in your terminal. It pulls competitor data through ScrapeCreators, writes the brief, sends it to the Higgsfield API, waits for the videos (2-3 hrs), and publishes to Instagram.

The .env file: the three keys you need

`` HIGGSFIELD_API_KEY=hf_xxxxxxxxxxxxxxxx INSTAGRAM_BUSINESS_ACCOUNT_ID=1784xxxxxxxxxxxxxxx INSTAGRAM_PAGE_ACCESS_TOKEN=EAAxxxxxxxx... SCRAPECREATORS_API_KEY=sc_xxxxxxxx ``

Higgsfield: higgsfield.ai → Settings → API Keys. Meta token: developers.facebook.com → My Apps → create an App → add Instagram → generate a token. Takes 5 minutes.

How the stack fits together

Three tools. One loop.

Claude *Strategist and analyst* - Scrapes competitors via ScrapeCreators - Spots the patterns in videos that took off - Writes the brief for 20 videos - Analyzes the metrics and optimizes the next batch

Higgsfield AI *Video generator* - Generates videos from a text brief - A batch of 20 clips in 2-3 hours - Styles: cinematic, product B-roll, avatar, cinematic intro - MP4 export in Reels / TikTok format

Meta Graph API *Metrics and publishing* - Publishes Reels via the API - Metrics: views, reach, saves, shares, ER - Publishing schedule - Competitor analytics (via Instagram Scraper)

The full workflow: 6 phases

From zero to 20 published Reels.

01 — Collect competitors Find 10-20 accounts in your niche whose Reels are growing. Not the biggest ones, but the ones going viral right now. ScrapeCreators pulls the top posts for a keyword over the last 7-30 days. → ScrapeCreators API → query = your niche + "reels" → sort by views → top 50 posts

02 — Claude analyzes the patterns Feed Claude the competitor data: video title, view count, likes, saves, publish date. It finds the patterns: which hooks, topics, lengths, and styles pull the most engagement. → Prompt below → you get a structured analysis with concrete takeaways

03 — Claude writes the brief for 20 videos Based on the patterns: 20 production specs for Higgsfield AI. Each one with a hook, topic, visual style, and CTA. Not abstract ideas, but concrete generation instructions. → Brief format: number, hook (first 3 sec), topic, Higgsfield style, CTA, hashtags

04 — Higgsfield generates the videos Go to Higgsfield AI, create a project, paste in the brief. Higgsfield takes your prompts and generates the clips: cinematic, product B-roll, AI avatar, or cinematic intro. A batch of 20 takes 2-3 hours with zero involvement from you. → higgsfield.ai → New Project → paste Claude's brief → pick a style → Generate Batch

05 — Publishing via n8n Download the finished MP4s from Higgsfield and load them into an n8n queue. Auto-posting on a schedule: 2-3 videos a day, at peak hours. The Meta Graph API accepts Reels directly. → Optimal schedule: Mon/Wed/Fri at 10:00, 13:00, 19:00 in your audience's time zone

06 — Claude analyzes the metrics → next batch Two weeks later, pull the metrics from the Meta API or Business Suite. Feed them to Claude: it compares the top 5 against the flops and extracts the winners' patterns. Then it writes an updated brief for the next 20 videos. The loop repeats. → Metrics: views, reach, saves (the most important one!), shares, ER = (likes+comments+saves)/reach

Prompts: copy and use

Three prompts that kick off the loop.

1. Competitor analysis

Prompt → Claude

``` Here's data on competitor Reels in my niche [PASTE: account name, video topic, views, likes, comments, saves, date].

Analyze: 1. Top 5 hooks by views: what do the first 3 seconds have in common? 2. Which topics pull the most saves (that's the value signal)? 3. Patterns in the CTAs: what provokes comments? 4. Optimal video length: short (7-15 sec) or long (30-60 sec)? 5. Top 3 visual styles that keep showing up among the leaders.

Give me a structured report with concrete takeaways, no abstractions. ```

2. Brief for 20 Higgsfield videos

Prompt → Claude

``` Based on the competitor analysis above, write a brief for 20 videos for Higgsfield AI.

For each video, specify: - Number (01-20) - Hook: the first 3 seconds, as specific text or action (nothing abstract) - Topic and the main point (1 sentence) - Generation style: pick from [cinematic / product-broll / ai-avatar / talking-head / text-motion] - Closing CTA (5-7 words) - 5 hashtags

Priority: replicate the patterns from the top 5 competitors, don't invent from scratch. Niche: [PASTE YOUR NICHE] ```

3. Metrics analysis → batch optimization

Prompt → Claude (2 weeks after publishing)

``` Here are the metrics for 20 Reels over 2 weeks [PASTE: video number, topic, views, reach, likes, comments, saves, shares].

Analyze: 1. Top 5 by engagement rate (ER = (likes+comments+saves) / reach × 100) 2. What do the winners have in common? Hook? Topic? Length? Style? 3. What flopped (ER < 2%) and why? 4. Saves leaders as a separate list: these are the topics people actually need

Deliver: - 10 formats for the next batch (with reasoning) - 5 hypotheses to test - What to cut entirely from the next iteration ```

Higgsfield styles

Which style for which job.

Higgsfield supports several generation modes. The style you pick in the brief is the key parameter.

Metrics: what to check and where

The data that actually matters.

Most people watch view counts. Saves are the real value signal. Here's the full metric set and where to get each one.

How to pull metrics via the Meta Graph API

``` GET https://graph.facebook.com/v19.0/{media_id}/insights ?metric=reach,impressions,saved,shares,video_views &access_token={PAGE_ACCESS_TOKEN}

→ media_id comes from the publish response (ig_media_id) → PAGE_ACCESS_TOKEN comes from the Meta Developer Console → Data becomes available ~6 hours after publishing ```

The optimization loop

A new batch every 2 weeks. Each one better than the last.

Weeks 1-2 1. Competitor analysis via ScrapeCreators 2. Claude writes the brief for 20 videos 3. Higgsfield generates the batch 4. n8n publishes on schedule (2-3 Reels a day)

Weeks 3-4 (next cycle) 1. Pull the metrics via the Meta API 2. Claude analyzes the top 5 and the flops 3. Updated brief: double down on the winners 4. The next 20 videos, now backed by data

What happens after 3 months After 6 batches of 20 videos, you have 120 published Reels and 6 optimization iterations. By that point Claude knows exactly what works in your niche, not from theory but from your own real data. ER climbs iteration after iteration because every new batch is built on the previous batch's winners.

Up and running in one evening

Minimal setup.

1. Sign up at higgsfield.ai (there's a free trial)
2. Connect ScrapeCreators: 100 free credits, enough to get started
3. Find 10 competitors in your niche via Instagram hashtag search
4. Paste their Reels URLs into ScrapeCreators → get the metrics as JSON
5. Copy Prompt #1 above → paste in the data → let Claude analyze it
6. Copy Prompt #2 → get a 20-video brief in 2 minutes
7. Paste the brief into Higgsfield → pick a style → start the generation
8. While Higgsfield generates (2-3 hrs), set up n8n for publishing

Stack cost

Total: ~$59-79/mo for the full loop. One viral Reel pays for the stack within a few weeks.