how we repurposed one video into 10+ pieces of content with higgsfield + claude
what we built (and why we are giving it away)
we kept hitting the same wall every creator hits. one video goes up, it performs, and then you are back to staring at a blank screen figuring out what to post tomorrow. so we built a system that takes one recording and turns it into two weeks of content across every platform - automatically.
the stack is higgsfield for visuals (static ads, hypermotion clips, carousel imagery) and claude for text (tweets, carousels, emails, blog posts). we wired them together into a repeatable workflow, documented the whole thing, and packaged it into downloadable files you can grab below.
this is not theory. this is exactly what we ran. here is a look at how it works.
the output - what one video actually produced
we recorded a single 12-minute video about content repurposing. from that one recording, we extracted: - 4 static product ads (generated via higgsfield from key frames) - 2 hypermotion clips (generated via higgsfield marketing studio) - 4 tweet threads (written by claude from the transcript) - 2 carousel scripts (written by claude, step-by-step format) - 1 newsletter email (written by claude, tightened from the full argument) - 1 long-form blog post (written by claude, SEO-optimized) - 3 short-form video clips (cut from the original)
17 pieces from 1 recording. that covered 10 days of posting across 4 platforms without creating anything new.
how we set it up
we built a project folder in claude code with a specific structure - a data/assets/ folder for reference images, a data/transcripts/ folder for video transcripts, and a .claude/skills/ folder for reusable prompt templates. the setup template (download below) has the exact folder structure.
we connected higgsfield via their CLI (3 commands from higgsfield.ai/mcp), ran the oauth login, and installed the agent skills. the full setup took about 30 minutes. after that, everything runs from prompts.
we pulled the transcript
we dropped the video file into claude code and asked it to transcribe. the transcript came back clean, split by topic. we saved it to data/transcripts/. this raw text became the source for every written piece claude generated later.
you can also use descript or whisper for this step. we used claude code because it was already open and it handled the formatting automatically.
we grabbed key frames
we pulled 4 frames from the video where the product/subject was clearly visible. these went into data/assets/ as reference images.
this is the most important step for higgsfield. every generation needs a reference image - the product has to appear exactly as shown. same label, same color, same everything. we learned this the hard way: skip the reference and the model invents something random. the agency playbook (download below) explains this rule in detail.
we generated visuals with higgsfield
this is where it gets interesting. we used the reference frames to generate platform-specific visuals in three formats:
static ads. we prompted higgsfield to generate 4 instagram-ready ads from our key frames. for each one we specified the angle (curiosity, stat flash, social proof) and the format (9x16 for stories). the skill file (download below) has the exact prompt template we used - it locks in the rules so the product always matches the reference.
hypermotion clips. we used higgsfield marketing studio to generate 6-second product launch clips. fast camera cuts, zoom on product detail, ambient background motion. these became our reels and tiktoks. the prompt structure is in the skill file under "hypermotion video template."
carousel imagery. we generated 3 visually consistent product images for instagram carousels. same subject, different environments. square format.
every generation was tracked in a google sheet - job id, result url, prompt used, status. the workflow file (download below) maps this tracking system end to end.
we generated text content with claude
we fed the transcript to claude and asked it to extract specific formats. here is what we ran:
tweet threads. we asked for 4 threads, each covering one distinct idea from the video. 3-5 tweets per thread, hook in the first tweet, conversational tone. claude pulled the key arguments from the transcript and restructured them for twitter/x.
carousel copy. we asked for 2 carousel scripts. slide 1 = hook, slides 2-8 = steps or points, slide 9 = CTA. one was a step-by-step process, the other was a list of key takeaways.
newsletter email. we asked claude to tighten the core argument into an email: hook, problem, solution, proof, CTA. under 400 words. this went straight into our beehiiv draft.
blog post. SEO-optimized long-form. we gave claude a target keyword and asked for an H1, intro, 3-5 H2 sections, and a conclusion with CTA. 800-1200 words. we published this on our site the same week.
the key insight: we were not asking claude to write finished posts. we were asking for 80%-done drafts. our voice and judgment closed the last 20%.
we scheduled everything
we staggered the posts across 10 days so the single video fueled almost two full weeks:
- day 1 - original video (native platform)
- day 2 - tweet thread #1
- day 3 - instagram static ad #1
- day 4 - carousel #1
- day 5 - hypermotion clip #1 (reels/tiktok)
- day 6 - tweet thread #2
- day 7 - newsletter email
- day 8 - instagram static ad #2 + tweet thread #3
- day 9 - blog post (SEO)
- day 10 - hypermotion clip #2 + carousel #2
buffer handled the scheduling. we batched everything into the queue in one sitting and moved on.
what we learned running this
the skill file is everything. the first batch of higgsfield outputs was inconsistent. some matched the reference, some did not. we reviewed the results, identified what worked, and updated the skill file with tighter rules. the second batch was significantly better. the third batch was dialed in. the skill file (download below) is the version we landed on after multiple iterations.
the reference image rule is non-negotiable. every time we skipped it or forgot to attach it, the output was unusable. the product looked different, the colors were off, the label was wrong. now our skill file has a hard rule: never generate without a reference image.
claude's drafts got better with examples. we started pasting our best-performing posts into the system prompt so claude could match our tone. by the third run, the drafts needed almost no editing.
google sheets as the tracker was the right call. having every generation logged with its prompt, result url, and status meant we could go back and see exactly what produced what. the workflow file (download below) has the full sheet structure.
what we would do differently
we would set up the google sheet tracker before generating anything. we started tracking halfway through and had to backfill. the setup template (download below) has the tracker as step 3 for this reason.
we would also run the 45-minute sprint exercise (download below) first to get comfortable with the workflow before going full scale. we jumped straight into a large batch and the learning curve slowed us down.
the stack we used
- higgsfield - image and video generation (static ads, hypermotion, carousel imagery)
- claude - text generation (tweets, carousels, emails, blog posts, transcripts)
- google sheets - tracker for all generations (planning, status, result URLs)
- descript - transcript extraction from video
- buffer - scheduling across platforms
grab the files
everything we used is downloadable below. the agency playbook has the full 8-part system. the workflow maps every phase. the skill file drops into your .claude/skills/ folder. the setup template gets your project folder ready in 30 minutes. and the exercise gives you a 45-minute sprint to run the whole thing end-to-end with your own product.
this is the exact system we run. not a course, not a teaser - the actual files. take them, use them, make them yours.
downloadable kit
grab the files
free - one email, every guide
read the full guide
we built a system that turns a single video into static ads, hypermotion clips, tweets, carousels, emails, and blog posts. here is exactly how we did it - with every file, prompt, and workflow included so you can do it too.
we only ask once - all guides stay open on this device. no spam.
want a system built for you?
we design and build ai-powered content engines - custom, done-for-you, yours to keep.
let's talk →