Ingest
Videos or a whole library
NVIDIA Crushes Quarter | OpenAI Cuts Off Cursor | Instinct Hits $2.5B Valuation
Build the edit once.
Run it on every episode.
By approval Workflows are approved team by team. Talk to us and we set them up with you.
Write your production as steps. A model analyzes each new video and picks the moments. A clone step gives every clip its own branch, which edits, checks, reframes and adds motion graphics on its own.
The real run made 8, chosen by the model. Selection picks 3 to 20.
A new 20VC episode is ingested. Analysis reads the whole 1:20:29 episode, and Selection picks 8 moments and saves clip_count. A clone step splits the run into 8 branches. Each branch edits its clip, checks it with Auto QA, which can send it back to Edit once, reframes it to 9:16 and adds motion graphics. The finished clips post to a callback URL for review. The clips shown are the real output of a run on 29 September 2026.
-
-
Analysis
Reads every word, speaker and shot
-
Selection
8 moments
- NVIDIA’s 70% guidance shock and the hyperscaler cash-flow squeeze
- Sam vs Elon: why the OpenAI–Cursor split was made for TV
- +6 more
clip_count = 8 -
Clone
8
One branch per moment
Clip 1: NVIDIA’s 70% guidance shock and the hyperscaler cash-flow squeeze
Clip 2: Sam vs Elon: why the OpenAI–Cursor split was made for TV
Clip 3: The AI agent that almost bought the team engraved Pateks
Clip 4: Jason Lemkin’s Clay reversal: “our agents will only use Clay”
Clip 5: Jason Lemkin on reward hacking and the $5,000 West End ticket
Clip 6: Rory O’Driscoll on why it’s time to panic about cyber
Clip 7: Jason Lemkin: never anthropomorphize AI agents
Clip 8: Jason Lemkin: why every startup must become a compound startup
Done
8 clips post to your callbackUrl with titles and download links, ready for review.
2,000 edits a month.
Or more than 50,000.
Every new video starts its own run of the same workflow, so the rules, the look and the review stay the same at any volume. We connect your sources: a YouTube channel notifies StarZero the moment a video goes live, and social accounts, storage buckets or web pages can feed in the same way. Each new video is ingested and starts a run on its own. Your own pipeline can start runs through the REST API, which we set up on request.
75 workflows × 8 clips = 600 edits a week
2,600 edits a month
Eight clips from one episode.
No one opened an editor.
The 20VC episode in the run above went through the Podcast Clips template. The model picked 8 moments, and each branch turned the three-person grid into a 9:16 clip with speaker framing, captions and a title. All 8 are here.
NVIDIA’s 70% guidance shock and the hyperscaler cash-flow squeeze Straight from the run above, with no edits after render.
From one 20VC episode
The same template on 8 episodes
69 clips. On each episode the model decided how many moments were worth cutting, from 6 to 12, and the clone step made that many branches.
-
NVIDIA Crushes Quarter | OpenAI Cuts Off Cursor
8 clips
-
-
ElevenLabs: Building an AI Sales Machine
12 clips
-
-
Coinbase Cuts AI Spend by 50% | Kalshi’s $40B Valuation
6 clips
-
-
How I Plan, Build, and Run Loops with Claude Code · Thariq Shihipar
6 clips
-
-
Grant Sanderson (3Blue1Brown): AI disproved a famous math conjecture
9 clips
-
-
Training Sand to Think: AGI & the Future of Physics
9 clips
-
-
NVIDIA GTC Keynote 2026
9 clips
-
-
From Revolut employee #3 to a £5B energy company · Alan Chang
10 clips
-
Steps you can read.
Variables you can change.
A template is an ordered list of steps that share one agent session and a set of variables. A task can save a variable, any later step can read it, and a clone step turns a number into branches. Here is a clipping template, written out.
-
setAgentSettingsShow rules
Give every agent the show’s rules once: caption treatment, framing, tone, the words to spell right.
caption_styleuser_notes -
conditionalTaskFix the transcript
If corrections are on, correct names and terms before anything is cut.
correct_transcript -
taskFind the moments
Read the whole episode, choose every moment worth a clip, and save how many there are.
clip_count -
cloneFan out
Split the run into one branch per clip. Each branch carries its own index and the same session history.
clip_countindex -
taskCut one clip
Cut the moment for this branch, reframe it to the aspect ratio, caption it, write the title and render.
indexaspect_ratio_numaspect_ratio_denpacing -
conditionalTaskScore it
If music is on, add a bed under the speech and render again. Otherwise keep the first render.
add_music -
callbackUrlDeliver
When the last branch finishes, the videos, their titles and every branch’s variables post to your endpoint.
results.video
- Branch on any variable
- conditionalTask and switchTemplate compare a variable with eq, contains, gte and more, so interviews and keynotes can take different steps or hand off to another template.
- Budget every step
- Set a credit cap per step. A step that runs over is stopped, and the other branches keep going.
- Review before it counts
- A critique agent scores each render against the rubric you write. Edits that miss the mark go back for another pass.
- Change templates safely
- Every edit is versioned, the last 50 are kept, and runs already in flight finish with the steps they started with.
Start runs from your stack.
Get every video back.
Trigger a run from your CMS, your ingest pipeline or an assistant. When it finishes, StarZero posts the videos, titles and variables to your callback URL.
By approval Workflows are approved team by team. The REST API is set up on request.
Talk to salesWorkflow API
Start one run per new video, or pass a list to POST /instances/bulk and start a run for each. Poll the run, or wait for the callback.
The REST API is set up on request. Talk to sales
Download links in the callback stay valid for 7 days.
POST /instances
{
"templateId": "<your template>",
"libraryId": "<your library>",
"selectedContentIds": ["<the new video>"],
"variables": {
"clip_count": 8,
"aspect_ratio_num": 9,
"aspect_ratio_den": 16,
"add_music": false
},
"callbackUrl": "https://example.com/hooks/clips"
}{
"instanceId": "…",
"results": {
"video": [
{
"title": "NVIDIA’s 70% guidance shock and the hyperscaler cash-flow squeeze",
"video_url": "https://…"
},
…
],
"variables": [
{ "clip_count": 8, "index": 0 },
…
]
}
}Remote MCP
Connect an MCP client to StarZero and sign in with your StarZero account. Your assistant can list your templates and start a run with the variables you give it.
https://mcp.starzero.ai/mcp
template_listLists the workflow templates you can run.instance_createStarts a run of a template on one of your libraries.
StarZero CLI
Sign in with your StarZero account, then describe a template to see the variables it expects, start a run, and wait for it from a terminal or a scheduled job.
starzero workflow template list
starzero workflow template describe <templateId>
starzero workflow instance create --template <templateId> \
--library <libraryId> --variables vars.json
starzero workflow instance watch <instanceId>
starzero workflow instance get <instanceId> By approval Bring a show and a brief.
Bring a show and a brief.
We’ll scope the run with you.
Workflows are approved team by team. Tell us where your videos come from, how many arrive each week, what each one should become and where the results need to go.