# Auto-Clips: Long Video to Vertical Shorts Point Auto-Clips at any long video and it finds the most engaging moments for you, ranked, titled, and scored. It then cuts, reframes, captions, voices, and renders each one as a finished vertical short, straight into your library and on to your social accounts. **On this page** - [What Auto-Clips does](#what-auto-clips-does) - [Why you'd use it](#why-youd-use-it) - [Before you start](#before-you-start) - [The pipeline at a glance](#the-pipeline-at-a-glance) - [Step-by-step: your first batch of shorts](#step-by-step-your-first-batch-of-shorts) - [The options, in detail](#the-options-in-detail) - [Source size and length limits](#source-size-and-length-limits) - [What each stage costs in credits](#what-each-stage-costs-in-credits) - [Where your shorts land](#where-your-shorts-land) - [Tips](#tips) - [Troubleshooting](#troubleshooting) - [Related pages](#related-pages) --- ## What Auto-Clips does You give Auto-Clips one long video: a podcast episode, a webinar, a talk, a stream, a vlog, gameplay footage. The AI watches the whole thing and returns a ranked list of **highlight moments**, each with a punchy title, a start and end time, a moment type (a hook, an insight, a funny beat, a payoff, and so on), and a 0–100 shareability score. You tick the moments you want, and one click turns each into a finished short: trimmed to exactly that moment, reframed to vertical **9:16** (or square, or widescreen), optionally overlaid with styled captions and a persona-matched voiceover, and rendered out as a real video file. Every short lands in your media library and appears in your Projects, each one a normal, fully editable project you can open in the studio afterward. When the batch finishes, a final panel lets you schedule the finished shorts to your connected social accounts, staggered on a cadence you choose. ## Why you'd use it - **One hour of footage becomes a week of content.** Instead of scrubbing a timeline hunting for clips, you review a short list of ranked moments and pick. - **The AI watches, it doesn't just listen.** The detector finds moments by watching the actual footage, so a quiet emotional beat, a visual gag, or a clutch play gets found even when nobody says anything quotable. - **The output is finished, not a rough cut.** Captions, voiceover, vertical reframe, and the render all happen in one pass. What comes out is ready to post. - **Everything stays yours.** Shorts render into your own workspace library, and each one is an editable project. Polish any of them in the studio like any other edit. ## Before you start 1. **Sign in.** Auto-Clips is part of your workspace; if you're signed out, the page asks you to sign in. 2. **Have a source video.** Pick from your library, paste a **public video URL** (YouTube / Vimeo / TikTok watch pages work; we import them), or **scan a channel** and multi-select up to 5 videos to batch. 3. **Credits, if metering is on.** **Core** detect, import, overlays, and the first free shorts of the day do **not** burn vendor credits. **Premium** video-AI detect and renders beyond your free daily allowance do. See [What each stage costs](#what-each-stage-costs-in-credits). 4. **For social publishing (optional):** connect your social accounts first, in the dashboard's **Publishing** section. CoreReflex publishes through a GoHighLevel connection, so you connect once and reach the platforms you've linked there (Facebook, Instagram, LinkedIn, X, Google Business, TikTok). Open Auto-Clips from the **Auto-Clips card on your dashboard** ("Open Auto-Clips →"), or go directly to `corereflex.com/auto-clips`. Paste a YouTube URL and click **Clip from URL (native)** for the oneshot path. Same oneshot from MCP (`corereflex_auto_clips` + `sourceUrl`) or CLI (`node scripts/cr-clips.mjs ""`). Premium video AI never runs unless you approve the **Premium Flag**. ## The pipeline at a glance **Single video** ``` Pick a source (library / URL / YouTube) → Detect (Core free · CR · Premium) → Review → Generate & render (first free shorts/day included) → Edit on timeline → optional Publish ``` **Channel batch** ``` Channel URL → List videos (free) → Multi-select up to 5 → Batch (import → detect → cut, live % progress) → Shorts on timeline → optional Publish ``` Detect quality: | Mode | What it does | Cost | |------|----------------|------| | **Core** (default) | Speech transcript + free local ranker | Free | | **Auto** | CoreReflex model when live, else Core | Free | | **CoreReflex** | Native-stack model when deployed | Free (fallback to Core if offline) | | **Premium** | Video AI watches the whole clip | Content credits | The status line (and for batches, a **percent progress** line) keeps you informed while work runs. ## Step-by-step: your first batch of shorts ### Step 1: Choose a source and set your options 1. Open **Auto-Clips** from the dashboard card. 2. **Pick your source**, any of: - A video from your library (click a tile), - A **YouTube / Vimeo / TikTok watch URL** (we import the master), - A **channel or playlist URL** → **List videos** → multi-select up to 5 → **Batch selected videos** (or double-click one video for single Detect). 3. Set the options (all have sensible defaults): - **Detect quality**: Core / Auto / CoreReflex / Premium (see table above). - **Clips** sets how many highlight moments to look for (1–20, default 4). - **Min sec / Max sec** bound the shortest and longest a clip may be (defaults 12 and 60 seconds). - **Aspect**: **9:16** vertical (default), **1:1**, or **16:9**. - **Overlays** are free the render engine graphics: Full (hook + brand), Hook only, Brand only, or Off. Brand-kit logo is used when available. - **Moments**: **General** or **Gaming**. - **Captions / Voiceover / Story** are optional; AI styled captions and voiceover spend credits. 4. Click **Detect highlights** (single video) or **Batch selected videos** (channel multi-select). What happens next, automatically: - The status line shows *"Preparing the source for analysis…"*. A long video is first condensed into a small analysis copy so the AI can watch all of it. That copy preserves the real timestamps, so the moments it finds map exactly back onto your original footage. (This step needs the render service on your instance; see [limits](#source-size-and-length-limits).) - If preparation isn't available on your instance, the page says so honestly and falls back to analyzing the source directly (which works for small clips), then reveals three extra fields (**Source width**, **Source height**, **Duration sec**) that you'll need to fill in before generating. - Then: *"Analyzing the video for the best moments…"*, and finally *"Found N highlights. Review and generate."* ### Step 2: Review and select highlights The **Review highlights** panel lists every moment found, best first. Each row shows: - a **checkbox** (all moments start selected, so untick any you don't want), - the **title** the AI gave the moment (e.g. "The one mistake every founder makes"), - the **time range** in the source (e.g. `312.0s – 355.5s`), - a **type chip** (hook, insight, funny, payoff, clutch…), - the **score**, a 0–100 estimate of how attention-grabbing and shareable the moment is. Moments never overlap each other, and the list is deliberately diverse: the detector picks the best moment of each *type* first, then fills the remaining slots by score, so you don't get five near-identical clips. Untick anything you don't want, then click **Generate selected shorts**. ### Step 3: Generate, watch them render Depending on your options, the status line walks through each stage as it happens: - *"Writing the story arc…"*: story mode writes one narrative across your selected clips. - *"Writing captions…"*: the AI writes your chosen persona's captions for the footage. - *"Recording voiceover 1/4…"*: a persona-matched narrator records each clip's voiceover. - *"Cutting and rendering N shorts…"*: each moment is trimmed, reframed, and queued to render. The **Your shorts** panel shows one card per clip, updating every few seconds: - **Queued…** → **Rendering…** → **Ready** (an inline player appears, plus an **Open / download** link), or - **Failed**, with the reason shown on the card. You can leave the tab open and come back; the cards keep polling until every short finishes. ### Step 4: Publish to your socials (optional) Once the batch finishes (and at least one short succeeded), the **Distribute to socials** panel appears: 1. Your connected accounts load as clickable pills, all selected by default. Click a pill to toggle it off or on. (If nothing is connected yet, the panel links you to the dashboard to connect.) 2. **Start (optional)** sets when the first post should go out. Leave it empty to start now. 3. **Every N hours** is the cadence between posts (default 24). Each finished short is scheduled one step after the previous one, so a batch of five shorts with a 24-hour cadence becomes five days of daily content. Set it to 0 (with no start time) to publish everything immediately. 4. Click **Schedule finished shorts →**. The panel confirms how many were scheduled, and tells you plainly if any failed. Each short posts with its highlight title as the caption. Scheduled posts also appear in the dashboard's **Publishing** section, where you can track their status. ## The options, in detail ### Moment presets: what the AI looks for - **General** (the default, right for talks, podcasts, vlogs, tutorials, any footage): *hook* (a scroll-stopping opener), *insight* (a valuable idea or "aha" explanation), *emotional* (an inspiring or heartfelt peak), *funny*, *surprising* (a twist or bold claim), *action* (high-energy, visually dynamic), and *payoff* (a satisfying result, punchline, or reveal). - **Gaming** (for streams and gameplay): *action* (combat and close calls), *funny* (fails and glitches), *clutch* (1vX wins and last-second saves), *wtf* (unexpected chaos), *epic fail*, *hype* (peak-excitement celebration), and *skill* (trick shots and impressive plays). ### The nine caption personas Captions are **AI-written** for your footage in the persona's voice, timed and spaced across each clip, never bunched at the start, never overlapping. Each caption may carry one fitting emoji. | Persona | The vibe | |---|---| | **Gaming** | Short, punchy creator captions: "HEADSHOT!", "wait for it…". ALL CAPS on the action beats. | | **GenZ** | Internet-native slang reactions: "no cap this is insane", "he's locked in rn". | | **Funny** | Meme-style, self-aware humor: "skill issue tbh", "well that escalated". | | **Dramatic** | Cinematic and impactful; lowercase for tension, CAPS for the climax. | | **Minimal** | Three words max, understated: "nice.", "well then." Less is more. | | **Story · News** | Professional broadcaster narration, with fewer, longer captions building excitement. | | **Story · Roast** | Sarcastic, playful commentary; mocking, never mean. | | **Story · Creepypasta** | Horror-style tension: ominous, slow reveals, ellipses… | | **Story · Dramatic** | Epic movie-trailer narration: "One shot. One chance." | **Auto (match moment)** lets CoreReflex pick a fitting persona for you instead of choosing one yourself. Captions in Auto-Clips are invented by the AI to *enhance* the footage, which is the right choice for gameplay, b-roll, and visual moments. If you want **verbatim speech captions** (an exact transcript of what's said), open the finished short in the studio (every short is an editable project) and use the caption tool there: see [Captions & subtitles](editing-and-mastering.md#captions--subtitles). ### Voiceover: a narrator matched to your persona Turn **Voiceover** to **On (match style)** and each clip gets a spoken narration whose voice matches the caption persona: an energetic hype voice for Gaming, a deep movie-trailer narrator for Dramatic, a near-whisper for Creepypasta, a confident broadcaster for Story · News, and so on. The voiceover speaks the clip's caption text (or, with no captions, the clip's title and hook). While the voiceover plays, the clip's original audio is automatically **ducked**, turned down to roughly a third of its volume, so the narration sits clearly on top without silencing the moment underneath. And if the narration runs slightly longer than the clip, the clip extends to fit it (playing a little more source footage) rather than cutting the voice off mid-sentence. It will never extend past the end of your source video. ### Story mode: one narrative across every clip With **Story mode** on, your selected clips stop being independent moments and become chapters of one story: the first clip sets it up, the middle escalates, the last one concludes, each part continuing from the one before. The narrative is written in a story persona (Story · News by default, or whichever Story persona you picked as your caption style), appears as on-screen captions spread across each clip, and, with Voiceover on, is spoken by the matching narrator. Story mode needs **at least two selected highlights**. It's the mode for compilation-style shorts: a match retold like a broadcast, a stream retold as a roast, an event retold as a trailer. ## Source size and length limits An honest account of what fits today: - **Small clips** (roughly up to 20 MB of video) can be analyzed directly, even without the preparation step. - **Longer sources** are handled by the automatic **prepare** step, which condenses the video into a compact, time-accurate analysis copy. Preparation accepts sources up to about **500 MB** and requires the render service to be deployed on your instance. If it isn't, the page says so plainly and small clips still work directly. - **Very long footage** (beyond roughly an hour) can produce an analysis copy that is still too large to analyze. If that's your case today, cut the source into parts and run each part through Auto-Clips. Support for hour-plus sources is planned. - **Batch size:** the detector finds up to **20** moments per run; a single generate cuts and renders up to **10** shorts (your instance's operator can adjust this cap). If you select more than the cap, the extra selections are not rendered, so run a second batch for the rest. - **Clip length:** each short can be 3 seconds to 3 minutes; the defaults (12–60 seconds) match what short-form platforms reward. ## What each stage costs in credits When billing metering is enabled on your instance, each stage draws credits from your workspace balance, always **before** the work runs, so nothing is ever partially charged. If your balance is too low, the stage stops with a clear message and no charge. (With metering off, these run without drawing credits.) See [Plans & Billing](plans-and-billing.md) for how credits work. | Stage | When it's charged | Credits | |---|---|---| | Prepare (long-video analysis copy) | Native proxy | Free | | Detect highlights (Core / Auto / CR) | Default native speech ranker | Free (no partner AI) | | Detect highlights (Premium) | Partner video AI; **Premium Flag** required (`confirmPremium`) | 3 | | Captions | Once per batch (written on the source, then shared across clips) | 3 | | Story arc | Once per batch | 3 | | Voiceover | Per clip | 5 | | Render a short | Per clip | 10 | | Publish to socials | Per short scheduled | 2 | **Worked example.** A long video into 4 captioned, voiced shorts, published: 3 (prepare) + 3 (detect) + 3 (captions) + 4 × 5 (voiceovers) + 4 × 10 (renders) + 4 × 2 (publishing) = **77 credits**. Skipping the extras is cheaper: 4 plain shorts with no captions, voice, or publishing is 3 + 3 + 40 = **46 credits**. ## Where your shorts land Every finished short lands in two places automatically: - **Your media library** holds the finished video file, alongside everything else you've uploaded or generated. It also appears in the Auto-Clips source picker itself, so a short can even become a source. - **Your Projects**: each short is saved as its own project, named `Auto-clip · `, in the dashboard's **Projects** view. Click **Open** on any of them to load it in the studio: the trimmed clip, the vertical reframe, the caption track, and the voiceover track are all there as ordinary editable layers. Tweak a caption, re-grade the color, swap the music (anything the studio can do), and render again. Nothing about an Auto-Clips short is special or locked; it's exactly as editable as a project you built by hand. ## Tips - **Match the Moments preset to the footage.** *Gaming* on gameplay footage finds clutches and fails that *General* would file under generic "action." *General* on a podcast finds insights and hooks the gaming preset isn't looking for. - **Detect more than you need.** Ask for 8 moments and keep the best 4. Reviewing is cheap; the expensive stages (voiceover, render) only run on what you tick. - **Trust the diversity, check the score.** The list is deliberately varied by type; the score tells you which ones the AI believes in most. A 90-score *hook* is usually your opener for the week. - **Story mode + Voiceover is the compilation combo.** Three or four clips, one arc, one narrator: it turns a highlight reel into an episode. - **Use library sources when you can.** They're already in your workspace with working links. If you paste an external URL, it must be a direct link to the video file, publicly reachable over https. - **Keep 12–60 seconds unless you have a reason.** Platforms reward complete-but-tight; the detector is tuned to find self-contained beats in that window. - **Polish the winner in the studio.** When one short takes off, open its project and cut a variant (new caption persona, different opening frame) without re-detecting anything. ## Troubleshooting - **"Enter a public https video URL."** Use a public **watch-page** URL (YouTube / Vimeo / TikTok; we import a master via the isolated worker) or a direct `https` media file. Private / non-allowlisted hosts are rejected. See [source import](./source-import.md). - **"Preparation unavailable. Analyzing the source directly (small clips only)."** The long-video preparation service isn't available on your instance right now. Small clips still analyze directly; to generate you'll be asked for the **Source width**, **Source height**, and **Duration sec** of your video (a video file's properties show these). - **"Long-video preparation is being provisioned and is not available yet."** Same cause, stated by the server: your instance's render service isn't deployed yet. You are not charged when this happens. - **"No standout moments found. Try a different source."** The AI didn't find self-contained moments in your bounds. Try the other Moments preset, widen **Min sec / Max sec**, or raise **Clips**. - **"That source asset is not in your workspace."** The link you pasted points at a CoreReflex asset belonging to a different workspace. Pick the video from your own library. - **"Enter the source width & height to generate."** Preparation was skipped, so CoreReflex doesn't know your video's dimensions. Fill in the manual fields that appeared under the options. - **A short is stuck on "Queued…".** The render service hasn't picked the job up yet: it's queued, not lost. Leave the tab open, or check the project's render status later under **Projects** in the dashboard. - **A short shows "Failed."** The card shows the reason. Re-run generate with just that highlight ticked; if it keeps failing, open the short's project from **Projects** and render from the studio to see more detail. - **"Captions skipped: …" or "Story skipped: …".** That layer hit an error and the run continued without it (your shorts still render). The message says why; the usual fixes are retrying, or checking your credit balance. - **A clip came back without its voiceover.** A single clip's voiceover can fail without stopping the batch; that clip simply renders without narration. Re-generate that highlight to try again. - **"Too many … requests. Try again shortly."** You've hit a rate limit. Wait a minute and continue. - **"No social relay connected" in Step 4.** Connect your social publishing account in the dashboard's **Publishing** section first, then return; the accounts load automatically once shorts finish. - **You were sent to the sign-in page mid-run.** Your session expired. Sign back in and re-run; detection results aren't kept across sessions, but your source video and any finished shorts are safe in your library. - **The run stopped with a credits message.** Metering is on and your balance was too low for the next stage; nothing was charged for it. Top up or upgrade under **Settings → Plan** ([Plans & Billing](plans-and-billing.md)) and retry. ## Related pages - [Editing & Mastering](editing-and-mastering.md): the studio where every finished short can be opened and polished; includes [Captions & subtitles](editing-and-mastering.md#captions--subtitles) (verbatim speech captions) and the [Shorts Studio](editing-and-mastering.md#shorts-studio) (reframe a *finished film* into a short, the manual sibling of Auto-Clips). - [Directing Films](directing-films.md): generate a film from a brief; its output videos are valid Auto-Clips sources. - [Voiceover & Music](voiceover-and-music.md): the narration and scoring tools behind the voiceover personas. - [Plans & Billing](plans-and-billing.md): how credits, metering, and tiers work. - [Getting Started](getting-started.md): sign-in, accounts, and your first project.