Came from an ad? You start with 50 free seeds — no card.

How it actually works

Every VidSeeds.ai workflow, explained straight

No fluff — what each tool does, the page it lives on, what it costs in seeds, and where you stay in control. Upload a file or connect a channel and follow along.

Metadata from your actual video•Thumbnails from real frames•Six platforms, 85 languages

Create & optimize

Start a project from a local file, check it before you publish, tighten the edit, and build the thumbnail.

Run a project end to end

Start from a local video, let the New Project screen do the analysis, then move through optimize, schedule, and results — you confirm every step.

1.
Pick your accounts first — Open /projects/new. If you've connected channels, the flow asks which ones before you upload, so the metadata is built for the right platforms.
2.
Upload a local file — Drag the file in. The browser probes it, extracts what it needs, and asks you to confirm the project name before anything runs.
3.
Let the pre-create analysis finish — Probing, extraction, transcription, metadata, and frame prep all happen before the project is created — no half-built projects.
4.
Edit each platform, then publish — You land on /projects/[id]/optimize. From there go to schedule and results to publish now or queue it for later.

Refresh a video you already published

Pick a published YouTube video, attach the source file, and push only the fields you choose back to YouTube.

1.
Open the video picker — Go to /reoptimize and pick from your connected channel's cached video list.
2.
Open that video's optimizer — Selecting a card opens /reoptimize/[videoId], tied to that one published video.
3.
Attach the original file — It asks for the local file so it can transcribe, pull frames, and generate fresh title, description, tag, and thumbnail options.
4.
Choose what to update — Review everything and push only the fields you want — leave the rest of the video untouched.

Check a video before you publish

Upload a cut before it goes live and see where the audio, visuals, hook, retention risk, and metadata actually stand.

1.
Upload a local file — Open /diagnose and drag a file in. Seeds are charged only when a run finishes.
2.
Let the pipeline run — It pulls audio, transcribes, samples frames, runs visual and audio checks, and turns them into plain findings.
3.
Read the score and the fixes — You get an overall score, per-platform publishability, content cues, strengths, and a prioritized fix list.
4.
Revise and run it again — Treat it as a pre-upload QA pass — fix the top issues, then re-check the new cut.

Tighten the edit with Precision Trim

Upload a file, let the analysis suggest a hook, cuts, and reorders, then export the trimmed version from your browser.

1.
Add a video or audio file — Open /precision-trim and drop a local media file.
2.
Let it build the critique — It extracts audio, transcribes, and proposes specific hook, cut, and reorder edits.
3.
Approve what you want — The flow opens /precision-trim/editor, where you keep only the suggestions you agree with.
4.
Export from the browser — Export supported files locally. There's a 500 MB cap on the client-side rebuild, so very large masters export in pieces.

Build a thumbnail in the Studio

Generate from a video, a short brief, or a full prompt — then download it or publish it straight to YouTube.

1.
Pick how you start — Open /thumbnails and start from a video, a short brief, or a prompt you write yourself.
2.
Add your channel's style — Load a connected channel's style, detect niche trends, pick a preset, and set how the on-image text is generated.
3.
Generate and compare — You get one or more options to compare side by side, and you can open the winner in the editor. The model renders any text inside the image — no server-side overlay.
4.
Download, swap, or publish — Download the image, export a video with the first frame swapped, or publish the thumbnail to a selected YouTube channel.

Refine a thumbnail with a prompt

Once you have a result, describe the change you want and keep iterating until it's right.

1.
Generate one first — Start in /thumbnails and create at least one result — the editor works on a saved thumbnail.
2.
Open the edit route — Use Edit Selected to open /thumbnails/[id]/edit.
3.
Say what to change — Type the change — background, lighting, text, composition — and submit a new pass.
4.
Keep the best one — Each edit replaces the current image with a new version you can download or push further.

Repurpose & localize

Translate metadata into 85 languages, manage the results, and cut vertical shorts from long video.

Translate a YouTube video's metadata

Pick a video, choose languages, review the results, and publish the localizations you approve to YouTube.

1.
Pick a YouTube video — Open /translate and choose one from your connected channel.
2.
Choose your languages — The dialog shows the seed cost from the number of languages before you start.
3.
Review on the results page — Finished jobs land on /translate/results, where you filter, edit, retry failures, and download everything.
4.
Publish what you approve — Selected title and description localizations go to YouTube through its localizations API — you decide which languages ship.

Work a translation results session

Use the results screen to sort languages, retry the ones that failed, download the package, and publish the approved set.

1.
Open a results session — Go to /translate/results after a run, or reopen an earlier session link.
2.
Check each language — Use search, status filters, and the detail view to see what's ready, published, pending, or failed.
3.
Retry or download in bulk — Select the languages you want, retry failures, and download the full package for an offline read.
4.
Publish the approved ones — Send only the languages you've approved back to YouTube from the same session.

Cut shorts from a long video

Upload long-form video, pick the moments that stand out, and reframe them into vertical clips.

1.
Upload a long video — Open /auto-clips and add a local file.
2.
Let it find the moments — It reads the duration, pulls audio, transcribes when it can, and falls back to visual analysis when there's no transcript.
3.
Choose the clips — On /auto-clips/select you review the detected moments and pick which ones to cut.
4.
Generate and reframe — Chosen moments are cut, then reframed to vertical so the finished shorts are ready to post.

Download the shorts package

Once the clips are done, grab them one by one or as a ZIP with captions and upload notes.

1.
Finish the clips first — This page fills in after /auto-clips/select produces at least one finished clip.
2.
Open the delivery page — Finished runs redirect to /shorts-download.
3.
Check captions and metadata — It builds title, description, hashtags, TikTok text, Instagram copy, and alt text from the main clip.
4.
Download the files — Take a single clip or the full ZIP with metadata, platform captions, and upload instructions.

Analyze & decide

Read your channel, find the next video, place mid-rolls, and find out why one underperformed.

Read your whole channel

Run one channel analysis and load 18 tabs of data — audience, meaning, visuals, integrity, competition, retention, and more.

1.
Start a run — Open /intelligence with a connected YouTube channel and trigger the analysis.
2.
Load the tabs — Audience, meaning, visual, integrity, pre-publish, metadata, competitive, retention, and the rest hydrate in batches.
3.
Read the numbers and summaries — Each tab shows typed analytics, and the AI cards explain the areas you dig into.
4.
Turn it into a plan — The pre-publish and action-plan views translate the findings into what to do next.

Find your next video

Get ten ideas built from your last 100 uploads plus the trend context the service assembles.

1.
Load your channel — Open /video-ideas with a connected YouTube channel — it checks your seed balance first.
2.
Generate ten ideas — The request sends your active channel, language, and a fixed count of 10 to the generator.
3.
Read the reasoning — Each idea comes with the channel analysis and the trend inputs behind it.
4.
Plan from them — It's built for ideation — regenerate for a fresh batch instead of editing in place.

Write a channel description

Pull your recent channel context, generate a description, and publish it back to YouTube.

1.
Load a channel — Open /channel-description-generator and pick the channel to analyze.
2.
Review what's there now — It loads your current description plus recent video context from YouTube or saved analysis.
3.
Generate a new one — It sends your channel name, current description, videos, language, and connection IDs to the generator.
4.
Publish to YouTube — When it reads right, publish to update the channel description.

Place mid-roll ads at the right spots

Read caption timing to find natural breaks, then jump into YouTube Studio to set them.

1.
Pick a video — Open /midroll-ads and choose one from your connected channel.
2.
It needs timecoded captions — It fetches captions first and stops with a clear message when there isn't enough timecoded transcript.
3.
Review the suggested breaks — You get timestamps with confidence, reasoning, and the surrounding context for each break.
4.
Open YouTube Studio — The results include a direct Studio link plus a guide on placing the breaks.

Find out why a video underperformed

Run an autopsy on a published video and read the report before you change anything.

1.
Pick a published video — Open /why and choose one from your connected YouTube grid.
2.
Run the autopsy — The page switches into report mode for that video and channel.
3.
Read the diagnosis — It combines performance context, best-practice analysis, key findings, and prioritized fixes.
4.
Compare another — Go back and run the same autopsy on the next video.

Manage & operate

Come back through the dashboard, reopen a project where you left it, connect platforms, and check your voice profile.

Use the dashboard as home base

Come back here for quick actions and recent work — it's built for resuming, not deep editing.

1.
Open the hub — Go to /dashboard for the welcome header, connect-channel banner, quick actions, and recent projects.
2.
Jump to the right tool — Quick actions link to projects, thumbnails, auto-clips, translate, diagnose, mid-rolls, and intelligence.
3.
Pick up recent work — Recent projects reopen on their current step instead of sending you back to the start.
4.
Make it your return page — Treat it as your operating home base between sessions.

Reopen an existing project

Use the projects list to jump straight back to where you left each one.

1.
Open the projects page — Go to /projects for the full project history, not just the recent few.
2.
Scan status and platforms — Each row shows status, the connected platform icons, and the created date.
3.
Jump to the current step — Links resolve to /projects/{id}/{currentStep}, so unfinished work opens where you stopped.
4.
Clear out old drafts — Delete the stale draft or finished projects you don't need.

Connect your platforms

Connect and group your social accounts — the available platforms depend on the domain you're on.

1.
Open social accounts — Go to /settings/accounts for your connected totals, account cards, and the platform sections enabled here.
2.
Connect or reconnect — Each section has connect, add-another, reconnect, or disconnect, depending on the state.
3.
Group for publishing — Bundle connected accounts into groups so you can select them later.
4.
Use them everywhere — Project setup, thumbnail publishing, and the YouTube flows all read from these connections.

Check your voice profile

See the per-channel voice fingerprint VidSeeds.ai builds, and re-run it when your channel changes.

1.
Open Voice Identity — Go to /settings/voice — it loads each connected channel and any saved fingerprint.
2.
Read the fingerprint — Each card shows personality, trait bars, signature words, and voice strength.
3.
Re-analyze when it drifts — Run Re-analyze to refresh the fingerprint and reload the saved profile.
4.
It feeds the AI — This same voice data shapes project optimization and thumbnail prompts.

Run it from your AI tools

Connect Claude, Cursor, Codex, or ChatGPT to the MCP server and run the same tools without leaving your editor.

Connect your AI tools to VidSeeds.ai

Point Claude, Cursor, Codex, or ChatGPT at the MCP server and run the same tools you use on the web — without leaving your editor.

1.
Open MCP Settings — Go to /settings/mcp-settings. Claude.ai and Claude Desktop connect over OAuth with no token.
2.
Create a token for everything else — For Cursor, Claude Code, Codex, VS Code, and other clients, create a Personal Access Token — it's shown once, so copy it right away.
3.
Point the client at the endpoint — Use https://vidseeds.ai/api/mcp (or /api/mcp/chatgpt for the ChatGPT app) and send your token as a Bearer credential.
4.
Start the 14-day trial — The free trial starts on your first connection. After it ends, an active subscription keeps the connector on.

Run the whole toolkit from a chat

Your agent can analyze a video, write metadata, build thumbnails, pull clips, localize, and prepare publishing — with your channels, voice profile, and seed balance already in context.

1.
Ask in plain language — Describe what you want; the agent calls the same tools the web app uses, so the two surfaces stay in parity.
2.
Work with your real context — Connected channels, your learned voice, and your seed balance carry over, so results match what you'd get in the browser.
3.
Watch the seed cost — Each connector has a daily included-call quota; past it, calls cost seeds at the same rates as the web. Checking your balance is always free.
4.
Keep local work local — Trimming, reframing, and frame extraction still run through FFmpeg on the machine your agent runs on.

Frequently Asked Questions

What is VidSeeds.ai, exactly?

It's a pre-upload video SEO tool. You bring an existing video or connect a YouTube channel, and the AI reads the actual content — speech, scenes, pacing — to write titles, descriptions, tags, chapters, and thumbnails for YouTube, TikTok, Instagram, Facebook, LinkedIn, and X. It is not a video generator, editor, or host. You review and edit everything before it publishes.

Where does a project go after /projects/new?

Creating it is just the entry step. The New Project flow finishes the analysis first, then lands on /projects/[id]/optimize. The same family also includes /projects/[id]/analyze, /projects/[id]/schedule, and /projects/[id]/results.

Which pages need a connected YouTube channel?

The YouTube-driven ones — Re-Optimize, Translate, Mid-Roll Ads, Why, Intelligence, Video Ideas, and Channel Description — need a connected channel. The upload-first ones — Projects, Thumbnails, Auto-Clips, Diagnose, and Precision Trim — start from a local file. Translation review sessions start from Translate after you pick a YouTube video.

Does Translate publish to every platform?

No. Translate creates a review session, lets you edit and retry languages, and publishes the title and description localizations you approve to YouTube — not to the other platforms.

What upload and export limits matter here?

Server uploads go up to 1 TB. Some browser-side exports are stricter — there's a 500 MB cap on the client-side rebuild used for thumbnail first-frame export and Precision Trim export.

Can my AI assistant run VidSeeds.ai for me?

Yes. Connect Claude, Cursor, Codex, ChatGPT, or another MCP client to https://vidseeds.ai/api/mcp — OAuth for Claude, a Personal Access Token from /settings/mcp-settings for the rest. It runs the same tools as the web app, with your channels and voice in context. There's a 14-day free trial from your first connection; after that it needs an active subscription, and tool calls use seeds at the same rates as the web.

Does it publish anything on its own?

No. Every title, description, tag, thumbnail, translation, and caption is yours to review and edit first. Nothing reaches your channels until you choose to publish — on the web or through the MCP connector.

Ready to run a real workflow?

Open the current flows, start with a local file or a connected YouTube channel, and follow the same routes the app uses today.

Metadata from your actual video•Thumbnails from real frames•50 free seeds to start

No card. You review every title, tag, and thumbnail before it goes anywhere.