On the evening of October 26, LibTV published an official announcement: it is now officially integrated with WorkBuddy. Inside a WorkBuddy task conversation, you can directly call LibTV's professional creation tools and director-level Skills, and you can also reference proven hit videos published on LibTV's own platform. The announcement frames it this way: "Your working context can become the starting point of video creation," plus a punchier slogan - one person is a video factory.
Background first: LibTV is the professional AI video creation platform under Liblib AI. It launched in March 2026 and self-reports 100+ AI director Skills and dozens of aggregated models on its website [official self-reported figures]. WorkBuddy is Tencent's AI Agent workspace, built around delivering finished work results inside a conversation. The logic of the integration is not complicated: WorkBuddy understands your request, breaks it into tasks, and manages your working files; LibTV actually produces the video.
This article sticks to practice: how to connect, how to write the request, ready-to-copy templates for three scenarios, how to read the cost claims, how to set up scheduled jobs, and finally the advanced route that skips the connector entirely - plus the boundaries you should know before committing budget.
One. Connect in two steps
According to the official announcement, setup takes two steps and never touches an API key.
Step one: in a WorkBuddy task conversation, tap the [+] button next to the input box, choose "Connectors," search for "LibTV," tap [Link], and follow the on-screen prompt to authorize your account. Once authorized, LibTV's creation capabilities are attached to WorkBuddy as callable tools.
Step two: return to the task conversation, upload your material, and state your request in one sentence. The material can be a product demo, an introduction deck, a story outline, or just a single idea.
The granularity of this integration is worth noting: this is not a shallow "hand you a link and send you off to LibTV" setup. WorkBuddy calls LibTV's professional creation tools and director-level Skills directly inside the conversation [official]. Request breakdown, creative planning, script and storyboard work all happen in the same thread; you never shuttle context between two windows. That is exactly what the official line about working context means.
Two. The four elements of a request
The official announcement is explicit about what a good request contains: what to shoot, where it will run, what style you want, and what must be preserved. These four lines are not filler. Each one directly affects model tier selection, storyboard direction, and re-roll costs. Here is one copyable example for each:
- What to shoot: "A 45-second brand video for a portable espresso machine, following the product from packed-away to full cup in a camping scene."
- Where it runs: "Paid feed placement on short-video platforms, vertical 9:16, product close-up within the first 3 seconds, purchase call-to-action frame at the end."
- What style: "UGC feel like an outdoor lifestyle creator, warm tones, handheld camera, no announcer-style voiceover."
- What must be preserved: "The product name and the slogan 'Hot coffee all the way' must appear verbatim and must not be rewritten."
One sentence of summary: the more specific the four elements, the fewer guesses WorkBuddy makes, the higher the first-pass rate at LibTV, and the more predictable your credit burn. The reverse also holds - throw in a bare "make me a video" and you will most likely torch your quota on re-rolls.
Three. Three ready-to-copy scenario templates
The official announcement lists three typical scenarios. Below, each scenario's official workflow is paired with a request template you can lift and edit.
Scenario one: brand video
Official workflow: upload a product demo or introduction deck; WorkBuddy extracts selling points and plans the creative; LibTV handles the script, storyboard, and video generation.
The deck I uploaded introduces our new portable espresso
machine. Read it fully, extract the 3 core selling points,
and write a 45-second script and storyboard in four acts -
pain point, product reveal, feature demo, call to action -
vertical 9:16, outdoor UGC look. Product name and slogan
must be preserved verbatim. After I confirm the storyboard,
call LibTV to generate the final video.The value here is skipping the "write the script yourself" step. The selling points in a deck are already your commercial conclusions; WorkBuddy translates conclusions into storyboard language, and LibTV turns storyboards into footage. The chain is coherent.
Scenario two: ad creative variants
Official workflow: take one validated creative concept; WorkBuddy plans variations across openings, selling points, and scenes; LibTV batch-generates UGC-style assets for paid testing.
The "see the cleaning effect in 3 seconds" asset already
has proven click-through. Around the same concept, plan 5
variants: two new openings (a dirt close-up; a surprised
user reaction), two new selling points (time saved; no
disassembly), one new scene (kitchen swapped for car
interior). Each under 20 seconds, vertical, generated via
LibTV, delivered as a list for my selection.Paid testing is sampling, not gambling: batch out variants of one proven concept and let the data pick the winner worth scaling. The fact that the official announcement calls this scenario out suggests it is aimed squarely at media-buying teams whose pain is asset burn rate and iteration speed.
Scenario three: short drama production
Official workflow: start from a story synopsis; WorkBuddy organizes the plot and breaks down production tasks; LibTV advances character and scene design, storyboard work, and continuous shot generation.
Synopsis: late at night, a delivery rider accepts an order
to his own childhood home; the person who opens the door is
the father he has not seen in ten years. Organize the full
plot line and break the production into tasks, listing the
characters and scenes that need design first. After my
confirmation, advance character design, then storyboards,
then generate continuous shots scene by scene. Plan for 5
episodes at 90 seconds each.Short drama is the scenario that punishes sloppy task breakdown: character consistency, scene continuity, shot-to-shot flow - break any one and the audience notices. Splitting design, storyboards, and generation into a task chain with dependencies is precisely what an Agent platform is good at.
Four. Reading the cost claims
"Three yuan for a brand video, five yuan for an episode of short drama" - this is the official announcement's own figure [official self-reported], quoted here verbatim as a claim, not a guarantee. The stated basis is model selection: the announcement credits the cost advantages of models such as Wan 3 and Seedance 2.0 Fast [official].
Third-party retellings of platform pricing (secondhand) go further: model credit pricing claimed to be up to 92% below competitors, membership SKUs 76% below. Treat those as marketing figures too; your real bill is decided by your own usage.
More interesting than unit price is how openly the hit-making process is documented. According to the announcement, productions like the viral hit "The One Who Came Before" have published their full creation canvases and prompts on LibTV, and WorkBuddy can reference those canvases while adapting them to your own product or story [official]. This is a rare "publish the recipe" move: what you see is not a finished-video black box but the storyboard nodes and the exact prompts behind each step. For a creator entering this space, dissecting a few public canvases beats fumbling in the dark.
Five. Scheduled jobs: a 7x24 production line
The announcement notes that you can combine WorkBuddy's scheduled tasks with your working context to turn a well-defined production request into a recurring job that batches out videos on schedule [official]. In practice, you write the four elements into a fixed instruction and attach a schedule - for example: "Every Monday at 9 a.m., read the 'featured product this week' document in my workspace; if it changed, generate three 20-second vertical UGC-style assets in the established style and output a list for my selection." Once fixed, every run follows the same standard.
Three use cases fit this pattern naturally: new product launches (an intro video for each month's new items), routine operations (three vertical shorts for social feeds every Monday), and paid testing (a weekly batch of different openings). One practical recommendation before you automate: run the full pipeline manually once, confirm both output quality and credit consumption, and only then hand it to the scheduler. The reasons are in the next section.
Six. Advanced: skip the connector, install the open-source Skills
If your Agent platform is not WorkBuddy, or you want the integration inside a local agent, LibTV has open-sourced its Agent Skills. Repository: github.com/libtv-labs/libtv-skills, i.e. libtv-labs/libtv-skills. This section is a secondhand retelling of the GitHub repository page; reachability was re-verified via the GitHub API before publication.
npx skills add libtv-labs/libtv-skillsThe repository is MIT-licensed, follows the OpenClaw skill specification, and is written in Python; the libtv-skill package includes scripts such as create_session, query_session, upload_file, and download_results [secondhand]. Authorization uses an access key: in the LibTV creation interface, open LibTV Skills from the top-right corner, create an access key, and configure it for your local agent [secondhand, from hands-on community write-ups].
Note that this advanced route is not the connector flow from the official announcement; it is a path the community has walked through. Follow the repository README for exact steps. MIT licensing means you can modify it freely - and own the results.
Seven. Cold water and boundaries
- The cost figures are official self-reported claims. Real spending depends on video length, model tier, and re-roll counts. "Three yuan" most likely assumes the cheapest model tier and a clean first pass; re-rolls and premium tiers change the math fast.
- Output quality tracks model tier and re-roll luck. Waste shots inside batch jobs are normal; the human selection step does not go away.
- Scheduled jobs depend on account quota and model queueing. When credits run dry or queues back up, a recurring job can spin idle.
- "One person is a video factory" is an official slogan, not a capacity guarantee. It describes a ceiling, not everyone's daily floor.
- Mind the derivative-work boundary when recreating hits: referencing public canvases and prompts is fine, but if your material includes other people's brands, likenesses, or music, sort out licensing before commercial use.
- Errors propagate down a long chain: if WorkBuddy misreads the task, every shot LibTV generates inherits the mistake. Manually reviewing the script and storyboard before generation is the highest-leverage intervention in the whole pipeline.
- Claims like "a short-drama studio crossed one million USD in daily revenue within a month" circulate without any official source; this article does not adopt them.
Eight. How this piece divides work with our earlier posts
Two earlier WorkBuddy articles on this site cover different ground. The WorkBuddy setup pitfalls SOP (July 23) handles the platform itself - installation, configuration, stability tuning, and credit tactics - none of which is repeated here. The Tencent Hy3 free-window article (August 10) covers the free-model angle, a benefits track. In short: how to install and tame WorkBuddy lives in the old posts; the video production flow after connecting LibTV lives in this one.
References
- LibTV official WeChat announcement on the LibTV x WorkBuddy integration, October 26, 2026 [official]: https://mp.weixin.qq.com/s/SbM-RBf1ZFPyG-CoXjlxHg
- LibTV official website (platform figures are official self-reported) [official self-reported]: https://www.liblib.tv/
- GitHub repository libtv-labs/libtv-skills (MIT license, OpenClaw skill spec, npx install command) [secondhand, reachability verified via GitHub API]: https://github.com/libtv-labs/libtv-skills
- Tahou platform profile page on LibTV (March 2026 launch and background) [secondhand]
- Community hands-on write-ups on connecting local agents to LibTV Skills via access key [secondhand]
Related reading
- WorkBuddy Setup Pitfalls SOP: covers WorkBuddy installation, configuration, and stability; this article covers the video production flow after integration.
- Tencent Hy3 Free Window on WorkBuddy: the free-model track - how to make the most of the Hy3 free period.
- Tencent Hy-Nomas Interactive World Model Open Source Deep Dive: same-batch open-source piece, from architecture to consumer-grade requirements.
FAQ
Q1: Can three yuan really produce a brand video? That is the official announcement's self-reported figure [official self-reported], not a promise. Real cost depends on video length, model tier, and re-roll counts: the cheapest model tier plus a clean first pass is what gets close to that number; repeated re-rolls and premium tiers push costs up noticeably. Test your own per-video cost on a small project first.
Q2: Can I use LibTV without WorkBuddy? Yes. The LibTV website is a complete creation platform on its own; WorkBuddy's connector simply wires it into an Agent workflow. The advanced option is connecting from any local agent through the open-source Skills and an access key, as described in section six.
Q3: Do scheduled jobs cost extra? Scheduling is a WorkBuddy feature, but every run makes real calls into LibTV and consumes account credits. Run the full pipeline manually first, confirm the per-run cost fits your budget, and only then turn it into a recurring job - otherwise idle runs can burn quota.
Q4: Will referencing hit canvases create copyright problems? The platform publishes creation canvases and prompts as an official, sanctioned path. But watch the derivative-work boundary: if your material includes other people's brand marks, likenesses, or music, confirm licensing before commercial use. An open recipe does not mean the platform has cleared every asset for you.
Q5: Which of the four request elements matters most? From experience, "what style" and "what must be preserved." A clear style keeps model tier and storyboard direction from drifting; locked-in content is what keeps slogans and product names from being rewritten. Drop any one element and you will usually pay for it with another round of rework.