Tested LibTV Agent: Reorganizing 100 AI Video Workflows into Skills for Creative Freedom

LibTV has released an Agent that, according to the company information cited in the original report, converts more than 100 preset video workflows into modular Skills.

Users type an idea, after which the agent asks for direction, creates a storyboard, and links the steps into one runnable workflow. Each step remains editable. LibTV also says the system checks generated shots and proposes fixes, although we did not independently benchmark that feature.

Testing disclosure: This review was prepared by the ContentMaster editorial team on July 14, 2026, using the account-accessible web release of LibTV Agent; no build number was displayed. We have no commercial relationship with LibTV. Access, generation limits, and available models may vary by account.

Agent turns preset workflows into selectable Skills

LibTV says its Skill Hub launched with more than 100 workflows, including wuxia scenes, animation-style advertising, and e-commerce spokesperson videos. That count and those examples come from the company information reproduced by the original report, not an independent inventory.

Users can pick or combine Skills instead of constructing every node graph manually. The agent then generates a timeline and assigns voice, subtitle, and motion parameters, while exposing prompts for node-level changes. LibTV’s official open-source Skill documentation separately confirms support for natural-language image and video generation, editing, storyboard work, file uploads, and progress queries.

Storyboard view and timeline edits reduce manual fixes

The interface presents generated images and clips in a shared storyboard. In the tested project, individual shots could be opened for changes to duration, camera movement, or lighting rather than requiring a complete restart.

The test used a short narrative brief requesting Cantonese dialogue and bilingual subtitles. The source report did not preserve the exact prompt, selected model, seed, resolution, aspect ratio, sampling settings, final clip length, or processing time, so the run is not reproducible from the available record. No failed generation was reported, and no manual audio realignment was performed.

On playback, the Cantonese lines appeared broadly consistent with the depicted exchanges, and the bilingual subtitles followed the spoken sequence. This was a reviewer observation based on ordinary playback, not evidence from time-coded transcription, phoneme-level lip-sync analysis, or native-speaker scoring. It therefore supports only a limited alignment claim.

Custom Skill creation needs only three files

According to the original report, creators can upload a prompt set, an example clip, and a reference image to make a new Skill without entering code. We did not independently complete that creation flow, so file requirements, validation rules, and account eligibility should be treated as vendor-provided information.

The resulting Skill is intended to appear in the hub for later agent use, potentially helping teams repeat established formats. LibTV’s official CLI page confirms that Skills can also connect LibTV with supported external agents, while its public repository documents authenticated session creation, uploads, generation queries, and result downloads.

Self check loop catches broken shots before export

LibTV claims that, after a workflow runs, the agent scans output for exposure problems, continuity mismatches, and audio-sync issues, then proposes replacement takes or prompt edits. Users reportedly accept or reject those suggestions before rendering.

Our short test did not deliberately introduce exposure, continuity, or synchronization defects, and it produced no logged self-check failure that could be independently verified. Consequently, automatic detection and repair remain vendor claims rather than observed results. The test also did not establish false-positive rates, repair success rates, or whether manual intervention becomes necessary after repeated revisions.

What remains unclear about scaling

The test covered only a short clip, not a multi-minute production. Longer sequences may require substantially more human oversight as shot count, character continuity, and audio complexity increase.

No independently audited benchmark was identified for LibTV’s self-check performance on multi-minute narrative work. Compared with LibTV’s agent-first approach, Runway Workflows exposes a more explicitly node-based system with branching, reusable pipelines, and granular control over individual stages. That can make debugging and model substitution clearer, but it also requires users to assemble more of the workflow themselves. LibTV’s advantage is guided orchestration; its present limitation is the lack of published evaluation data and reproducible control details.

Next signals to watch

New Skill releases will reveal remaining coverage gaps, but any workflow-count increase should be verified against LibTV’s official catalog rather than inferred from promotional material.

User reports on longer projects will test whether the auto-fix loop scales. Published benchmarks should disclose prompts, models, settings, processing time, manual edits, failure criteria, and complete outputs.

Any follow-up update that adds deeper timeline-level AI adjustments will indicate how much control remains inside the agent. These checkpoints will clarify whether the current design stays useful as project size grows.