Communitygithub.com

Newdvn2025/topic-to-published-video

Generate, extend, and download B-roll videos via the AutoDL Art API; supports text, reference images, and first/last frame input. Designed for user-authorized video material generation tasks.

What is topic-to-published-video?

This skill automates B-roll video generation through AutoDL's Art API, handling text prompts, reference images, and first/last frames. It first analyzes speaking footage, asset gaps, and user budget to create a manifest, then performs dry-run validation before user authorization to avoid unnecessary costs. After approval, it submits requests, tracks task IDs to prevent duplicate payments, and supports resuming incomplete tasks. Compatible with platforms like Claude Code and Cursor, it runs as a Python 3.10+ script with no third-party dependencies, reading API keys from environment variables or config files. The skill validates generated files for non-emptiness, correct parameters, and semantic accuracy, and points to detailed prompting guides and manifest examples for quick setup.

Works with~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/Newdvn2025/topic-to-published-video/tree/HEAD/addons/autodl-broll

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

AutoDL B-roll

First, examine actual speaking footage, material gaps, and user budget to generate a manifest. Users should not submit until the plan is ready. Default text generation workflow: minimax_h3_lightx2v_no_pic, 4–5 seconds, 768p landscape; for action demonstrations, extend duration as requested by the user. The generated aspect ratio may not be exactly 16:9; after download, verify and derive outputs per project requirements while keeping the original clip.

Configuration & Execution

Python 3.10+, no third-party Python dependencies. Read the API token first from the environment variable AUTODL_ART_TOKEN, then from topic-to-published-video/autodl.env in the user's config directory; use --env-file to specify another file. Environment variables take precedence. Never place the key in chat, command arguments, manifest, or state files.

Relative reference image paths are resolved from the manifest directory; local images are converted to data URLs.

python3 scripts/autodl_broll.py manifest.json --prepare-only --output-dir output

This command only validates and saves the request body, without calling the API. After user authorization:

python3 scripts/autodl_broll.py manifest.json --output-dir output

The task ID is saved immediately; re-running with the same output directory continues the recorded task without submitting and paying again. Failed tasks are not automatically retried; for additional generation, use a new ID and output directory, and obtain corresponding budget authorization. Do not modify the content of an already-submitted request for the same ID.

Workflows & Acceptance

Text: minimax_h3_lightx2v_no_pic. Reference image: minimax_h3_z0902, requires at least ref_image_0. First/last frame: minimax_h3_b99_002, first_frame and last_frame are mandatory. For more workflows and exact enumerations, see the API documentation; do not mix resolution values across different workflows.

API availability and pricing depend on the current account and service; fixed unit prices or guarantees of being cheaper than other services are not given. After a request returns SUCCESS, still verify non-empty files, parameters, and complete action; the flow director is responsible for semantic acceptance.

When crafting prompts, read the prompting guide; to start, copy the text example or first/last frame example.

Related Skills