Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add genfeedai/skills --skill video-model-benchmarkgit clone --depth 1 https://github.com/genfeedai/skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/genfeedai/skills/video-model-benchmark)<a href="https://agentmods.dev/skills/genfeedai/skills/video-model-benchmark"><img src="https://agentmods.dev/badge/skills/genfeedai/skills/video-model-benchmark/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/genfeedai/skills/video-model-benchmark"><img src="https://agentmods.dev/badge/skills/genfeedai/skills/video-model-benchmark.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00080 | $0.01676 |
| Opus 5.5 | $0.00032 | $0.00670 |
| Sonnet 5.5 | $0.00016 | $0.00335 |
| Haiku 4.5 | $0.00008 | $0.00168 |
Grade A, and why
video-model-benchmark scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 91 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Video Model Benchmark
You turn one prompt into a side-by-side comparison video. Each model generates the same prompt as a 9:16 clip; the clips are composited into one square or landscape video that plays in sync, each panel labelled with its model name, the shared prompt shown. The output is made to post on X.
It sells the output, not the tooling: the viewer sees two results from one prompt and judges for themselves.
Run It
# default pair, square 1080x1080, 4 seconds, no audio
bun run scripts/compare.ts --prompt "A courier carries four flat boxes past a lamp post, handheld, daylight"
# landscape 1920x1080 with the prompt in a left column, branded footer
bun run scripts/compare.ts --prompt "..." --canvas landscape --footer "genfeed.ai"
# check the plan and the cost ceiling without spending anything
bun run scripts/compare.ts --prompt "..." --dry-run
Requires Bun 1.1+, ffmpeg and ffprobe on PATH, and REPLICATE_API_TOKEN in the environment (read from env only, never written to disk). Zero npm dependencies.
Flags
| Flag | Default | Notes |
|---|---|---|
--prompt |
required | Shared by every model, shown on the video. Keep it under ~140 characters for the square canvas (three lines max, then truncated with an ellipsis). |
--models |
veo-3.1-fast,seedance-2.0-fast |
Two or three aliases from the table below, or any owner/name Replicate id. |
--canvas |
square |
square is 1080x1080, prompt on top. landscape is 1920x1080, prompt in a left column (two models) or on top (three). |
--duration |
4 |
Seconds per clip. Snapped to what each model accepts (Veo 4/6/8, Kling 5/10). The final video is trimmed to the shortest clip. |
--resolution |
720p |
Passed to models that take one. |
--audio |
none |
--audio <alias> keeps that one model's audio track. Default is silent, so no model gets a sound advantage and it is cheaper. |
--blind |
off | Shuffles the panel order and labels them A / B (C). The key is written to manifest.json. Good for "which one is better?" posts, then reveal in a reply. |
--footer |
none | Small text in the corner, for example the handle. |
--out |
./video-benchmark |
Holds clips/, the final mp4 and manifest.json. |
--reuse |
off | Reuse clips already in clips/ instead of paying to regenerate. Use it to re-composite as another canvas. |
--clip "Label=path.mp4" |
none | Repeat 2 or 3 times to composite clips you already have. No generation, no spend. |
--model-input alias='{"seed":1}' |
none | Merge extra JSON into one model's input. |
--font |
auto | Path to a font file when none of the built-in candidates exist. |
What ships with it
5 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 91 lines · 80 tokens per session scan A 31f81a80f639
video-model-benchmark is a skill published in the GitHub repository genfeedai/skills (3 stars, last pushed today), licensed MIT. It adds 80 tokens to every session and 1,676 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-10-28.
Other skills, from other repositories
self-media-content-brief
A short planning document for deciding what content to make, for whom, on which platform, and with what evidence.
self-media-content-workflow
A general workflow for planning, creating, adapting, reviewing, and archiving social-media content.
self-media-short-video
A planning and production guide for vertical short videos on platforms such as TikTok, Xiaohongshu, and WeChat Channels.
self-media-video-publisher
A workflow for publishing a finished video to platforms such as YouTube, Douyin, WeChat Channels, and Xiaohongshu. It checks each platform's files and text, then either provides a manual publishing package or uploads only after separate permission.
self-media-platform-copywriting
A writing workflow for turning an approved topic, evidence, and brief into posts for platforms such as X, Xiaohongshu, WeChat, and short-video apps.
self-media-wechat-publisher
A publishing workflow that turns approved Markdown into formatted WeChat articles or image posts and saves them as drafts.