vidgen-fal

name

vidgen-fal

description

The fal.ai worker for video generation; produces a Kling-rendered MP4 as the master output (always kept). Pass --webm to also encode a VP9 WebM loop alongside it (default 640px wide, ~1–3 MB for a 5s clip). Targets Kling v1.6 by default (standard tier); pass --pro for the pro tier. No router hub exists today; invoke this skill directly. Triggers: /vidgen-fal, "generate a video", "make a looping video", explicit mentions of "fal video", "Kling video". Skip for image-only tasks and non-video generation requests. Skip when the user supplies a reference image wanting image-to-video — this script is text-to-video only.

vidgen-fal

This skill is a direct-call fal.ai video worker. It calls a Kling v1.6 endpoint, downloads the MP4 (always kept as the master), then optionally runs a local ffmpeg pipeline — scale + VP9 encode — to produce a WebM alongside it. Pass --webm to opt into the WebM transcode. No router hub exists today; the user chose this as a direct-call worker.

Models

Tier	Endpoint	Cost
Standard (default)	fal-ai/kling-video/v1.6/standard/text-to-video	$0.28 / 5s, $0.56 / 10s
Pro (--pro)	fal-ai/kling-video/v1.6/pro/text-to-video	$0.49 / 5s, $0.98 / 10s

Both tiers share the same request schema (prompt, duration, aspect_ratio, negative_prompt, cfg_scale). Pro produces higher quality output at roughly double the cost.

Invocation

doppler run -p claude-code -c std --no-fallback -- bun ~/.claude/skills/vidgen-fal/scripts/vidgen-fal.ts --prompt "<text>"

Pass the file path directly — bun run is not used because it would interpret the path as a package.json script name.

Flag reference

Flag	Default	Description
--prompt <text>	required	Text description of the video to generate.
--out <dir>	~/Pictures/vidgen	Output directory for all generated files.
--duration <5|10>	5	Clip length in seconds. Enum — only 5 or 10 accepted.
--aspect-ratio <16:9|9:16|1:1>	16:9	Output aspect ratio.
--cfg-scale <0–1>	0.5	Prompt adherence (float). Values outside 0–1 are rejected.
--negative-prompt <text>	blur, distort, and low quality	Negative prompt passed verbatim to Kling.
--pro	off	Switch to the pro tier endpoint.
--webm	off	Also encode a WebM loop (VP9) alongside the MP4. Both files are kept and emitted.
--scale <number>	640	Width in pixels for the WebM output. Must be a positive even integer. Height is computed automatically. Has no effect unless --webm is set.
--webm-crf <number>	32	VP9 quality, 0–63. Lower = larger file / better quality. Out-of-range values are rejected. Has no effect unless --webm is set.

Workflow

// Gather
intent      = do("understand what motion or scene the user wants")

// Expand prompt
prompt = do("""
  Rewrite the request as a descriptive paragraph.
  Cover: subject, motion type, environment, mood, lighting, style.
  Do NOT include loop or WebM wording — that is handled by the pipeline.
""")

// Choose flags
duration_flag     = do("--duration 10 if user wants a longer clip; default 5 is fine for most")
aspect_flag       = do("--aspect-ratio 9:16 for portrait/vertical output; omit for landscape default")
cfg_flag          = do("--cfg-scale <n> only if user wants tighter (closer to 1) or looser prompt adherence")
negative_flag     = do("--negative-prompt '<text>' to override default")
pro_flag          = do("--pro if user asks for higher quality or if standard output is unsatisfactory")
out_flag          = do("--out <dir> when the user specifies an output path")
webm_flag         = do("--webm if user wants a WebM loop encoded alongside the MP4 (default: MP4 only)")
scale_flag        = do("--scale <n> to override the default 640px width (must be even, e.g. --scale 480 for smaller output)")
crf_flag          = do("--webm-crf <n> to override the default quality (e.g. --webm-crf 28 for higher quality)")

// Invoke  (pipeline: fal call → MP4 kept; if --webm: ffmpeg scale + VP9 → WebM also kept)
Bash(doppler run -p claude-code -c std --no-fallback -- \
  bun ~/.claude/skills/vidgen-fal/scripts/vidgen-fal.ts \
  --prompt "<prompt>" \
  [duration_flag] [aspect_flag] [cfg_flag] [negative_flag] [pro_flag] \
  [out_flag] [webm_flag] [scale_flag] [crf_flag])

// Relay output
do("print the 'video: ...' path so the user can open it")
do("if --webm was set, also print the 'webm: ...' path")
do("note the log path and elapsed time if emitted")

Notes

Source fps. Kling v1.6 standard renders at 24 fps at 1280×720 (confirmed via smoke test on a real output).

MP4 as master. The MP4 from Kling is always kept. It is the highest-quality output and serves as the source for any downstream transcode. Pass --webm to produce a WebM alongside it.

FAL_KEY is injected by doppler run -p claude-code -c std --no-fallback. It is never passed on the command line.

Output files. The script emits vidgen-<timestamp>.mp4 into ~/Pictures/vidgen/ (created automatically) or the directory given by --out. When --webm is set, a matching .webm is also produced in the same directory. A log.jsonl entry is appended per run.

If you opt into --webm

WebM size. A 5s source at 24 fps × 1280×720 scaled to 640×360 with VP9 crf 32 produces roughly 1–3 MB. Compare that to the 144 MB GIF the same source produced — WebM is the practical format for looping video in HTML5 contexts.

Looping in HTML5. WebM loops automatically with <video loop autoplay muted playsinline>. No special encoding or server configuration is required.

Tuning levers. Use --scale 480 to cut size further at the cost of resolution. Use --webm-crf 36 to reduce file size at the cost of visible compression artifacts. Use --webm-crf 24 for higher quality at a larger file size. --scale and --webm-crf are independent.

Convert MP4 to WebM later

If you have the MP4 and want to produce a WebM after the fact, use the same pipeline the script runs internally:

ffmpeg -i in.mp4 -vf scale=640:-2 -c:v libvpx-vp9 -crf 32 -b:v 0 -row-mt 1 -pix_fmt yuv420p -g 1 -keyint_min 1 -an out.webm