Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

ffmpeg video mixing · varghq | DecimalAI

varghq/ffmpeg-mixing

SkillScore100Maintained

Mix, trim, and concatenate video clips with ffmpeg without audio/video desync. Use when stitching generated clips into an original video, inserting scenes at timestamps, or any ffmpeg filter_complex work involving trim/concat with a continuous audio track.

design-mediaby varghq

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Coding pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/varghq-ffmpeg-mixing/SKILL.md

decimalai skills pull varghq-ffmpeg-mixing --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/varghq-ffmpeg-mixing/SKILL.md https://app.decimal.ai/s/varghq-ffmpeg-mixing/SKILL.md

Raw file: /s/varghq-ffmpeg-mixing/SKILL.md ↗ · machine-readable summary at /s/varghq-ffmpeg-mixing.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 38d ago

Yes — +32% lift+3% tokens0% turns100% of whole cases pass with skill (22/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-01	✗→✓	▲ Improved	-10%	0%
case-02	✗→✓	▲ Improved	-14%	0%
case-03	✗→✓	▲ Improved	-35%	0%
case-05	✗→✓	▲ Improved	4%	0%
case-06	✗→✓	▲ Improved	-18%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

varghq / sdk

.claude/skills/ffmpeg-mixing/SKILL.md

View on GitHub ↗

License MITImported 8/3/2026Version v1

Skill contents

SKILL.md

ffmpeg video mixing

Lessons for mixing video clips with ffmpeg while keeping audio and video in sync.

problem: audio/video desync when mixing clips

what went wrong

using -ss X -t Y to pre-trim input, then applying relative trim filters caused timing drift:

bash
# BAD: relative timestamps after pre-trim
ffmpeg -ss 64 -t 36 -i original.mp4 ...
  -filter_complex "
    [0:v]split=5[orig1][orig2]...;
    [orig1]trim=0:4,setpts=PTS-STARTPTS[o1];
    [orig2]trim=4:11,setpts=PTS-STARTPTS[o2];
    ..."

this produced wrong duration (38s instead of 36s) with audio desync.

solution: use absolute timestamps from full input

trim directly from full original using absolute timestamps:

bash
# GOOD: absolute timestamps from full file
ffmpeg -i original.mp4 -i scene1.mp4 -i scene2.mp4 ...
  -filter_complex "
    [0:v]trim=64:68,setpts=PTS-STARTPTS[o1];
    [1:v]scale=1280:720,trim=4:6,setpts=PTS-STARTPTS[s1];
    [0:v]trim=70:75,setpts=PTS-STARTPTS[o2];
    ...
    [o1][s1][o2]...concat=n=N:v=1:a=0[outv];
    [0:a]atrim=64:100,asetpts=PTS-STARTPTS[outa]
  "
  -map "[outv]" -map "[outa]"

key points

absolute timestamps: trim from full input file, not pre-trimmed
separate audio handling: use atrim on audio stream independently
setpts reset: always use setpts=PTS-STARTPTS after trim to reset timestamps
scale before trim: when mixing different resolutions, scale first then trim
video duration = audio duration: ensure total video segments match audio segment length

example: inserting clips into original

to insert generated clips at specific timestamps while keeping continuous audio:

bash
ffmpeg -y \
  -i original.mp4 \
  -i generated-scene.mp4 \
  -filter_complex "
    [0:v]trim=START1:END1,setpts=PTS-STARTPTS[o1];
    [1:v]scale=1280:720,trim=0:DURATION,setpts=PTS-STARTPTS[s1];
    [0:v]trim=START2:END2,setpts=PTS-STARTPTS[o2];
    [o1][s1][o2]concat=n=3:v=1:a=0[outv];
    [0:a]atrim=AUDIO_START:AUDIO_END,asetpts=PTS-STARTPTS[outa]
  " \
  -map "[outv]" -map "[outa]" \
  -c:v libx264 -preset fast -crf 18 \
  -c:a aac -b:a 192k \
  output.mp4

timestamps must add up: (END1-START1) + DURATION + (END2-START2) = AUDIO_END - AUDIO_START

Illustrative Want to see what varghq/ffmpeg-mixing would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/21/2026, 11:38 UTC · 22 cases

Lift

+32%

100% vs 68%

Agent turns

1.0 vs 1.0

Tokens

+3%

2,158.59 vs 2,089.73

fail → pass · 7pass → pass · 15

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 I have a main MP4 file and two secondary B-roll video clips. I want to write an ffmpeg filter complex command that inserts these B-roll clips at specific points along the main video timeline while keeping the original background audio track playing continuously without getting out of sync. Can you generate the full ffmpeg command for me?	fail→pass	16,901	9,270	-45%	1	1	0%	3,205	2,869	-10%	0	0	—
case-02 I'm trying to splice together parts of a long video recording with a few external scene files, but my current attempts keep causing audio desynchronization and incorrect duration outputs. How should I structure an ffmpeg command to cut and combine these segments while preserving precise audio and video alignment throughout?	fail→pass	19,084	11,382	-40%	1	1	0%	3,550	3,057	-14%	0	0	—
case-03 I am writing an ffmpeg script to extract three segments from a 10-minute 1080p source video input.mp4 and combine them into a single video file output.mp4. To optimize processing speed, I am considering using -ss and -t before -i input.mp4 on the input side and then using relative trim filters. How should the timestamp trimming be structured in ffmpeg to prevent timing drift?	fail→pass	22,269	9,138	-59%	1	1	0%	4,046	2,610	-35%	0	0	—
case-04 I wrote an ffmpeg command to cut two sections from a video file and join them using concat=n=2:v=1:a=1. The output plays, but seeking is broken and the second segment stutters upon transition. What filter step must be applied immediately after every trim operation?	pass→pass	6,004	3,665	-39%	1	1	0%	1,086	1,491	+37%	0	0	—
case-05 I have a main 4K MP4 recording main.mp4 and a 1080p clip insert.mp4. I need to slice a 5-second section from insert.mp4 and mix it into main.mp4 using ffmpeg filter_complex. What order should the scale and trim filters be chained for insert.mp4?	fail→pass	10,176	5,981	-41%	1	1	0%	1,791	1,866	+4%	0	0	—
case-06 I need to construct an ffmpeg command where three video clips from different sources are concatenated into a single output video stream, while preserving continuous audio extracted from the primary source clip. How should the audio track be isolated and processed?	fail→pass	14,313	7,980	-44%	1	1	0%	2,677	2,208	-18%	0	0	—
case-07 I am building an automated ffmpeg command generator that cuts a 100-second original video into two sections (0 to 30s and 45 to 100s) and inserts a 15-second external clip in between. I need to ensure the audio track matches the exact total video output duration. What mathematical duration relationship must hold between video segments and audio trim parameters?	pass→pass	15,720	8,000	-49%	1	1	0%	3,058	2,282	-25%	0	0	—
case-08 Generate an ffmpeg command to cut video segment A (0 to 10s) and segment B (20 to 30s) from original.mp4, insert scene1.mp4 (scaled to 1280x720) between them, and output a synced output.mp4 with continuous audio from original.mp4 spanning 0 to 30s duration.	pass→pass	12,053	7,583	-37%	1	1	0%	2,365	2,360	-0%	0	0	—
case-09 When trimming multiple distinct non-overlapping video segments from a single video stream input [0:v] in ffmpeg filter_complex, how should the input stream be handled across multiple trim calls?	fail→pass	12,853	11,288	-12%	1	1	0%	2,512	2,750	+9%	0	0	—
case-10 In an ffmpeg filter_complex command that joins three trimmed video streams [v1], [v2], [v3] into [outv] and a trimmed audio stream [outa], what -map flags should be passed to the final output file to select these streams?	pass→pass	5,144	3,318	-35%	1	1	0%	854	1,397	+64%	0	0	—
case-11 I want to replace a 10-second middle portion (15s to 25s) of main.mp4's video track with scene.mp4 (which is 10 seconds long), while keeping the original main.mp4 audio uninterrupted across the entire 0s to 40s duration. How should the video trim ranges and audio trim ranges be specified in ffmpeg?	fail→pass	10,095	7,221	-28%	1	1	0%	2,056	2,344	+14%	0	0	—
case-12 When working with secondary input video clips in ffmpeg filter_complex that have different frame sizes, what is the exact filter chaining syntax for adjusting resolution before cutting the clip?	pass→pass	10,696	6,758	-37%	1	1	0%	1,920	2,046	+7%	0	0	—
case-13 After applying the atrim filter to an audio stream in ffmpeg filter_complex, what specific audio filter must be chained immediately afterward to reset presentation timestamps?	pass→pass	3,680	2,470	-33%	1	1	0%	506	1,205	+138%	0	0	—
case-14 If video trim segment 1 is 0 to 5 seconds, video trim segment 2 is 10 to 20 seconds, and an inserted scene clip is 8 seconds long, what range length must be provided to atrim for the background audio starting at timestamp 0?	pass→pass	5,936	3,884	-35%	1	1	0%	1,105	1,538	+39%	0	0	—
case-15 When concatenating 3 video-only stream labels [o1][s1][o2] in ffmpeg filter_complex while keeping audio mapped from a separate stream, what parameter values should be passed to the concat filter?	pass→pass	7,413	3,370	-55%	1	1	0%	1,466	1,446	-1%	0	0	—
case-16 In an ffmpeg command replacing multiple sections of video while maintaining background audio from the original input file, should audio be trimmed using relative input seek flags or directly inside filter_complex?	pass→pass	13,327	6,306	-53%	1	1	0%	2,204	1,925	-13%	0	0	—
case-17 Why does using -ss 30 -t 20 on an input file prior to passing it to filter_complex trim filters cause timing drift and wrong output duration in ffmpeg?	pass→pass	17,094	15,073	-12%	1	1	0%	2,748	3,234	+18%	0	0	—
case-18 Provide an ffmpeg command snippet for mixing video clips that encodes the final output using H.264 video with fast preset, CRF 18, and AAC audio at 192k bitrate.	pass→pass	6,863	4,069	-41%	1	1	0%	1,313	1,608	+22%	0	0	—
case-19 How should labeled stream pads (such as [o1], [s1], [o2]) be structured when feeding multiple trimmed video segments into the concat filter in an ffmpeg command?	pass→pass	13,900	9,624	-31%	1	1	0%	2,540	2,499	-2%	0	0	—
case-20 How do I extract a single PNG thumbnail image from a video at timestamp 00:01:30 using ffmpeg?	pass→pass	5,910	4,068	-31%	1	1	0%	1,059	1,458	+38%	0	0	—
case-21 How do I re-encode an MP4 video file to reduce its file size using two-pass AV1 encoding in ffmpeg?	pass→pass	12,865	12,692	-1%	1	1	0%	2,211	3,279	+48%	0	0	—
case-22 How do I burn hardcoded SRT subtitles into an MP4 video stream using ffmpeg?	pass→pass	9,469	6,477	-32%	1	1	0%	1,702	2,017	+19%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +32 percentage points is the difference between those two pass rates over the 22 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 68.2%
Pass rate with the skill: 100%
Cases attempted: 22
Cases counted toward lift: 22
Cases that regressed: 0
Skill version measured: v1
Run completed: 8/21/2026, 11:38 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-01

Prompt

I have a main MP4 file and two secondary B-roll video clips. I want to write an ffmpeg filter complex command that inserts these B-roll clips at specific points along the main video timeline while keeping the original background audio track playing continuously without getting out of sync. Can you generate the full ffmpeg command for me?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run100

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+14% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 37 days ago

Other measured skills in the registry, with their headline benchmark lift.