Fix a yellow or blue color cast in a video by API (colortemperature)

To warm or cool a clip, run colortemperature in the filtergraph of Video filter. It takes a temperature in Kelvin from 1000 to 40000, with a default of 6500, and a mix from 0 to 1 that blends the result with the original. On a mid-gray test frame with ffmpeg 9.0.1, 4000 came out warm and 9000 came out cool, so a yellow-looking clip is the one to try above 6500. Treat the direction as something to confirm on your own frame.

Which colortemperature options are there?

How do I test a correction before I pay for it?

The check is free. Run a candidate such as 7500, encode a short clip with it, and compare against the source. Because the encode bills per job, not per second, trim first with video trim so each test is a few seconds long, then apply the winning value to the full clip.

curl -X POST https://api.sume.com/v1/video-filter/check \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/interview.mp4",
    "filtergraph": "colortemperature=temperature=7500:mix=0.8"
  }'

Is this the same as white balance in a camera app?

No. A camera app can sample a gray card and solve for the cast. This filter applies the number you give it, so the correction is a guess you tune by eye. For a gentler tint without Kelvin math, see sepia or blue tint with colorbalance, which uses channel-level controls.

What if only one shot in a long video has the cast?

Cut that range out with video trim, run the filter on the piece, and put it back with Timeline 1.0. A mix below 1 keeps a heavy-handed correction from looking obvious when the neighbouring shots are fine.

What does a filter job cost and where are the limits?

The docs list $0.02 per encode job, with a source of at most 300 seconds, a filtergraph of at most 2048 characters and 32 named filters. The /v1/video-filter/check route runs the same allowlist and source checks for free and returns valid and next_action; a graph that passes can still fail on the worker (bad expression, memory, time), which comes back as a structured job error.