12 articles about ai-video.
Gemini 4 Argon video input doubles to $4 per 1M tokens on Dec 1, 2026. Five FFmpeg prep levers that cut the token bill, plus the one-call API equivalent.
Convert 4K video for TikTok: downscale 10-bit AI output to 1080x1920 8-bit SDR H.264, cap the bitrate, normalize audio. FFmpeg command plus one API call.
FFmpeg onnx dnn filter in 9.0: the real dnn_processing chain, NCHW float32 model limits, silent CUDA-to-CPU fallback, and what hosting it actually costs.
Veo 3.1 deprecated on the Gemini API: three preview model IDs die Oct 22, 2026. The exact migration to gemini-omni-1.1-flash, plus the n8n and Make swap.
gemini-omni-flash-preview was deprecated September 30, 2026. Swap to gemini-omni-1.1-flash in n8n, Make or curl, then reframe the 1080p output to 9:16.
TikTok's AI generated label isn't automatic: set it with is_aigc in the Direct Post call, then burn a visible disclosure that survives your re-encode.
The Sora API shutdown alternative for n8n: swap in an async Veo 3.1 or Kling 3.0 call, then normalize the output with reframe, concat, captions and audio mix.
Extract last frame FFmpeg commands return the wrong frame after a fast seek. Use -sseof with -update 1, then join Veo 3 and Sora 2 clips with no seam.
Veo and Sora render only 16:9 and 9:16. Get the AI video aspect ratio math and FFmpeg crops that turn one 9:16 generation into 4:5, 1:1, and 16:9 files.
AI avatar video automation stalls after the talking head renders. The assembly chain: stitch b-roll, burn captions, add a music bed, and ship 9:16 vertical.