Skip to main content
Video is the most expensive output in Pixio. Almost every question below has the same underlying answer: decide on cheap stills first, commit to video once.

​
Choosing a model

Kling — the largest and most complete video family here, covering text-to-video, image-to-video, video-to-video, audio-driven video, and lipsync. Then compare ByteDance, Google, and PixVerse.
Image-to-video, almost always. Generate a still until the composition is right, then animate it. Text-to-video makes you pay video prices to discover the framing was wrong. Use the Create Video chained action.
Filter for video-to-video. LTX and Luma have the most depth there.
HeyGen for avatars and presenters. To sync an existing face to an audio track, Sync is the specialist.
That’s Video Agent — segment generation with character memory and a stitched export. Plan it in Cinema Studio first.

​
Control

Prompting a camera angle is unreliable. CamView lets you block the shot in 3D — place objects and figures, position the camera, keyframe the move — so framing is geometry instead of a description. Blocking is free; you only pay to render.
Video Agent carries character memory across segments. For a specific recurring subject, train a LoRA and use it to generate consistent stills, then animate those.
Storyboards in Cinema Studio — frames with their own descriptions, shot language, and per-frame references. Changing a storyboard frame is free.
Duration limits are per model and shown in the form. For anything beyond one clip, generate segments and assemble them in the Video Editor or via Video Agent.

​
Costs

Many video models bill per second, and cost also scales with resolution. A short test clip and a final render can differ by an order of magnitude.
In order of effect:
  1. Get the first frame right as a cheap image, then chain into video
  2. Test at the shortest duration the model allows
  3. Test at lower resolution before the final run
  4. Block the camera in CamView — free — instead of re-rolling prompts
No. Failed generations are never charged.

​
Finishing

The Video Editor — a real timeline. Note that rendering is app-only; it can’t be triggered through the API.
Yes — Upscale Video, Caption Video, Lipsync Video, Edit Video, and Remove BG are all chained actions on a finished video.
Check status — video is reported as its own service group, and per-model status can show one model degraded while the rest are fine. Failed runs aren’t charged, so don’t retry into a known incident.