---
name: video_editing
description: Source-conditioned video flows: animate a photo, audio-driven motion, video style transfer, promptless FlashVSR video upscaling, Seedance/HappyHorse/MiniMax H3 references, stitching, orbits, dance-montage compositions, segment extend/replace, and pure-ffmpeg post-production (overlay, subtitles).
always_loaded: false
tool_names:
  - animate_photo
  - sound_to_video
  - video_to_video
  - upscale_video
  - stitch_video
  - orbit_video
  - dance_montage
  - extend_video
  - replace_video_segment
  - overlay_video
  - add_subtitles
---

# Video editing

Convert a still image, audio track, or existing clip into video, plus Seedance and MiniMax H3 multimodal references, HappyHorse image references, stitching, orbits, dance-montage compositions, segment extend/replace, and pure-ffmpeg post-production (overlay, subtitles) over previously rendered or uploaded clips.

## Tools

- `animate_photo` — photo-to-video animation with LTX/WAN/Seedance/HappyHorse/MiniMax H3 i2v and first/last-frame routing.
- `sound_to_video` — audio-synced video generation.
- `video_to_video` — video style transfer with ControlNet.
- `upscale_video` — promptless FlashVSR upscale of one uploaded or generated video to 1080p or 1440p, keeping every frame, the frame rate, the aspect ratio, and the audio.
- `stitch_video` — concatenate previously rendered clips.
- `orbit_video` — 360° orbit composition with optional dialogue.
- `dance_montage` — beat-synced dance-style composition over uploaded photos.
- `extend_video` — append new tail content to an existing video without rewriting the rest.
- `replace_video_segment` — swap a bounded time window inside a video while preserving the unchanged portion and original audio outside the replaced window.
- `overlay_video` — burn-in a static text/logo overlay onto an existing video via ffmpeg.
- `add_subtitles` — burn-in subtitle cues onto an existing video via ffmpeg.

## Constraints

- Per-clip retry and the batch progress contract are sacred — never collapse a multi-clip render down to a single waterfall call.
- `animate_photo` errors with `all_failed` must surface to the user; do not auto-retry from inside the chat loop.
- Use `upscale_video` when the user only wants a sharper, higher-resolution copy of an existing video. Pick the source with `sourceVideoIndex` (generated results are 0-based; -1 is the first upload) and optionally `targetResolution` 1080 or 1440 (default 1440, or 1080 for sources under 720p). It cannot produce 4K. Sources must be ≤768px on the short edge and at most about 1344×768 pixels (768×1344 portrait), 1-60 fps, and ≤100 MB. The server sets the maximum clip length; relay its error if a source is too long, without quoting a fixed limit. Never invent a prompt or substitute `video_to_video`/`generate_video` unless the user names a generative model such as Seedance for a re-render.
- For multiple prompt-only takes from one fixed source/end image and shared settings, prefer one Dynamic Prompt request with `numberOfVariations`/`-n`, then stitch the returned clips if the user asked for a single final video.
- Keep per-clip prompt arrays and source/end image arrays when clips need different assets, durations, audio windows, or other per-output settings.
- HappyHorse i2v uses exactly one first-frame image and r2v uses 1-9 image references; do not attach reference audio, reference video, ControlNet, or negative prompts.
- MiniMax H3 i2v uses one first frame and H3 flf2v uses both first and last frames. For a loose image/video/audio set use `generate_video` with `minimax-h3-r2v-balanced` (Balanced R2V) unless another tier is explicitly requested; r2v references are roles, not frame anchors.
- MiniMax H3 keyframes: when one H3 clip should pass through extra images at chosen moments, pass `keyframes=[{imageIndex, atSeconds}]` (up to 8; `imageIndex` indexed like `endImageIndex`, -1 = first upload) on that one `animate_photo` or `sound_to_video` call instead of chaining clips, and make the clip long enough that every keyframe falls strictly inside it. Say in the prompt what each keyframe shows at its time, and start a new shot at a keyframe that changes the angle, place, or light. Keyframes add to the first and last frame and are not references, but the prompt names each one `<Picture N>`, numbered in time order after the first/last-frame pictures. Two are included in the price; each additional keyframe adds a little. Error 4100 means no worker can pin keyframes right now: retry shortly or render without them.