Models / Higgsfield / Video to video
Higgsfield video to video prompting guide
Last checked 2026-10-11Higgsfield (host platform) · Host: Kling 3.0/O3/2.6, Seedance 2.0/2.5, Wan 2.6/2.7/3.0, MiniMax H3/Hailuo 2.3, LTX-2.5, Grok Imagine 1.5, PixVerse V6, Happy Horse; own: Cinema Studio 4.0, Genjutsu
How to prompt Higgsfield in video-to-video mode. Every point links its source; community tips are marked. Your agent gets the same guide through the Atlas MCP (get_guide).
Prompt structure
edit instruction→what to keep→ref roles
- Higgsfield hosts 22 video model families (Seedance, Kling, Wan, MiniMax, LTX, Grok, PixVerse, Happy Horse) plus its own Cinema Studio 4.0 and Genjutsu; each workflow has its own schema. source ↗
What the text should describe
- Genjutsu Restyle: source video ≥4 s (first 30 s used, audio kept) + required preset_id style UUID + optional ≤5 character images; prompt is optional guidance ('Preserve the original movement and camera composition'), not a custom style. source ↗
- Kling O3 Video edit: one source 3–15.5 s; duration derived from source; prompt like 'Change the background to a sunlit garden while preserving the subject and motion.' source ↗
What not to describe
- Genjutsu has no reference-to-person mapping: describe roles in the prompt for multiple characters; no new-soundtrack control. source ↗
Duration and limits
- Kling on Higgsfield: chain scenes with the final frame as next start frame; add characters as Elements. source ↗
- Kling on Higgsfield: top-level prompt ≤2,500 chars (truncated), multi_prompt shot ≤512 chars, 1–6 shots. source ↗
Negative prompt
Supported via .
Real prompts, credited
Change the background to a sunlit garden while preserving the subject and motion.
Next
All Higgsfield modesHiggsfield model, prices & shotsFrom $0.0578/s · pricesSeedance video to videoKling video to videoRunway video to videoLuma video to video