higgsfield-video-explainer

Run the video-explainer workflow through the higgsfield CLI: pick a style preset, set duration and language, generate one Seed Audio narration take and one Gemini Omni 10-second clip per block, then assemble them with the explainer_video model into a final MP4.

It creates or resolves one universal style-key image from a CMS preset, custom description, or style-reference images, then attaches that same key to every clip and repeats the same STYLE descriptor and non-photorealism negatives in every prompt.

Videos are built from fixed 10-second blocks, with total durations of one to ten minutes. Output supports 16:9 (1280x720) or 9:16 vertical (720x1280) aspect ratios, with optional burned subtitles in patrick, caveat, marker, or anton fonts.

Yes, you can pass local style-reference images with repeated --image flags or use uploaded media IDs. Only the rendering style and color grading are borrowed; people, text, logos, and objects from the references are never copied.

Avoid it for photorealistic films, ads or UGC content, talking-head videos, podcasts, motion typography reels, one-off clips without narration, or editing an already finished video. It is designed strictly for non-photoreal narrated explainers.

Only the failed block is regenerated or re-submitted; running jobs are rejoined with higgsfield generate wait rather than duplicated. Two identical failures indicate the prompt or parameters must change before retrying.