PixVerse·par PixVerseVideo extend
Pixverse v6 Video-Extend
Pixverse v6 Video Extend model. High-quality video generation from image prompts.
Ouvrir dans l'espace de travail →
pixverse/v6/video-extendParamètres
| Paramètre | Type | Par défaut | Plage ou options |
|---|---|---|---|
duration durationThe duration of the generated media in seconds (1-15). | select | 5 | 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 |
seed seedRandom seed for reproducibility. | number | — | — |
quality qualityThe resolution of the generated video. | select | 720p | 360p, 540p, 720p, 1080p |
Entrées
Fichiers que ce modèle accepte en plus du prompt.
| Entrée | Accepte | Fichiers max. | |
|---|---|---|---|
Video video | video | 1 | Obligatoire |
Guide de prompt pour PixVerse
TARGET MODEL: PixVerse (AISphere video). Sequential parser — earlier tokens weigh most. Priority order is official: subject identity → action → environment → technical cues LAST.
Règles à respecter
- Text-to-video is THREE sentences: S1 = subject + defining traits + ONE action + location. S2 = ONE camera move + named style/lens/lighting/composition cues. S3 = positive stability constraints. Sweet spot 50-80 words — past ~200 the prompt dilutes its own control.
- ONE action, restrained — competing verbs produce artifacts. Show speed as physical evidence (motion blur, streaking lights); never repeat "fast".
- ONE camera move per shot, named simply in prose ("slow macro push-in", "steady lateral tracking", "low-angle tilting upward").
- KILL the praise words — PixVerse's own guide says "cinematic", "beautiful", "epic", "professional" sample too broadly. Replace each with a named physical cue: "warm rim light", "cyan-magenta contrast", "35mm", "anamorphic 2.39:1", "one-point perspective".
- The prompt box takes POSITIVE constraints only: "Hands remain natural", "Cup shape remains stable", "Product silhouette stays intact". Naming a defect noun ("no bent fingers") can summon it — convert every exclusion to its positive form.
- Don't write "realistic" as a style word — realism is PixVerse's unstyled base; named cues carry the look.
- When the user wants sound, write the audio into the shot as content ("audio includes soft room tone and faint spoon clink") — never as a toggle or promise.
Ce qui fonctionne le mieux
- Reference shape (official, verbatim rhythm): "A ceramic coffee cup sits on a dark wooden table as steam rises in slow curls. Slow macro push-in, warm tungsten side light, shallow depth of field, quiet morning cafe background. Cup shape remains stable, no text overlay, audio includes soft room tone and faint spoon clink."
- Character consistency is a vocabulary lock: keep the character's description a fixed paragraph (face, hair, outfit, build) and paste it WORD-FOR-WORD every shot — identical phrasing carries identity; drift in words is drift in face.
- Technical cues ride the tail, never the lead.
Issu du manuel LUVI pour cette famille : les mêmes règles qu'appliquent le réécriveur de l'espace de travail et le moteur de prompt MCP.