PixVerse·by PixVerseReference to video
Pixverse v6 Reference-to-Video
Pixverse v6 Reference-to-Video model. High-quality video generation from image prompts.
Open in workspace →
pixverse/v6/reference-to-videoParameters
| Parameter | Type | Default | Range or 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 |
aspect ratio aspect_ratioThe aspect ratio of the generated video. | select | 16:9 | 16:9, 9:16, 1:1, 4:3, 3:4, 2:3, 3:2, 21:9 |
seed seedRandom seed for reproducibility. | number | — | — |
quality qualityThe resolution of the generated video. | select | 720p | 360p, 540p, 720p, 1080p |
sound soundWhether sound is generated simultaneously when generating a video. | select | true | true, false |
Inputs
Files this model takes in addition to the prompt.
| Input | Accepts | Max files | |
|---|---|---|---|
images images | image | 4 | Required |
Prompting guide for PixVerse
TARGET MODEL: PixVerse (AISphere video). Sequential parser — earlier tokens weigh most. Priority order is official: subject identity → action → environment → technical cues LAST.
Rules that must hold
- 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.
What works best
- 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.
From LUVI's own manual for this family, the same rules the workspace rewriter and the MCP prompt engine apply.