Pixverse V6 Reference to Video

Pixverse V6 Reference to Video — Image-to-Video Generation

What is Pixverse V6 Reference to Video?

Pixverse V6 Reference to Video composes a brand-new video scene from reference material instead of animating a single still. You supply up to 10 reference images — characters, products, or backgrounds — and up to 2 reference video clips, tag each in your prompt (@image1, @video1), and the model generates a fresh shot that keeps every referenced element on-model. It is PixVerse's reference-to-video (Fusion/omni) mode in the V6 family, built to cut the identity drift that single-image image-to-video shows over a 15-second window.

Key Features

• •Multi-image referencing: up to 10 tagged subject, character, or background images in one request, each addressed by @imageN.
• •Reference videos (omni mode): up to 2 clips (15 seconds total) that carry motion, camera movement, and visual style — enabling motion imitation and subject replacement.
• •Consistency: keeps characters, scenes, and motion coherent across the clip, reducing visual drift.
• •Native audio: optional synchronized soundtrack generated in the same pass.
• •Flexible output: 1–15 second duration, 360p–1080p quality, and auto or fixed aspect ratios.
• •Reproducibility: a seed for repeatable results, synchronous MP4 response.

Best Use Cases

Reach for reference-to-video when specific, recognizable elements must stay exact: product teasers and ad hooks where a branded item must look right; recurring characters across campaign variations and short-drama prototypes; multi-subject scenes that place a chosen character with a chosen prop or environment; and motion matching, where a reference clip's camera move is applied to a new subject. In testing, a mechanic and a cafe-racer from two separate images held their identities through a roll-out, mount, and engine kick — a two-reference composition single-image models cannot do.

Prompt Tips and Output Quality

Name each reference (@image1, @video1) and leave a space after every tag — a trailing comma triggers a ref-name error. Describe one clear action and camera move, and give each subject its own clean-background image for the strongest identity hold. Draft at 540p, then move to 720p or 1080p for finals. With a reference video, duration is ignored and the output matches the longest clip; enabling audio or a reference video both raise the per-second rate. Complex action and talking close-ups may need retries.

How to Use Pixverse V6 Reference to Video

Pixverse V6 Reference to Video turns a set of references into one coherent clip. The workflow is: gather clean reference images (and up to two motion clips), tag each in the prompt, describe one clear action, then pick resolution, duration, and audio.

Writing the prompt

Name every reference by its tag and leave a space after it — @image1 rolls @image2 out of the garage. A trailing comma or punctuation stuck to a tag causes a ref-name error. State the subject, the action, and the camera move in literal terms; avoid stacking several simultaneous actions.

Choosing references

Give each subject or object its own image on a clean background so the model can hold its identity. Use reference_images (up to 10) for characters, products, and backgrounds. Use reference_videos (up to 2, 15 seconds total) when you want to borrow motion, camera movement, or style from an existing clip.

Recommended Settings

These are the values tested end to end for this model:

• •Quality: 540p for drafts, 720p or 1080p for finals (higher raises the per-second price).
• •Duration: 5s is a good default; ignored when a reference video is used (output matches the longest clip).
• •Aspect ratio: auto follows the references; set 16:9 for landscape or 9:16 for vertical social.
• •Audio: enable generate_audio_switch for a synchronized track; expect occasional re-rolls.
• •Seed: set a fixed integer to reproduce a result.

Example prompts to try

Each example below ran end to end at the settings shown; what carried over is the reference identity, the framing, and the motion.

• •Multi-image consistency (default). Prompt: "@image1 rolls @image2 out of the garage bay, throws a leg over the saddle and kicks the engine to life; exhaust smoke curls up and the rear wheel spins." A mechanic and a cafe-racer from two separate images both stayed on-model through the roll-out and mount. Tip: name both subjects and keep each reference on a plain background.
• •Environment reference. Prompt: "@image1 leans over the planting benches inside @image2 and mists a row of ferns with a brass sprayer while water droplets scatter and steam drifts between the glasshouse arches." The referenced greenhouse held as the backdrop while the botanist acted. Tip: pass the setting as a second image to lock the environment.
• •Native audio. Prompt: "@image1 drives a fast rock groove, sticks snapping across the snare and ride cymbal, hi-hat chattering, the kick drum pulsing." The drummer played on-model with a synchronized percussion track. Tip: set generate_audio_switch true; re-roll if the audio is rough.

Reference-guided motion

This model takes up to two reference videos alongside your images. Address the motion clip as @video1 and the new subject as @image1.

• •Motion imitation / subject replacement. Prompt: "@image1 performs the jump and spin from @video1 and lands in a low crouch on the gym floor." The referenced gymnast performed the clip's exact crouch-spring-flip-land, with the original performer swapped out. Tip: with a reference video, duration is ignored and the output matches the clip length.

Further scenes the model is built for: "place @image1 and @image2 together in one shot", "apply the camera move in @video1 to @image1", or "keep @image1 consistent across a longer 1080p take".

How many references can I use?

Up to 10 images and 2 video clips (15 seconds total) per request.

How do I name references in the prompt?

Tag them @image1, @image2, @video1 — with a space after each tag.

Does it generate audio?

Yes. Enable the audio switch for a synchronized native soundtrack in the same pass.

What is the maximum length?

15 seconds. With a reference video, the output matches the longest clip's length.

Can it keep a character consistent across a scene?

Yes — multi-image referencing anchors appearance to reduce drift over the clip.

How is it different from image-to-video?

Image-to-video animates one still; reference-to-video composes a new scene from multiple tagged references.