Introducing PixVerse C1 Reference-to-Video on WaveSpeedAI

Lock Your Subject and Background, Generate the Video

A persistent weakness in AI video has been subject consistency — generate a clip of “a woman in a red coat walking down a Paris street,” and two seconds later it’s a different woman in a different coat. Reference-driven generation fixes this: you supply the subject, you supply the background, and the model composites them into video.

We’re excited to announce that PixVerse C1 Reference-to-Video is now live on WaveSpeedAI — production REST API, fine-grained reference control via @ref_name prompt syntax.

What Is PixVerse C1 Reference-to-Video?

PixVerse C1 Reference-to-Video is the reference-guided variant of the PixVerse C1 flagship video model. Instead of generating purely from text or a single start image, you upload reference images with names, and then cite them in your prompt using @ref_name syntax — the model composes a video that keeps the subject and background consistent with what you uploaded.

Example:

Upload ref_1.png (a specific character) and ref_2.png (a specific beach). Prompt: “@ref_1 walks along @ref_2 at sunset, slow tracking shot from behind.” Output: a 10-second clip where the character looks like ref_1 and the environment looks like ref_2, not some AI approximation.

This is a big deal for branded content, character-based storytelling, and any IP work where the subject has to be recognizable.

Key Features

@ref_name Prompt Syntax Reference multiple uploaded images by name in your prompt. The model resolves each reference to the uploaded asset.

Subject + Background Consistency Keeps character identity and environment style locked across the full clip duration.

Up to 1080p, 1–15 Seconds Same quality tier as the rest of the PixVerse C1 family. Choose 360p / 540p / 720p / 1080p.

Multiple Aspect Ratios Portrait, landscape, cinematic — match your delivery format.

Optional Native Audio Enable generate_audio_switch for one-call delivery with synchronized audio.

Real-World Use Cases

Character-Driven Shorts and Web Series

Produce consistent-looking protagonist across a series of clips — same face, same wardrobe, different scenes. Essential for serial content.

Branded Content with Brand Assets

Upload brand characters, logos, or product references — the model composites them into new video scenes without drift.

Storyboarding With Locked Talent

Previs a scene where “this actor” meets “that location” without shooting either. Lock both via reference images.

IP Work and Licensed Characters

Generate marketing clips, games cinematics, or social content featuring a specific character IP, with identity fidelity.

Educational and Instructional Content

Anchor a consistent instructor or mascot across an educational video series.

Personalized Content at Scale

Apps that insert a user’s photo (as @ref_user) into cinematic scenes — birthdays, celebrations, personalized greetings.

Getting Started on WaveSpeedAI

1. Upload your references — one or more images, each with a ref_name label.
2. Write a prompt using @ref_name to direct which reference appears where.
3. Pick duration, resolution, aspect ratio, and audio — up to 15s at 1080p.
4. Submit — the WaveSpeedAI API handles the rest.

Full schema on the model page.

Pricing

Per-second pricing scales with resolution and audio:

• 360p: $0.03/sec (no audio), $0.04/sec (with audio)
• 540p: $0.04/sec (no audio), $0.05/sec (with audio)
• 720p: $0.05/sec (no audio), $0.065/sec (with audio)
• 1080p: $0.095/sec (no audio), $0.12/sec (with audio)

A 10-second 720p clip with audio runs about $0.65.

Why Run PixVerse C1 Reference-to-Video on WaveSpeedAI

• One API, 890+ models. Chain reference-to-video with upscaling, lip-sync, and editing tools.
• No cold starts. Production-grade latency.
• Per-second pricing, no minimums.
• Scales horizontally — batch-generate personalized content for large audiences.

Pro Tips

• Good reference images produce good results. Clean composition, single subject, clear lighting.
• Name references descriptively. @hero and @cafe read cleaner in prompts than @img1 and @img2.
• Use multiple references sparingly. 2–3 references per prompt produces the cleanest composition; past that, the model spreads attention.
• Describe the scene action clearly — motion is still directed by text, even with reference anchors.
• Lock identity first, style second. If you want the reference character but a different mood, keep the reference but push style via prompt keywords.

Start Creating Today

PixVerse C1 Reference-to-Video is the strongest answer in the PixVerse family for consistent-subject video at scale — and now one API call away.

Try PixVerse C1 Reference-to-Video now on WaveSpeedAI and generate video with identity you can rely on.