People treat AI video like a slot machine — prompt, generate, hope. That works for one clip. For a sequence it’s a waste of credits, because consistency comes from structure, not luck.

Why Storyboard Before You Generate?

Two reasons. Cost: reference generations burn more credits, so you want a plan that minimizes re-rolls. Consistency: the model anchors on whatever reference you give it, so the references have to be decided before the first generation, not patched in later.

A simple three-column board — shot, reference, action — is enough. You don’t need storyboard illustrations.

How Do You Structure the Reference Set?

• Character sheet: one clean front-facing portrait per recurring character.
• Location sheet: one wide establishing image per set.
• Prop sheet: anything that appears in multiple shots and must look identical.

Keep the set small. Six or seven references for a one-minute short is plenty; piles of near-duplicates confuse the model more than they help.

What Does a Nine-Shot Short Look Like?

1. Establishing wide — location reference only.
2. Character medium — character + location.
3. Character close-up — character reference, tight framing.
4. Insert shot of a prop — prop reference.
5. Two-shot — both characters.
6. Action beat — character + location.
7. Reaction — character close-up.
8. Wide payoff — location.
9. End card.

Note the rule: use the character close-up sparingly, because that’s where drift shows most.

How Do You Keep Shots Looking Related?

Repeat the lighting description in every prompt. “Soft window light from the left” in shot two and “harsh noon sun” in shot three will break the illusion faster than any facial drift. Lock camera language too — if shot one is a “35mm look,” keep that phrase throughout.

What If a Shot Breaks?

Don’t re-prompt from scratch. Go back to the reference, generate two variants, and pick the closer one. If it still fails, change the framing angle rather than the content — a different angle often sidesteps whatever the model is struggling with.

Budget two shots of re-rolling per ten generated. Anyone claiming first-try perfection is either lucky or not looking closely.

Quick Summary

• Decide all reference images before generating, not during.
• Keep the reference set small: character, location, prop sheets.
• Repeat lighting and camera language in every prompt to keep shots related.
• Re-roll from the reference or change angle instead of rewriting the prompt.

Consistency is the problem that has kept AI video out of real production. You generate shot one, then shot two, and the character’s face has quietly changed. Vidu Q2 attacks that directly with reusable reference subjects.

What Is Vidu Q2’s Consistency Feature?

You upload a reference image for a character or a scene and attach it to each generation. The model locks the subject’s appearance and reuses it across shots, so a three-shot sequence keeps the same face, outfit, and set.

It’s not perfect — drift creeps in over many shots and strong changes in pose or lighting — but it’s far better than prompting a name and hoping.

How Do You Build a Consistent Sequence?

1. Lock your character reference first. Use a clean, well-lit portrait on a plain background.
2. Generate a “master shot” and keep it. This becomes your visual anchor.
3. For each new shot, attach both the character reference and the master shot.
4. Describe only what changes: the camera angle, the action, the location.
5. Generate two variants per shot and pick the closest match to your anchor.

What Does Vidu Q2 Cost?

• Free tier: a monthly credit allowance with watermarks.
• Paid: subscription tiers from roughly $10-$30/month depending on credits and resolution.
• API: per-generation pricing for developers, with volume discounts.

Reference-based generations cost more credits than plain text-to-video, so plan the bill around how many shots you actually need.

How Does Consistency Compare?

Against competitors, Vidu Q2 leads on subject consistency for character-driven shorts. Veo still produces more photoreal motion and better physics. Luma Ray3 wins on color range. Where Vidu stands out is a coherent cast across a sequence — the thing you need to tell a story rather than post a single loop.

Where Does It Still Drift?

Three situations. Large pose changes. Dramatic lighting shifts between shots. And close-ups after wide shots, where the model re-interprets facial detail. Reducing shot-to-shot contrast in framing helps a lot.

Also expect the usual limits: fast motion blurs, hands are hit or miss, and on-screen text is unreliable.

Quick Summary

• Vidu Q2 uses reference images to keep characters and scenes consistent across shots.
• Clips run 4-8s; free tier is watermarked, paid from ~$10-$30/month.
• Leads on multi-shot consistency; Veo leads on physics, Ray3 on color.
• Drifts on big pose changes, lighting shifts, and wide-to-close-up cuts.