Vidu Q1 vs Q2: What Actually Changed
One generation apart, and the difference is narrower than the marketing suggests — but it lands exactly where multi-shot work succeeds or fails.
Vidu Q1 launched globally on 21 April 2025 with 1080p output, natively generated synchronised audio at 48 kHz, cinematic effects and multi-reference support. Vidu Q2 arrived on 25 September 2025 and pushed in a different direction: facial performance, reference conditioning, faster inference and 4K.
Both are selectable in the OpenFilmo generator panel, which is the only comparison that matters in practice. Run the same prompt and the same references through each and judge on your own footage — demo reels are chosen for their best frames, not their average ones.
Where the two generations differ
Four differences that change a decision.
Q2: performance
Micro-expression modelling for blinks, gaze and mouth movement. This is the change most people notice, and the one hardest to judge from a still.
Q2: references and speed
Reference-to-video with multiple subject inputs, at an inference speed Shengshu describes as roughly three times the Q1 reference workflow.
Q1: a flatter, more literal look
Q1 tends to move more mechanically. That is a defect in a performance shot and occasionally an advantage when you want plain, unemphatic motion.
Both: the same credit rate here
Model choice does not change the price. A five-second clip is five credits either way, so comparing them costs the same as generating with one.
How to settle it with your own footage
A comparison that costs credits, not a day.
Fix one prompt and one reference set
Change a single variable. If you change the prompt and the model together you learn nothing about either.
Generate the same short beat on both
Keep it to one beat and a few seconds. The difference between the generations shows up in facial timing within the first second.
Judge in motion, not on stills
Export both and play them at speed. Micro-expression work is designed to be read in motion and often looks unremarkable frame by frame.
Default to Q2 and keep Q1 as the exception
Q2 is the better default on anything with a person in it. Reach for Q1 deliberately rather than by habit.
Which jobs pick which model
By shot type rather than by benchmark.
Anything with a face
Q2. The performance modelling is the entire point of the newer generation and it is where the difference is visible.
Scenery, product and silhouette shots
Either. Without a face in frame, Q2’s advantage largely disappears and the simpler generation is not a downgrade.
Iterating on a prompt
Use the faster mode for exploration regardless of model, then render the approved take in the cinematic mode.
Reproducing an older look
Q1, when a previous project’s output has to match and the earlier generation’s behaviour is part of that match.
Why this comparison matters less than it looks
The models differ. The workflow problem does not.
Choosing between Q1 and Q2 is a real decision, but it is a smaller decision than it appears. Neither model solves continuity on its own, and both reduce drift only when the reference images are clean and consistent. A team with good references will get usable sequences from either generation.
The honest ranking is that Q2 is better on the thing that usually breaks a sequence — facial continuity across shots — and that everything else is close enough that the difference is unlikely to be the reason a project succeeds. If you are deciding where to spend attention, spend it on references rather than on the model toggle.
There is also a version trap worth naming. Model generations move every few months. Building a workflow that depends on one version’s exact behaviour means rebuilding it regularly, which is why the skill worth developing is reference preparation rather than prompt memorisation.
Credits and plans
Generation is metered in credits, shared across every tool.
A five-second video costs 5 credits and one image costs 2. The Free plan gives you 30 credits every month with no credit card, no watermark and commercial use included. Paid plans start at $9 for 400 credits and add faster generation, parallel tasks and higher export resolution.
Related guides
Longer reads on the same workflow, from the OpenFilmo blog.
How to Use Vidu: A Practical Guide
How to use Vidu: reference-to-video workflow, keeping subjects consistent across shots, prompt approach, and where it fits compare…
AI Video Cost Per Minute: The Real Numbers in 2026
How to calculate true AI video cost per finished minute: generation price, regeneration waste, edit time and iteration, with worke…
Best AI Video Generators 2026: 15 Tools Tested
We tested 15 AI video generators on the same prompts and ranked them for 2026. Compare quality, speed, credit pricing and commerci…
Related workflows
Closely related tools and guides on the same problem.
Frequently asked questions
Is Q2 simply better than Q1?
On facial performance, yes. Elsewhere the gap is small enough that shot type matters more than the version number — a scenery shot will not improve much either way.
When was each released?
Q1 launched globally on 21 April 2025. Q2 was released on 25 September 2025, with reference-to-video following on 21 October 2025.
Does Q2 cost more here?
No. Model choice does not change the credit rate — five credits per five-second clip for either, drawn from the same balance.
Can I mix them in one project?
Yes, and it is often sensible: Q2 for the shots where a face carries the frame and Q1 where it does not.
What matters more than the model choice?
The reference images. Clean, consistent, evenly lit references do more for multi-shot continuity than any version difference between these two generations.