Back in February I called Seedance 2.0 the best AI video model by far.

Two months later I wanted to find out if that still holds when you put it directly against the competition.

I ran it head-to-head against Kling 3.0 across five specific tests.

Here’s what happened. But first, some catchup:

Outskill is expanding their mentor network and looking for practitioners who've built real GenAI systems.

They train senior tech professionals (10+ years exp) at India's biggest companies and need people who can teach AI hands-on, not just talk about it.

Mentors run live workshops and training sessions, fully online, across engineering, data, DevOps, cybersecurity, and more.

If you've got the experience and want to teach the next generation of AI builders, applications are open now.

1. Skills Janitor

A Claude Code plugin that audits, tracks usage, and compares your installed skills with nine slash commands, zero dependencies. It surfaces which skills you invoke, flags duplicates or broken ones, and lets you search GitHub for new ones to add.

2. GhostDesk

A Windows AI overlay that floats above any app and stays invisible during screen shares, so your notes and AI queries stay private on Zoom, Teams, Meet, and 10+ other platforms. It uses OS-level window exclusion APIs to keep your workflow uninterrupted without switching tabs.

3. Ray

A local-first AI financial advisor that connects to your bank via Plaid and answers questions about your actual spending, debt, and savings goals and not generic advice. All data stays encrypted on your machine, with PII stripped before anything reaches the AI model.

Seedance 2.0 is ByteDance's video model.

It can do native audio and video in one pass, multi-shot storytelling, up to 12 reference files per generation.

Currently top of the Artificial Analysis leaderboard for both text-to-video and image-to-video.

Available globally through Higgsfield and fal.ai.

Kling 3.0 is Kuaishou's model.

It can do native 4K at 60fps, clips up to 3 minutes, free tier of roughly 6 videos a day, and the most accessible of any current model.

Now, four tests.

Test 1: Dialogue with camera moves

Prompt: A 15-second argument scene with fast whip pans between two speakers. "You promised... Where were you? Got held up at work... What do you want from me, an apology?"

Seedance put the argument in a realistic modern kitchen. The fast, dynamic whip pans, grounded environment, feels like something from a TV show.

The problem: it scrambled the dialogue entirely, one character firing off both sides of the argument back to back.

Kling set the scene against a plain white background with me in a maroon sweater and a woman in a dark blazer.

The camera did the whip pans and the dialogue followed the right logical order, accusatory lines to one person, defensive lines to the other but the audio sync drifted slightly.

My take: Both models got the camera move right. If you could combine Kling's dialogue sequencing with Seedance's kitchen, you'd have the perfect clip.

Winner: Tie

Test 2: Epic scale zoom

Prompt: Wide battlefield shot that zooms from aerial down into the macro detail of a single arrow mid-flight.

Seedance started from the air exactly as asked.

The camera drops fast, a heavy arrow enters the centre of frame, and the shot holds on it in macro while the army blurs beneath.

It did what the prompt said.

Kling ignored the aerial instruction completely and put the camera at ground level racing through a muddy trench battlefield with wooden stakes and black smoke.

An arrow enters the frame and the camera tracks it in tight, giving you a sharp macro of the wood grain, the fletching, the metal tip mid-flight.

Impressive close-up work but wrong camera move.

My take: Kling generated the wrong shot beautifully, which is still getting the wrong shot.

Winner: Seedance

Test 3: Match strike

Prompt: Close-up of a match striking against a box, flaring to life, lighting a candle. Wood, phosphorus, wax, and fire all interacting.

Seedance stayed at a standard close-up and kept it as one continuous shot.

The full sequence is in order, no cuts.

Kling went macro from the start, the unlit matchhead resting on the striker strip in extreme close-up.

Impressive texture but then it hard cuts. You hear the strike sound but the match is already lit.

The physics of lighting a candle are wrong, and the hard cuts break the sequence into unrelated stills.

My take: Kling's individual frames are more impressive but Seedance's version of events is believable as a sequence.

A usable clip has to show the thing happening, not a series of well-lit moments that don't connect.

Winner: Seedance

Test 4: Precise instruction following

Prompt: A royal greenhouse. Blue watering can on the left, red tulip on the right, reflecting pond in the centre. A hummingbird enters at second 2, hovers for three seconds, exits at second 5. Can and pot remain fixed throughout.

Seedance got it right, almost.

Blue watering can on the left, terracotta pot with red tulips on the right, reflecting pond in the middle. Ambient audio, stationary camera, lush setting.

The hummingbird arrives at roughly the 1-second mark, slightly early, lands on the water whose physics is pretty sus, and then darts out just before second 5.

Kling struggled.

The watering can came out silver, not blue. The scene is spare rather than royal.

My take: This one is close. Kling followed a dense set of specific instructions accurately but I like Seedance’s output more.

Winner: Tie

Where each one wins

Seedance is better at reading a camera direction and following it.

It understood the aerial-to-macro move, kept the match sequence coherent, and treated the prompt as a shot list and executed it.

However, its individual frame quality is often softer than Kling, and it can hallucinate specific details when instructions get precise, like the silver watering can.

Kling produces sharper, more textured frames and handles precise object placement better.

But it sometimes substitutes its own camera interpretation entirely, ignoring directions it either can't or won't follow.

February's call still holds, with one correction.

Seedance is the better director's tool.

It thinks about the sequence, respects the camera move, and keeps the physics coherent across a clip.

For narrative work, product commercials, anything where you're describing a specific shot rather than a mood, it's ahead.

But Kling surprised me in this round.

The greenhouse test in particular as it’s a technically demanding prompt with exact colour, position, and timing requirements, and Kling hit almost all of them.

For high-volume social content where you're generating a lot of clips and visual sharpness matters more, Kling at $6.99 a month is a hard offer to argue against.

If you're using either of these, reply and tell me what you're making with them.

Until next time,
Vaibhav 🤝

If you read till here, you might find this interesting

#AD 1

Most people with $1,000,000 or more saved have a number. Fewer have a plan for turning it into reliable income. Fisher Investments' Definitive Guide to Retirement Income helps you calculate future costs and build a portfolio strategy around them.

Download your free guide.

#AD 2

Some people use Claude to write emails. Others use it to basically run their entire business while they play Wordle.

This isn't just ChatGPT's cooler cousin. It's the AI that's quietly revolutionizing how smart people work – writing entire business plans, planning marketing campaigns, and basically becoming the intern you never have to pay.

The Hustle's new guide shows you exactly how the AI-literate are leaving everyone else behind. Subscribe for instant access.

Get our Claude playbook