All right, so this week I want to break down something I’ve been testing that I think is genuinely one of the more significant updates we’ve seen in AI video this year. And I also want to catch you up on a few other things that dropped in the last two weeks.

Let’s start with the main thing.

Kling dropped Motion Control for their 3.0 model on March 4th, and the concept is straightforward: you give it a still image of a character and a reference video showing a motion, and it transfers that motion onto your character.

Here’s the thing about Motion Control. It’s not a magic button. The quality of your output is directly tied to the quality of your reference video.

Two weeks after Kling’s Motion Control launch, Higgsfield dropped Cinema Studio 2.5 on March 18th. Worth paying attention to if you’ve been using Higgsfield as part of your stack.

One more thing worth mentioning: OpenAI quietly shut down the Sora standalone app this month.

Adobe Firefly expanded this month and added Kling directly into its model lineup. They’re now at 30+ models including Google, OpenAI, and Runway alongside Kling. They also launched Quick Cut, turns raw footage into a structured first cut automatically.

Midjourney dropped V8 Alpha on March 17th, and the headline number is speed: 4 to 5 times faster than V6. Faster iterations mean you can actually test more concepts in a session instead of waiting on generations.

Suno released 5.5 on March 26th, and this one adds three features worth paying attention to if you use it for background music or score work.

A lot moved this week. Motion Control is real and worth testing. Cinema Studio 2.5 is a meaningful update if Higgsfield is in your stack. Midjourney V8 is worth keeping an eye on even in alpha. Suno 5.5 is a bigger shift than the version number suggests. And the broader market is settling into a shape that makes tool selection less overwhelming, you don’t need everything, you need the right combination for what you’re building.

One more thing, I’ve been busy behind the scenes. I’m building out some new courses and guides that I think are going to be genuinely useful for where this space is heading. Nothing to announce just yet, but they’re coming soon. Stay tuned.

I’ll catch you in the next one.

- Khalil

Docs piling up faster than you can write them? Same.

Every team knows the feeling — product ships, docs don't. Changelogs get forgotten. Style violations quietly accumulate. Broken links go unnoticed for months.

Mintlify's new Workflows feature fixes this. Define automation rules, and the agent handles the recurring maintenance work for you — on your schedule, by your rules.

Draft docs when a PR merges. Generate changelogs every Friday. Run a style audit on every push. Flag translation lag before it becomes a problem. Each workflow is version controlled, fully configurable, and fits into your existing review process.

You decide when it runs, what it checks, and whether changes get committed directly or opened as a pull request for review.

The result: documentation that actually keeps up with your product, without someone manually chasing it down.

Try Workflows

Kling dropped Motion Control for their 3.0 model on March 4th, and the concept is straightforward: you give it a still image of a character and a reference video showing a motion, and it transfers that motion onto your character.

Not a rough approximation. We’re talking full-body, posture, joint movement, hand gestures, and facial expressions, all extracted from the reference video and applied to your image.

The reference clip can be anywhere from 3 to 30 seconds. So you could shoot yourself walking, dancing, or performing a scene, and then apply that exact motion to any AI-generated character.

When this update dropped, my main question wasn’t about dance videos or memes. It was: could it actually work for detailed dialogue scenes and real AI filmmaking? Because if you can reliably transfer subtle performances, not just body movement but nuanced facial expression, then it becomes a genuinely useful tool for building cinematic scenes.

A few things worth noting about how it actually works under the hood:

The system uses something Kling calls Element Binding, which keeps the character’s face consistent across every angle, emotion, and occlusion in the generated video. That’s been one of the bigger problems with motion-based generation: you start with a clear face and by the end of the clip it’s drifting. Element Binding is designed to solve that.

The motion transfer is powered by “3D Spacetime Joint Attention”, the model processes motion in three dimensions, not just matching 2D poses frame by frame. That means it understands physics. Gravity, balance, momentum, how fabric moves, how a body decelerates.

I tested both orientation modes, matching the video vs. matching the image, and the results were noticeably different. When the orientation matched the video, the motion looked more natural and facial movement stayed more consistent. When I switched to match the image, things started to fall apart. So if performance accuracy is your goal, that’s the setting you want.

I also stress tested head turns and face blocking. With 3.0 it actually recovered pretty well. The identity stayed relatively consistent even after the face was partially hidden. Certain mouth movements can still break the model though, licking your lips confused it quite a bit.

I also tested voice binding, since Kling 3.0 Omni lets you bind a voice to a character, I created a voice in ElevenLabs and bound it. But when I ran it through Motion Control, the audio didn’t automatically switch. It just kept the original audio from the reference clip. You still need to swap the voice manually afterward. If Kling ever adds automatic voice replacement when a voice is bound, that will be a big deal.

Read the full Kling 3.0 announcement

Try Kling 3.0 Motion Control for yourself

Here’s the thing about Motion Control. It’s not a magic button. The quality of your output is directly tied to the quality of your reference video.

Shaky reference = shaky transfer. Unclear motion = unclear transfer. If you’re serious about using this in real projects, you need clean angles, clear body visibility, and controlled movement in your reference footage.

I tested it inside an actual cinematic situation, a small scene from the Life of the Lazy Mon story I’m building. An airport customs line, first trip down to Costa Rica. Something became very clear watching it back: the model is only as good as the acting driving it. In my case, the acting was not great. Which means for someone like me, it might actually be better to prompt the performance than to film it myself.

But here’s where I think the real value is. Kling Omni already does a lot. When it lands dialogue and acting, it’s amazing. But sometimes it refuses to nail a specific word or phrase no matter how many regenerations you run. Motion Control gives you another option, you perform the line, upload the clip, map the character onto it, and now you have the exact beat you wanted.

The combination of human performance and AI generation is where I think the workflow is heading. Real actors provide the nuanced moments. AI handles the supplementary shots, filler scenes, in-between moments. That hybrid approach is where this tool starts to make real sense.

Two weeks after Kling’s Motion Control launch, Higgsfield dropped Cinema Studio 2.5 on March 18th. Worth paying attention to if you’ve been using Higgsfield as part of your stack.

The headline feature is the Soul Cast integration, now built directly into the generation workflow. Version 2.5 puts your AI characters at the center of the process before the first frame is generated. You can have up to 3 Soul Cast characters in a single scene.

The ‘waxy’ plastic look that’s been a consistent complaint with AI actors? They’ve made real progress on that. Skin textures are noticeably better, and the characters follow the visual language of whatever era or genre you set.

I covered Cinema Studio 2.5 in depth in my most recent video, go check that one out if you want the full breakdown and real tests.

One note: Higgsfield’s blog post mentions native color grading as a 2.5 feature, but I didn’t actually see that in the interface. Take that one with a grain of salt until it’s clearly there.

Read the Cinema Studio 2.5 announcement

One more thing worth mentioning: OpenAI quietly shut down the Sora standalone app this month.

It didn’t make as much noise as you might expect, and honestly, by this point, Sora had already been lapped by the competition. The market moved fast and Sora didn’t keep pace.

The market has split into pretty clear tiers now. Runway for quality-first. Kling for cost efficiency and control. Veo 3 if you’re in the Google ecosystem. And open-source options pushing hard from below. You don’t need all of them, you need the right combination for what you’re building.

What to know about the Sora discontinuation

Adobe Firefly expanded this month and added Kling directly into its model lineup. They’re now at 30+ models including Google, OpenAI, and Runway alongside Kling. They also launched Quick Cut, turns raw footage into a structured first cut automatically.

Worth keeping an eye on as the pipeline evolves. The question for me is always whether it fits into a real creative workflow, not just whether the feature list is impressive.

Read the Adobe Firefly expansion announcement

Midjourney dropped V8 Alpha on March 17th, and the headline number is speed: 4 to 5 times faster than V6. Faster iterations mean you can actually test more concepts in a session instead of waiting on generations.

Native 2K output via the --hd flag, and noticeably improved text rendering. Quoted text inside images is actually readable now, which has been a persistent weak spot. Still alpha, not on the main site yet, but if you’re in the Discord you can test it.

Keep an eye on this one, when alpha becomes the default model, the workflow implications are real.

Read the Midjourney V8 Alpha release notes

Suno released 5.5 on March 26th, and this one adds three features worth paying attention to if you use it for background music or score work.

Voices is the big one, voice cloning for singing, available for Pro and Pro+ subscribers. You can create a custom AI singing voice that sounds like you, and use it across tracks.

Custom Models lets you train on your own music style. Tempo, genre, tone, you can start locking that in as a model rather than re-prompting from scratch every time.

And My Taste is an auto-personalization layer. It learns your genre and mood preferences over time and starts shaping the default outputs toward what you actually use.

If you’re using AI music in your workflow at all, background scoring, intros, original tracks, 5.5 is worth revisiting.

Read the Suno v5.5 announcement