Harsh Verma

• Artificial Intelligence
• 130k+ Followers On Instagram

[Harsh Verma @harshxaii]

Seedance 2.5 just went live on Runway.

You now get more reference inputs and longer clip lengths to work with, giving you a lot more room to actually build something out.

Try it today.

Video via @runwayml

[Harsh Verma @harshxaii]

Layers just went live in Luma Agents.

You can ask the agent to pull apart any image into layers, whether it was generated in Luma or something you uploaded. From there you just keep chatting: swap the product, change the background, update the headline.

Change one part, everything else stays untouched. Powered by Uni-1.

Video via @LumaLabsAI

[Harsh Verma @harshxaii]

Same character sheet, two different Claude models, one video model rendering it.

Magnific tested Opus 5 against Opus 4.8 on the same character prompt, then ran both outputs through Seedance 2.0. What this actually reveals isn't which model performed better, it's how much the language model you choose shapes what the video model produces. The prompt is the real variable here, not just the renderer.

Character consistency is turning into a prompting problem, not a rendering one.

Video via @magnific

[Harsh Verma @harshxaii]

I turned one phone photo into a full Pixel ad using just a single prompt.

Made this on @FlowbyGoogle

No shoot, no studio, no editing timeline. Just one photo in, and the AI built the lighting, framing, and polish to make it look like an actual product campaign.

This is how fast ad creative is moving now.

@GoogleIndia @GooglePixel_US @GeminiApp

[Harsh Verma @harshxaii]

Kling keeps quietly pushing the ceiling on AI video higher.

A tiny meow-squito buzzing around causing chaos, rendered with the kind of motion and character detail that used to need a full VFX pipeline. What's impressive is how consistent the little creature stays throughout the clip while the movement still feels physically believable.

This is the new baseline for playful short-form AI video now.

Video via @Kling_ai

[Harsh Verma @harshxaii]

Ad creative testing is becoming as simple as writing a prompt.

This Claude Opus 5 vs GPT 5.6 comparison ad was made entirely with Arcads. You pick a concept, AI actors deliver the script, and you walk away with a finished ad ready to run, no shoot, no studio, no editing timeline slowing things down.

Your ad pipeline just got a lot faster.

Video via @arcads_ai

[Harsh Verma @harshxaii]

Kling just broke down what actually makes a shot feel cinematic.

It's not about writing a fancier prompt. It's about thinking like a director, building in camera movement, framing, and pacing as part of how you describe the shot. Once you stop treating the model like a random generator and start treating it like a cinematographer, the output changes completely.

Small shift in how you prompt, big difference in what comes out.

Video via @Kling_ai

[Harsh Verma @harshxaii]

Introducing Gulab Music.

You bring the intent, a feeling, a story, a vibe. Gulab takes it from there and puts together a complete music video: original track, synced visuals, final cut.

One sentence in, a finished music video out.

Video via @trygulab

[Harsh Verma @harshxaii]

Character design is turning into a single step inside Luma.

Before a character can do anything on screen, someone has to nail down exactly how they look from every angle. A Luma Skill handled that entire process in one shot, generating a full character sheet with every side rendered and staying consistent. That's the kind of unglamorous prep work that normally eats up days on a real production. Made by Misho.

Pre-production is quietly getting automated.

Video via @LumaLabsAI

[Harsh Verma @harshxaii]

The best AI video right now isn't about spectacle, it's about restraint.

Jason Chalkley built the quiet moment right before a dragon takes off, letting the whole scene hold its breath. No big action, just tension and timing, made entirely in Luma. The skill here is in what he left out, and that kind of patience is what most AI clips are missing.

This is the bar AI filmmaking should be aiming for.

Video via @LumaLabsAI

[Harsh Verma @harshxaii]

Kling keeps closing the gap between a prompt and an actual film set.

This disaster scene looks like it was shot on location, not generated. The wreckage physics, lighting, and camera movement all hold together in a way that normally needs a full crew and a real budget. You describe the scene, and the model builds the set for you.

The film set is turning into a text box.

Video via @Kling_ai

[Harsh Verma @harshxaii]

The best AI video doesn't come from controlling every prompt, it comes from working with the model like a collaborator.

Set up a Luma Skill, give it space to explore your idea, and shape the story from what it brings back. Instead of micromanaging each step, you're directing an agent that actually gets what you're going for.

This is what a creative agent looks like when it's actually working well.

Video via @LumaLabsAI

[Harsh Verma @harshxaii]

Runway just introduced Workflows inside Runway Agent.

You can now build, run, or edit node-based workflows using plain language, no manual node setup needed. This means you can get high-quality, scalable outputs without wrestling with the technical side.

Try it by calling the /Workflow skill at the link below.

Video via @runwayml

[Harsh Verma @harshxaii]

ChatGPT Voice just moved to desktop, and it's turning your voice into the control layer for agents.

You can now talk directly to your computer and direct multiple agents in ChatGPT Work or Codex at the same time. Running on GPT-Live, it listens, responds, and coordinates the work in the background while you keep working on other things. No more switching windows to check on every task.

Voice becoming the actual interface for agent work, that's the real shift here.

Video via @OpenAI

[Harsh Verma @harshxaii]

Runway just introduced Media Router.

Instead of manually picking which model to use for every single request, you set your preference once, whether that's cost, quality, or speed, and the router automatically picks the right video, image, or audio model for the job.

One setup, and it handles the model-picking for you every time after that.

Live now in Runway Dev.

Video via @runwayml

[Harsh Verma @harshxaii]

Seedance 2.0 is solving one of AI video's hardest problems, continuity.

One generation gives you multiple shots, and the characters actually keep the same look across every cut, all wrapped in one consistent style from start to finish. Full prompt is in the thread if you want to try it yourself.

Keeping the same character consistent across cuts, that's the real breakthrough here.

Video via @magnific

[Harsh Verma @harshxaii]

HeyGen just launched companion mode.

Instead of just generating a video from a prompt, your agent now pitches multiple angles first, shows you a storyboard to review, then sketches out frames before anything gets built. You approve each step along the way.

This is the shift from ordering a video to actually directing one.

Video via @HeyGen

[Harsh Verma @harshxaii]

AI music videos are starting to feel like actual films.

BARIDEGI blends Korean mythology with K-pop visuals, and the creators just won Kling's NEXTGEN challenge for best music video. What makes it work is the consistency, an entire story holding together across two full minutes, not just a string of nice-looking clips.

This is the shift happening in AI video. Not clips anymore, whole stories.

Video via @Kling_ai

[Harsh Verma @harshxaii]

Reframe from Luma just fixed the aspect ratio headache for AI video.

You shoot one master clip, and Reframe adjusts it to fit whatever format you need, vertical, square, or wide. It doesn't just crop the edges off, it repositions the framing so nothing important gets cut out.

One shoot, every platform covered.

Video via @LumaLabsAI

[Harsh Verma @harshxaii]

Magnific just moved into Adobe InDesign.

You can generate images right inside InDesign now. The image comes in as an editable vector, so you can resize it for print without it going blurry. You can also fix the lighting, remove the background, or make mockups, all without leaving the page.

So you don't need five different tools anymore. Just one place to do it all.

Video via @magnific