Col

TV producer by day, AI enthusiast/tinkerer by night. Strengths: Scottish. CPP: Runway, Luma, Kling, Freepik, Hailou, Higgsfield, Nim, PixVerse, Pika

[Col @LiveBetween2B]

Minimax, how I’ve missed you! The new H3 model’s t2v game - and prompt adherence - is off the charts. It’s made me proper lol more than once. Bravo! @Hailuo_AI #MinimaxH3

[Col @LiveBetween2B]

@FishAudio All the DRAMA please!!

[Jeff_synthesized @JeffSynthesized]

What an incredible experience this was. One of the most inspiring days I have ever experienced. I’m going to post about it tomorrow. @PeterDiamandis you’re an incredible human being, thank you. ❤️

[Col @LiveBetween2B]

Like the sage Notorious BIG once said…

Mo Pixels, Mo Problems

[Col @LiveBetween2B]

@araminta_k The tell is how easy it is to get in... I remember 3 or so years ago, when Runway opened up their CPP (and they may have been the first, I'm not sure) getting in was like trying to find one of Wonka's golden tickets. Now it's just tokens-for-promotions.

[Col @LiveBetween2B]

@MOVIETVWRITERX Case in point: every microdrama I've seen this year 😂

[Col @LiveBetween2B]

Better than all the microdramas I’ve seen this year combined 😂🧟‍♀️🩸☠️

[Dave Clark @Diesol]

You're on your first date.

You took a driverless car.

Oh and your date turns into a monster...

This is 30 Second Horrors Part 3

Created in Seedance 2.5.

[Col @LiveBetween2B]

Surprised this isn't getting more reposts. I'm sure there will be disagreement ... but I get it. It makes sense.

[Jon Finger @mrjonfinger]

White walls, green screens, grey walls, and other flat color “voids” are worse for tracking your Ai Hybrid production than more natural environments.

[Col @LiveBetween2B]

@andymac3d FFS already sold out!

[Col @LiveBetween2B]

@weberwongwong It’s a grind - in a good way. Try this, tweak the skills, try again. Like training a dog. But crucially - this only works if you’re using your own agents + MCP. The closed-off agents on creative sites just can’t be trained in the same way.

[Col @LiveBetween2B]

@andymac3d Ha - I vibe coded one a month ago. So easy to do now. However … mine does have one feature that I haven’t seen in ANY of these alts.

But I can’t tell you.

Cos everyone would vibe-copy it 😂

[Col @LiveBetween2B]

@kattlatte That is pretty spot on!

[Col @LiveBetween2B]

When the creative AI history books are written, there will be a whole chapter for the @fal team.

If you've been in this game for the last 3 years, you already know who they are. But ... the work they've done in the last 6 months (special mention to @OdinLovis) has been astounding.

[Gorkem Yurtseven @gorkem]

Try an early preview of our post trained version of H3.

It's incredibly fast.

Co-optimized with our custom inference stack for higher throughput with no compromises on output quality.

[Col @LiveBetween2B]

@andymac3d @Hailuo_AI PhD level intelligence ... and you use it for googly eyes!! 😂

[Col @LiveBetween2B]

@aizd117 I don't think anything at this stage is pixel accurate - but it's a little more pixel accurate than other models.

[Col @LiveBetween2B]

AI as a relighting tool is still improving ... but getting better every day. I crept down some dark suburban alleyways to see if Minimax H3 on MiniMax Design could do a complete day-to-night transformation on my footage.

Verdict: we're not a million miles away from being able to shoot horror movies in the daytime!

@Hailuo_AI #MiniMaxDesign

[Col @LiveBetween2B]

@guicastellanos1 @Hailuo_AI Agree. Also - and it's much more subtle - the wet tank in the last clip is a highlight for me.

[Col @LiveBetween2B]

Down another rabbit-hole with @Hailuo_AI Minimax Design, this time attempting to create a ‘wet weather’ skill for existing video.

Got close. Actual rain creates too much conflict as its primary instruction is to keep the structure of the video the same. But, wet hair/clothes, wet environment, overcast sky, diffused lighting … all things that are a preproduction concern could soon be a postproduction decision.

Again, courtesy of H3. A decent, repeatable result. But a skill is not a LoRA - that’s the logical next step! #minimaxdesign

[Col @LiveBetween2B]

@Walala_ai @Hailuo_AI Yes, it's not perfect - but it has retained the elements of the original video better than other models. That's the thing we can build upon.

[Col @LiveBetween2B]

@AlinTiganus @Hailuo_AI Yeah, but every step degrades the output. The goal is to try and do it in as few steps as possible - ideally just one.

[Col @LiveBetween2B]

@kickingkeys This is what AI video gen *used* to be like ... and it's beautiful!

[Col @LiveBetween2B]

@BittuVishn9048 Just an omni, but they let you build a Skill that - in theory at least - is repeatable across any clip.

[Col @LiveBetween2B]

@cfryant @Hailuo_AI Just text. Adding an image seems to change the video in some way.