Roope Rainisto

THE JET My latest short movie - longest I've ever made, 12 minutes. Made with @runwayml Runway MCP, BeatBandit, SeeDance (and lots of other things). 🧵notes

@maxescu The roadmap for 2028 development is reviewed by the committee set for 2027 to evaluate the market changes of 2026, and it could threaten the team bonuses by setting "risky milestones" - best to keep targets low so that expectations can be exceeded

It still baffles me why Adobe can't do this kind of stuff...

@dfsaracino @runwayml If you keep watching it carefully, you may encounter some other moments that will make you think: "wait, is this real?!"

@rainisto @runwayml Love how she walks up holding nothing and then suddenly has the box. Top notch work

THE JET My latest short movie - longest I've ever made, 12 minutes. Made with @runwayml Runway MCP, BeatBandit, SeeDance (and lots of other things). 🧵notes

@SEAWPSG @runwayml Fortunately this IS my day job! ;)

@rainisto @runwayml Don’t mean to be rude, but don’t quit your day job yet.

@BlackEyedHatMan @runwayml Yes that would be been possible. Then of course if I would be... there's "a few" technically inconsistent details here in this story. :)

@rainisto @runwayml This is extraordinary 👍 Of course someone from the "peanut gallery" must say something 😉 Perhaps a "post process" 💡 e.g. at 1:11 "Fastened" is supposed to be the correct word - could you reproduce the voice (say eleven labsl) and have AI "splice in" the corrected word?

well, i connected GPT-5.6 Sol to Blender and... let's just say the world of 3D graphics (at least via Blender's MCP) is safe for now... but still... this whole scene was generated by GPT-5.6 with very little human intervention feels like *EARLY* pixar, like too early lol

AI video - today (2026)

AI video - 3 years ago.

To achieve full spatial consistency - everyone sitting exactly where they were last time - the cabin looking exactly the same - you would need to create and add even more references. But I don't think it would made the story 'better'. It's the case of the performance now. A solid performance from an actor is more important than mechanical continuity. Plans are also challenging in the late stage of the creation process - once you start editing the movie, the videos guide the work. I don't look at the script anymore. Plans become out of date. The edit tells me what to do next. Storyboards were useful, but they're just artefacts. The script is useful, but it's also just an artefact. Nobody reads the script or sees the storyboards. They see the finished result!

This is the first short I've made mostly by talking to my LLM - which then talks with BeatBandit MCP to define the shots, reference images, storyboard, and then talks with Runway MCP to create videos, review them. The actual edit still needs to be done by hand, and this now takes most of the time. Maybe one day AI can also give a reasonable first edit, but that problem is far from trivial.

@IamEmily2050 @Dani__oros Meta Spark looks actually pretty good.

@Dani__oros Who? No one even at Seedance V2 level yet.

ByteDance Jimeng finally opened the Annual subscription for the new tier plan, and the price is shocking. This is really not looking good, $650 US dollars per month 😭😭😭😭

behold: the anti AI CTO vs John Gaeta (Matrix bullet time inventor) AI Technologist

Definitely - AI coverage has lots of valid points and lots of silly points. I'm blind to a lot of things in my own script - having this alien reader point out things that are obvious in my head but not to it is super helpful. It's then up to me / us to decide what to do with these comments.

@rainisto The best AI results come from refinement not one shot prompting alone

Playing with Sol to improve a screenplay. (One I wrote myself, a few years ago.) One agent improves the work, sends it to a second one that reads it cold, always same evaluation criteria + detailed notes. Agent one looks at the comments - I discuss its comments through each round, say which directions to go for each point raised. It's always proposing something silly, but mostly the suggestions are good. I give the direction. Sol applies them, second round. Rinse and repeat. Don't shoot the messenger, but it works really well. From 79 to 90 score in a few hours. I've tried this process before, but now with Sol (on high) it works much better than before. Yeah don't put yourself out from the loop, review and guide the suggestions each round. (And yes can you do this also on a multi-agent council loop - but Fable and Sol are quite much aligned nowadays.) Anyways - probably nothing!

Yes the scores aren't exact - there's about a 2-3 point difference if you give the same screenplay with same prompt to two independent agents, but the actual written comments themselves are usually very much aligned.

@ggg78g89 Yeah subs for Opus 4.8 only - it's not a good proposition now with worthwhile competition available.

🚨SCOOP: MY Friend at Anthropic says things are VERY tense internally. Dario's running tough meetings — GPT-5.6 Sol is strong and Grok 4.5 is right on Opus's heels. Pulling Fable from subs on July 12 would trigger mass cancellations (why keep Max for Opus 4.8?), so they're now pushing to keep Fable 5 in subs permanently.

@fellazb Yes, the use of colors here is really helpful - not just have a gray blocking scheme.

@rainisto Such an effective outcome, thanks for sharing this workflow. It also shows that it is important to have some control on basics such as composition and lighting for a basic mood in 3D as a starting point.

Asking Sol to use MCP Blender to create a scene of a car driving down the strip in Miami at night Positioning the camera, taking an image, then asking gpt-image-2 make this primitive 3d scene into a polished "Miami nightlife" scene

more examples