12:26Isa does AI · Tutorial
How to Make Social Media Designs in Minutes With Claude + OpenArt
A 12-minute walkthrough of a five-post-type design system built on a single brand sheet and two AI tools.
June 4thThe argument in one line.
OpenArt's Director chat can turn a single face reference sheet and a one-paragraph pitch into a fully cast, scripted, captioned 90-second live-action drama, folding story generation, shot breakdown, and editing into one conversation instead of three separate tools.
Read if. Skip if.
READ IF YOU ARE…
- A creator who wants scripted, narrative short-form video without actors, a crew, or a location.
- Someone already juggling separate AI tools for story, prompts, and editing who wants to see a one-chat alternative.
- A vertical-content creator curious what a full 'micro drama' pipeline looks like, cast to captions to export.
SKIP IF…
- You want full manual control over every shot's camera angle and prompt wording, this workflow is built around the AI proposing and you reacting.
- You're not willing to pay per-second render credits, 1080p output runs about 230 credits/second, which adds up fast even on a 90-second episode.
The full version, fast.
OpenArt's Director tool builds a complete AI micro drama from one chat: attach a face reference sheet to lock your look, describe the story you want, and it proposes three openings, then expands your pick into a full cast, beat sheet, and five-step script built around a single twist. You set aspect ratio, resolution, and length (1080p runs about 230 credits/second), review AI-generated character and location stills, then generate the whole episode in one pass instead of stitching separate clips together. When a handful of shots come out wrong, you regenerate just those and the tool rejoins them into the existing cut. Auto-captions and export finish the job, though leftover audio and physics glitches still need a manual review pass.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:24
01 · Cold open, the promise
Isa states the pitch: a full 90-second AI micro drama, cast and twist included, made from a few prompts in one tool.
00:25 – 01:23
02 · Opening OpenArt's Director
Sets Guide-me mode so the tool checks in at every step, then attaches a face reference sheet before typing the story pitch.
01:23 – 01:58
03 · Three story openings, one pick
Director returns three story openings; Isa picks 'the overlooked heiress' and tells it to run with option one.
01:58 – 02:42
04 · Building the cast and beat sheet
The chat expands the pick into a full cast built around one twist: Isa as the overlooked assistant, Julian as the humiliating board member, plus a host and background guests.
02:42 – 03:41
05 · Five beats, one chat
Director lays out the whole 90 seconds as five named beats, then folds shot breakdown into the same chat instead of needing a separate tool to convert story into prompts.
03:41 – 04:34
06 · Aspect ratio, resolution, duration
Three production calls before anything renders: 9:16 vertical, a resolution tier (480p/720p/1080p, priced per credit-second), and a roughly 90-second runtime.
04:34 – 05:22
07 · Reviewing the generated cast
Isa reviews AI-generated stills of herself, Julian, the host, and the gala location before committing to a full render.
05:22 – 06:42
08 · First render and first watch-back
One confirmation renders the full multi-shot episode; Isa watches the first cut and starts flagging which shots don't hold up.
06:42 – 07:41
09 · Fixing broken clips without a full redo
Three of the six shots need regenerating; Director redoes just those and rejoins them into the existing cut.
07:41 – 09:20
10 · Auto captions, then export
One action auto-captions the whole timeline, then the episode exports at 1080p as a finished vertical video.
09:20 – 09:47
11 · Final watch-back, remaining flaws
Isa calls the episode finished but flags two leftover errors: a character stepping over nothing, and dialogue that carries too far across the room.
09:47 – 10:07
12 · The pitch
Isa closes with a limited-time OpenArt discount through her referral link and a call to build a first episode around one character and a twist.
Lines worth screenshotting.
- A single AI chat tool can now generate a complete 90-second scripted drama, cast included, from one paragraph of description and a face reference photo.
- Attaching a face reference sheet in the first message is what keeps a character's likeness consistent across every generated shot afterward.
- Resolution is the real cost lever in AI video generation: OpenArt bills roughly 60 credits/second at 480p, 110 at 720p, and 230 at 1080p.
- 'Vibe directing' means picking a direction from AI-generated options and reacting to what it builds, rather than writing exact prompts for every shot.
- A five-beat structure, hook, sting, turn, reversal, button, is enough to script a satisfying 90-second twist story.
- When three of six generated clips come out wrong, the fix is to regenerate just those clips and let the tool re-stitch them into the existing cut, not redo the whole render.
- Auto-generated captions matter most for vertical video, since most viewers watch with sound off.
- Even a live-action-styled AI drama still needs a human review pass: dialogue can carry too far across a room, and physical actions can miss what they're supposedly interacting with.
- Traditional AI video pipelines split story-writing, shot breakdown, and prompt-writing across separate tools; Director folds the story and its scene structure into the same chat.
Five beats and one face sheet make an AI twist story work
STORY STRUCTURE
A face reference photo locks a character's identity across every generated shot, and a five-beat hook-sting-turn-reversal-button structure is enough to carry a satisfying 90-second twist.
02Opening OpenArt's Director
- Guide me mode makes the tool check in at each step instead of generating the whole video in one shot, which matters when you're still learning a new AI tool's judgment.
- Attaching a face reference sheet in the very first message is what locks a character's look before any story content gets generated.
03Three story openings, one pick
- Treat the AI's first pass as a menu of directions, not a finished output, picking one and reacting to what it builds is a faster loop than trying to fully spec the story upfront.
04Building the cast and beat sheet
- A strong twist works backward: build supporting characters specifically to set up the one reveal, rather than fleshing out a full ensemble.
- A single antagonist whose reaction the audience gets to watch is what makes a twist land, more than a large cast.
05Five beats, one chat
- The hook-sting-turn-reversal-button structure compresses a satisfying twist story into five beats, short enough for a 90-second runtime.
- Doing story-writing and shot-breakdown in the same chat removes the step of manually converting a script into shot-by-shot prompts for a separate video generator.
06Aspect ratio, resolution, duration
- Resolution is the actual cost lever in AI video generation: 1080p runs roughly four times the credit cost of 480p, so match resolution to where the video will actually be watched.
- Vertical 9:16 is the default choice for anything built for short-form feeds, decided before any generation starts.
07Reviewing the generated cast
- Review every generated character and location still before committing to a full render, a face that looks 'slightly off' is worth catching at the still stage, not after rendering dialogue.
08First render and first watch-back
- A full multi-shot episode can render from one confirmation once the cast and beat sheet are locked, rather than generating and assembling each shot separately.
- Always watch the first full render back critically, this is where you catch which specific shots didn't hold up in motion.
09Fixing broken clips without a full redo
- When only some shots in a multi-clip render fail, regenerating just those clips and re-joining them into the existing cut avoids redoing the whole video.
- Being able to step back into any single shot at any stage, without restarting downstream work, is what makes iterative fixes cheap.
10Auto captions, then export
- Auto-captioning the full timeline in one action beats manually timing captions line by line, and matters most for vertical video watched with sound off.
- Export resolution is a separate decision from generation resolution, choose the final output quality at the export step.
11Final watch-back, remaining flaws
- Even a finished AI-generated scene can carry small physics or audio errors, a character stepping over nothing, dialogue volume not matching the room, worth a final critical pass before publishing.
Terms worth knowing.
- Director (OpenArt)
- OpenArt's chat-based video generation mode, branded 'Vibe Direct Your Next Video,' that builds a story, cast, and shot list from a conversation instead of individual prompts.
- Face reference sheet
- A set of reference photos of a person's face, attached at the start of a project, that an AI video tool uses to keep that person's likeness consistent across every generated shot.
- Vibe directing
- A workflow where you pick from AI-generated story or shot options and react to what it builds, rather than writing exact prompts for every individual shot.
- Credits (render cost)
- OpenArt's per-second billing unit for video generation, which scales with output resolution, roughly 60 credits/second at 480p up to 230 credits/second at 1080p.
Things they pointed at.
Lines you could clip.
00:04
“It's not just one short video, it's a full 90 second episode with a cast, a story, and a twist ending.”
01:18
“That face sheet is the important part because it locks my look from the very first message and keeps me the same person in every shot later.”
01:23
“This honestly feels more like vibe directing than prompting.”
03:52
“480p runs about 60 credits a second, 720p is around 110, and 1080p is about 230.”
06:50
“So now I've got three inconsistent videos inside an otherwise finished video, and normally that would mean redoing the whole thing.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphor
If you've ever seen these micro drama videos and wanted to make your own, now you can do that by just using AI. And it's not just one short video, it's a full 90 second episode with a cast, a story, and a twist ending. In this video, I'll walk you through the entire process, from creating the characters to making a full drama episode, by just using one tool.
And the best part is that the only thing that I'll need is just a few prompts. For all of this, I'm using Open Art Director, and every prompt I'll use will be in a file in the video's description, so you can copy them and follow along step by step. Normally, every video is only a few seconds, and you put them all together after, but the director makes minutes of finished video in one generation, with the characters, the voices, and the music already in it.
So this one's actually worth learning properly. When you open Director from the menu on the left, you land in a chat box. The first thing I'll do is set it to guide me, so it checks in with me at every step instead of making the whole video in one shot.
Then I'll hit the plus to attach my character's face reference sheet and type in a couple of lines about what I want, because keeping it that way is what gets it to suggest angles I wouldn't have thought of on my own. That face sheet is the important part because it locks my look from the very first message and keeps me the same person in every shot later.
It sets up the project, locks my face to my reference sheet, then gives me three different story openings to pick from.
and all of them make sense. But each one goes into detail, like describing the whole emotional baggage, which already makes it sound serious. So from one message, I've already got three real story directions to choose from.
But those are still just story ideas. The one I pick is where it actually starts turning into a real episode. So I'll go with the less obvious one here.
I'll just tell it to run with number one, the overlooked heiress, where everyone thinks I'm just staff. And the one thing I've learned from this is that you never treat that first pass the finished thing.
You pick the direction you like, let it build from there and react as it goes. That's the whole vibe directing thing again.
So it takes that one line and builds out the actual cast and it's smart about it because every character is there to set up that one twist in the end. I'm the assistant nobody looks at twice and the twist is that I'm secretly the new majority owner of the company. Then it gives me a villain, Julian, the board member who humiliates me in front of the rest of the guests Because a twist like this only works if there's one person whose reaction you get to watch.
There's the host up at the podium who makes the announcement, plus a few background guests so the gala actually feels full. And then it lays out the whole 90 seconds across five quick steps. I'm walking through the gala carrying a coffee tray.
Julian snaps his fingers and treats me like staff. That's when the host taps the mic to introduce the company's new majority owner and points right at me. It even breaks down the reveal, my walk to the stage, and the final line that ends the episode.
And to get to this exact point with any other setup, you'd already be using at least two separate tools. You'd have one of them write the whole story out. Then you'd take that and break it into separate scenes by hand, turning it into shot -by -shot prompts so the image and video generators can actually use it.
But here it's doing both in the same chat, the story and the structure. So there's nothing to copy out or reformat between tabs. So now I've got a full cast and a finished 90 second story to use.
But all of this is still just names and lines on a screen. And now I want to see if it can actually turn these people into something real. So before it shows me a single face, it stops and asks how I want the whole thing built.
It gives me three quick calls to make. The aspect ratio, the resolution, and the length.
Aspect ratio is the easy one, so I'll set it to 9 by 16, since this is all for vertical feeds. Resolution is the one that actually costs you, because 480p runs about 60 credits a second, 720p is around 110, and 1080p is about 230, so it's the most expensive one, but for a real live action look, it's the best one to go with.
and the length I'll just set to around 90 seconds. So I'll tell it the cast is perfect, keep Julian exactly as he is, and let it move on. Then it builds out the whole cast and the location, and I'll look at them one by one.
So first off, I go with my character. And on this close -up photo on the right, I feel like I look slightly different than usual. But it's nothing serious, so I'll go with it.
Then Julian, the board member who humiliates me, so he's the face you're watching when it all turns around. And he looks really realistic, especially with the wrinkles around his eyes, so I definitely like that one.
The host there at the podium, the one who makes the announcement. And again, it looks like a real portrait photo of her, so nobody would ever realize she isn't real. And the last one is the gala itself, the hall dressed for the reveal with a clear stage and a podium.
The whole place looks really realistic and official, so I'll definitely use it for the video later on. And what makes all of this work is that my face is built straight off that reference sheet, so it's locked from the start. And now I've got the whole cast and the gala built, with my face locked to the one I started with.
But still, I'll need to make sure whether all of this holds up the second it starts moving. So I'll give it the go ahead, and now I actually find out if it can pull the whole thing off. I'll just tell the model everything looks good.
And that single confirmation kicks off the next big render.
Over there. And keep it moving. Of course.
The help stays at the back. I was only passing through. Then pass through somewhere else.
Ladies and gentlemen. Thank you all for coming. It's my honor to introduce the company's new majority owner.
Wait, the assistant? That's not possible.
Miss, I didn't realize - Thank you, Julian. That'll be all. I wouldn't say it's perfect though.
Watching it back, a few moments clearly need another pass. So now I've got three inconsistent videos inside an otherwise finished video, and normally that would mean redoing the whole thing. But all I'll have to do here is tell it which videos didn't look good and regenerate the second, fourth, and fifth videos.
And this is the part I really like about the director. You can step back in and fix individual videos anytime, no matter what stage of the video you're at. It redoes the three broken ones with the same cast, the same parts, and the look, while leaving the first and the third videos completely un - touched then it joins all three back into the cut right where the old ones were so I fixed it without ever redoing the whole video the three are back in place but the way they cut into each other still isn't smooth and that's exactly what the edit is for the editing page has three tabs shot scene and timeline but I don't actually need to make any changes here so I can go straight to adding captions what I'll do here is I'll go to the timeline tab I'll open the caption section and hit auto captions so it lays captions a the entire video on its own, without me typing or timing a single line.
For something going on vertical feeds, you really want captions, since so many people watch with the sound off. And normally, picking shots and captioning would mean a whole separate editor, but here it's all on the same page as everything else. Then, once it's all cut together and captioned, I'll hit export, choose 1080p, and see how the final cuts look.
Over there, and keep it moving. Of course.
The help stays at the back. I was only passing through. Then pass through somewhere else.
Ladies and gentlemen. Thank you all for coming. It's my honor to introduce the company's new majority owner.
Wait, the assistant? Her? She's the assistant?
Miss I didn't realize Thank You Julian that'll be all And honestly, it's looking really good now. The cuts flow properly, the captions are in, and the whole episode feels finished.
And that's a finished 90 -second microdrama made without filming a single thing. But a few details still feel off, like the moment where I appear to step over something that is not there. And the conversation with Julian somehow carries across the entire room, even though neither of us are speaking that loudly.
But overall, I am really happy with the result. But the best part is the limited time deal. Upgrade to Advanced or above before July 31st and unlock up to 40 % off some of the best AI image and video models for the rest of 2026.
So sign up through my link to OpenArt and build your first episode around one character and an emotional twist in the end. Thanks for watching and I'll see you in the next one.
The Hook
The bait, then the rug-pull.
Isa opens with the promise every AI-video tutorial makes, one tool, one prompt, but backs it up by building a full cast, a five-beat twist script, and a finished 90-second gala scene inside a single chat with OpenArt's Director, then fixing the shots that came out wrong without re-rendering the whole thing.
Named ideas worth stealing.
02:43list
The Five-Beat Twist Structure (Hook / Sting / Turn / Reversal / Button)
- Hook
- The sting
- The turn
- The reversal
- The button
Director breaks a 90-second twist story into five named beats: the setup image, the insult that plants the misdirection, the announcement that flips it, the crowd's reaction, and a final quiet line before cutting to black.
Steal forany short-form scripted story built around a single reveal
03:46concept
OpenArt render cost by resolution
- 480p ~60 credits/sec
- 720p ~110 credits/sec
- 1080p ~230 credits/sec
Resolution is the primary lever on render cost inside Director; 1080p costs roughly four times what 480p costs per second of finished video.
Steal forbudgeting AI video render costs before committing to a resolution
How they asked for the click.
VERBAL ASK
09:47link
“Upgrade to Advanced or above before July 31st and unlock up to 40% off some of the best AI image and video models... sign up through my link to OpenArt.”
Spoken discount pitch paired with an affiliate link (tolt.link/microdrama) repeated in the description, framed as a limited-time offer rather than a hard sell.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
