How to Make AI Videos With Claude Fable 5 (Higgsfield Supercomputer) — Transcript
Full transcript
- 0:00Claude Fable 5 just completely changed
- 0:02how AI videos get made, and every video
- 0:04you're watching right now was made by
- 0:05it. Four ads and a week of vlogs, and I
- 0:07didn't edit a single frame. I just
- 0:09approved steps in one chat. I've been
- 0:11testing this setup for the past week,
- 0:12and nothing else I've tried even comes
- 0:14close. And so, in this video, I'll show
- 0:16you exactly how to make AI videos with
- 0:18Fable 5 from one brief to a finished
- 0:20batch, so you can make your first one
- 0:21today. The idea is you stop making the
- 0:23videos yourself and start directing an
- 0:25agent that makes them for you. With a
- 0:27normal video model, you write a prompt
- 0:28and take what comes back. But, here you
- 0:30hand over the whole job and approve the
- 0:31plan, the cast, and each render as it
- 0:33builds. Higgsfield supercomputer runs
- 0:35all of Higgsfield from a single chat, so
- 0:37you describe what you want once, and it
- 0:39plans the steps out, matching each one
- 0:40to the right model and rendering the
- 0:42lot. And the reason it can take on a
- 0:43brief this big is that it's running on
- 0:45Claude Fable 5. The way it works is that
- 0:47Fable 5 is the planner, and each step
- 0:50gets handed to whatever specialist fits.
- 0:52An image model for the frames, a video
- 0:54one for the renders, and a separate
- 0:56model for the voices. So, Fable 5's real
- 0:58job is lining them all up and passing
- 1:00the right thing to each one. Normally,
- 1:02four ads like these would mean jumping
- 1:03between a few different tools and an
- 1:05editor. A week of vlogs on top of that
- 1:07is days of work. Here, all of it happens
- 1:09in one chat. The only piece I make
- 1:11myself is the product, so that's where
- 1:12we start. Every ad is going to reuse
- 1:14that one product shot, so it's worth
- 1:16getting right before anything else.
- 1:17Everything today runs on Higgsfield, and
- 1:19the link is in the description if you
- 1:21want to follow along step-by-step. So,
- 1:22before I even open supercomputer, I'll
- 1:24go to the image tab, pick GPT image 2,
- 1:27set it to a one-to-one square, and
- 1:28describe the product I want. For this
- 1:30video, the product is a simple mug, a
- 1:32matte cream ceramic one with a little
- 1:34line art sun on the front. And I've got
- 1:35a plain product shot of the mug on a
- 1:37white background, which every ad is
- 1:39built around from here on. You make it
- 1:40once and reuse it, because if you let
- 1:42the agent regenerate the mug for every
- 1:44ad, the glaze and the shape come out a
- 1:45bit different each time, and your four
- 1:47ads stop looking like the same product.
- 1:49The lighting on this shot matters more
- 1:51than it looks. Keep it flat and with no
- 1:52harsh shadows or reflections, because
- 1:54the more the lighting is locked into the
- 1:56product shot, the harder it is for the
- 1:57agent to relight it convincingly in a
- 1:59sunny kitchen or a dim reading corner
- 2:01later. I use GPT image two for it
- 2:03because it holds fine product detail
- 2:05well. The shape and the little sun come
- 2:06out exactly right, which counts for a
- 2:08lot when this one image is about to
- 2:09appear in every ad. I keep it one-to-one
- 2:11on a plain background, too, since a
- 2:13simple product plate is the easiest
- 2:14thing for the model to drop into any
- 2:16scene later. Now, I'll open
- 2:17Supercomputer, and there are only two
- 2:19settings to change here. I'll pick
- 2:20approve before generation, so after
- 2:22every prompt it lays out its plan and
- 2:23pauses until I hit run instead of
- 2:25spinning on its own. I'll set it to
- 2:27Claude Fable 5, since that's the part
- 2:29that reads the brief and plans every
- 2:30step. And then I hand it a single brief
- 2:32that has to come back as four ads that
- 2:33all feel different. I'll give it that
- 2:35brief in one message. The mug goes in
- 2:36first, so it's locked as the product
- 2:38from line one. Then I tell it what I
- 2:40want. UGC style ads, four of them, each
- 2:42a different angle, morning coffee, cozy
- 2:44reading, office desk, and a gift idea.
- 2:46All casual and honest, like real
- 2:48creators shot them. I want them
- 2:50vertical, 9:16, so they're ready for
- 2:52reels and shorts. And I ask it to walk
- 2:54me through each step, so I can sign off
- 2:56as it goes. This is why I made the image
- 2:58first, instead of just describing the
- 2:59mug in the brief. If you only describe
- 3:01it, the agent pictures it a bit
- 3:02differently every time. An attached
- 3:04photo gives it one fixed thing to
- 3:06[music] copy, so the mug stays identical
- 3:07across all four ads. From that one
- 3:09message, it comes back with a full plan.
- 3:11Four ads and four different avatars,
- 3:13each one worked out down to the hair and
- 3:15the styling. Everyone gets its own
- 3:17setting and its own script. The casting
- 3:18is properly varied. The coffee ad gets a
- 3:20late 20s woman with ash blonde bangs in
- 3:22a bright kitchen, and the reading one a
- 3:24softer brunette on a couch with a throw
- 3:26and books. For the office ad, it picks a
- 3:2830-something guy with light stubble at a
- 3:29lived-in desk. And the gift one a
- 3:31bearded guy boxing it up under string
- 3:33lights. The scripts are just as
- 3:34different. [music] One leans on the mug
- 3:35making the morning feel calmer. Another
- 3:37plays it as a break from a desk full of
- 3:39monitors. And the gift version sells it
- 3:41as the present that always works. So, no
- 3:42two of them sound the same. It all comes
- 3:44down to the brief. I tell it to start,
- 3:46and it builds the first avatar, a late
- 3:4820s woman in a bright kitchen. Once Once
- 3:50it she's good, it puts that frame
- 3:51together, her holding the mug with the
- 3:53sun clearly on it, and starts rendering
- 3:55that first video while it moves on to
- 3:56the next creator. That start frame step
- 3:58is what keeps each ad consistent. It
- 4:00locks the creator's face and the mug
- 4:02into one still image first. So, when the
- 4:04video renders, it has a fixed target to
- 4:06animate instead of reinventing her frame
- 4:07by frame. [music] I make it check in at
- 4:09each step, mostly to save money. Catch a
- 4:11bad avatar right here and it's a
- 4:12one-line fix. Miss it and let it run and
- 4:15you've paid to render four ads you'll
- 4:16just throw out. The second creator is a
- 4:18woman with soft brown waves and a
- 4:19reading mug. And once I approve her, it
- 4:21goes straight into her start frame and
- 4:22renders the video off that. Then I tell
- 4:24it to build the last two at once instead
- 4:26of stopping to check with me on each and
- 4:28it works the same way through both. The
- 4:29avatar first, then the start frame, then
- 4:31the video straight off that frame. So,
- 4:33now there are four creators who all look
- 4:34different, two women and two men with a
- 4:36finished ad each. Then I change
- 4:37direction. I tell it I want them looking
- 4:39at the camera and talking. The video's
- 4:41at 10 seconds and the scripts tighten to
- 4:43fit. [music] And it regenerates all four
- 4:45as 10-second on-camera videos with
- 4:46proper lip sync, rewriting each script
- 4:48shorter to fit, and it never once starts
- 4:50over. I push it to 10 seconds and
- 4:52on-camera on purpose. 10 seconds is long
- 4:54enough to make one point and short
- 4:55enough that nobody scrolls off before
- 4:57it's done. And having them talk to
- 4:58camera is what makes it read as a real
- 5:00person.
- 5:01>> This is the mug I reach for first. It's
- 5:03just a simple ceramic thing with a sun
- 5:05on it,
- 5:07but pouring my coffee into it makes the
- 5:09whole morning feel calmer.
- 5:11This mug is my perfect reading
- 5:13companion.
- 5:15Tea in one hand, book in the other.
- 5:17The little sun on the side just fits the
- 5:20vibe.
- 5:21>> My desk used to be all screens. This mug
- 5:24didn't change much,
- 5:26but having it next to my keyboard makes
- 5:28the work day feel less robotic.
- 5:31Whenever I don't know what to get
- 5:32someone, I get this mug.
- 5:35Simple, beautiful.
- 5:37Everyone deserves a cup that makes them
- 5:39smile.
- 5:41>> Watching the four back, it's clearly the
- 5:42same mug in everyone, just re-lit for
- 5:44each scene. Ads are one thing, but the
- 5:46same agent will build you a whole
- 5:48content series. Now, let's do something
- 5:49completely different, a week of casual
- 5:51vlogs. This time I want a brand new
- 5:53creator instead of reusing any of the ad
- 5:55avatars. So, I have it build her
- 5:57character sheet first, full body and
- 5:58face, and then each day's video comes
- 6:00straight off that sheet, so there's no
- 6:02start frame to make this time. It's
- 6:03seven vlogs, one for each day of her
- 6:05week, 10 seconds each and all on camera.
- 6:07Then I'll paste the brief in. It plans
- 6:09her out, too, a warm mid-20s woman with
- 6:11dark hair and light freckles. Then it
- 6:13maps out 7 days for her, a morning
- 6:15routine, a gym session, errands around
- 6:17town, cooking dinner, then a night out
- 6:18with friends, a day outdoors, and a slow
- 6:20reset day at home. I'll approve, and it
- 6:22builds the character sheet, then
- 6:24generates all seven vlogs off that one
- 6:26sheet, so it's the same face all week.
- 6:27Get the character sheet right, and
- 6:29you've basically got a creator you own.
- 6:30Once she exists, you can put her in any
- 6:32video from here on. So, it's worth
- 6:34nudging her face and build until they're
- 6:35right before you generate a single day.
- 6:37It's also why the vlogs come together
- 6:39quicker than the ads did. The ads needed
- 6:41a fresh start frame for every video,
- 6:43since each one was a different creator
- 6:44holding a product. Here it's one person
- 6:46all week. So, the character sheet on its
- 6:47own is enough for the model to hold her
- 6:49face the whole way through. The sheet
- 6:51keeps her face locked, but the clothes
- 6:52and the settings can still change
- 6:54day-to-day. If you want a tighter look,
- 6:55put that in the brief, too. This vlog
- 6:57style is a different job to the ads.
- 6:59It's how you'd run a personal brand or a
- 7:01faceless channel. I do want to fix the
- 7:02voice, though, so I ask it to put a
- 7:04single consistent voice across all seven
- 7:06days. It offers me a few presets that
- 7:08suit a young creator. I pick one called
- 7:10Quinn, and it applies that same voice to
- 7:12every vlog. Preset voices like that are
- 7:14quick, and they fit, but they can sound
- 7:15a little generic. So, for a real
- 7:17channel, you'd probably clone your own
- 7:18voice instead. This is just the fast way
- 7:20to keep the whole week sounding like one
- 7:22person.
- 7:25>> Morning coffee.
- 7:28Let's see if I can be a functioning
- 7:29human today. Same walk every day, and I
- 7:31still love it.
- 7:33Three stops, and I've already lost the
- 7:35list. Post office, groceries, and one
- 7:39thing I definitely forgot.
- 7:42Wish me luck.
- 7:44Only got like 40 minutes today, so we're
- 7:46keeping it short.
- 7:48Legs,
- 7:49then I'm out.
- 7:50Honestly, the hardest part is just
- 7:52showing up.
- 7:58Attempting to cook something that isn't
- 8:01pasta. Wish me luck. Said I'd be ready
- 8:04at 8:00. It's 8:15.
- 8:06Meeting the girls in 20 minutes.
- 8:10This is 5:00 now.
- 8:15Needed some sun.
- 8:17Found a spot and honestly, might just
- 8:20stay here forever. Doing absolutely
- 8:23nothing today and I've never felt
- 8:25better.
- 8:26Laundry's going, tea's made. That's the
- 8:29whole plan.
- 8:30Reset day.
- 8:32>> That's a full week of content built from
- 8:33one character sheet and one voice, with
- 8:35me doing nothing but describing it. Once
- 8:37you've got a recipe that works, you
- 8:38don't have to baby sit it forever. You
- 8:40can take it off confirm and let it run a
- 8:42bigger batch unattended, which is how a
- 8:44week of vlogs turns into a month without
- 8:46you approving every frame. Because it
- 8:47remembers between sessions, the mug, the
- 8:49creators and the voice all carry over to
- 8:51your next chat. So you're not rebuilding
- 8:53your product or your cast every time you
- 8:55come back. For anyone making content at
- 8:57any volume, this is the change that
- 8:59matters. One person can now brief and
- 9:00approve what used to take a small team
- 9:02and the hard part stops being the
- 9:03production and becomes coming up with
- 9:05the ideas. The whole batch, all 11
- 9:07videos, cost a fraction of what a single
- 9:09shoot or an agency day would, because
- 9:11you're paying in credits per render
- 9:13rather than day rates and revisions. It
- 9:15suits anyone who needs volume and
- 9:16consistency more than one cinematic hero
- 9:18video, a brand running UGC or a creator
- 9:20posting daily. And if you only ever
- 9:22wanted one perfect 30-second film, you'd
- 9:24still art direct that yourself. But for
- 9:26a steady stream of content, this is the
- 9:27fastest route I've found. So you hand
- 9:29Claude Fable 5 and the supercomputer
- 9:31your own product or your own week and it
- 9:33builds the batch while you approve each
- 9:35step. And the part I can't do for you is
- 9:37pointed at your own product. [music]
- 9:38What it builds from a brief of your own
- 9:39is the part I want you to see. Open
- 9:41Higgsfield through the first link in the
- 9:42description and hand it your first brief
- 9:44today. Thank you for watching and I'll
- 9:46see you in the next one.
About this transcript
This page contains the full transcript of How to Make AI Videos With Claude Fable 5 (Higgsfield Supercomputer) by Parker Prompts, generated from the public captions YouTube serves with the video. The transcript has 2,250 words across 321 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.