From the creator community
GPT-6 Astra’s Five-Minute Launch Video: The Shared Workflow
Yatharth Gupta’s thirty-second Seam Studio launch film attracted requests for the workflow on X. The opening post credited GPT-6 Astra and fal’s H3 Max and described a five-minute result. In the replies, Gupta linked the visual-thinking skill he used. That link is more useful to a creator than the speed claim.
The published example uses seven keyframes and six five-second motion clips, followed by a separate edit. This article follows the state-by-state idea and develops an original example for a fictional product called Fieldnote. You can prepare its visual assets in Ciyo while keeping the original demo’s tools and our adaptation clearly separate.
Watch the demo and open the actual breakdown
The creator’s repository identifies Nano Banana 2 for frames, H3 Max for motion, Music 3 for instrumental audio and FFmpeg for assembly. Seam Studio itself is a concept product about a designer developing a garment and preparing a small preorder campaign. Its generated interface is concept imagery, not proof of working software.
Replies asked for more detail and questioned how much creative and editing work the short timing included. Treat the timing as a creator report. For your own project, estimate research, approved frames, retries and editing separately. The repeatable part is making visual states that can support a coherent transition.
Original community post and creator media. Our Ciyo adaptation follows the case study.
Choose the transformation your product actually provides
Fieldnote is our fictional workshop-planning app. Its film should show scattered research becoming a useful workshop board. Start with three states: loose notes, a person sorting them, and an organized board. The viewer should understand the change with the sound off.
For a real app, capture those states from the actual interface. Remove private content and use the captures as the source of truth. Decorative AI artwork can surround a screenshot, but a generated screen containing invented controls is not evidence that your product has those features.
Plan six beats before making six clips
The table gives the editor a job for every beat. It also prevents a common launch-film problem: six attractive shots that all communicate “our product exists.” The progression needs a problem, a change and a result.
Create the most difficult pair first, usually the action and result states. In Ciyo, make one approved presentation frame for each, then compare their framing and visual identity. Extend to the opening and ending only after this pair reads clearly.
| Beat | Visible state | Asset needed |
|---|---|---|
| 1 · Arrival | Loose workshop notes cover the desk | Opening still with a clear focal point |
| 2 · Friction | Two notes repeat the same idea | Closer detail that explains the problem |
| 3 · Action | Notes gather into a small set of groups | Start and result references for a transition |
| 4 · Result | The workshop board has a readable order | Approved actual screen for a real product |
| 5 · Use | A facilitator can see what happens next | One supporting scene or real capture |
| 6 · Invitation | Product name and one next step | Quiet ending background plus editable text |
Separate the frame brief from the motion brief
A still prompt specifies the objects, viewpoint, palette and empty space. A motion prompt specifies what changes over time. Repeating a long product description in the motion instruction can distract from the single movement you need. Keep complex interface text in the original capture or in the edit.
For Fieldnote, an appropriate first motion test is a restrained push toward the organized board. It does not require the video model to invent a working drag-and-drop interaction. Use Ciyo’s Image to Video on the approved frame and inspect whether labels, proportions or objects drift.
Plan a short launch film for Fieldnote, a fictional workshop-planning app. The story moves from scattered notes to an organized workshop board. Return six beats, each with a visible starting state, an ending state, a still-image brief and one restrained motion instruction. Identify where a real product screenshot would be required. Keep interface copy separate from generated artwork. Explain what each beat adds to the story and flag any redundant beat. Do not estimate a five-minute completion time.Budget the iteration you can actually see
Keep an asset ledger while working. Count accepted frames and usable clips, not just total generations. If one transition repeatedly changes the interface, use a cut between accurate stills or screen recordings in the final edit. More generations are not automatically a better answer.
Look at Ciyo’s displayed quote before each generation. The creator’s provider stack and timing do not establish a price or completion time for a different model and workflow. Your useful estimate starts with your own accepted-frame rate and the number of shots still missing.
| Stage | Record | Decision it supports |
|---|---|---|
| Frames | Approved state, reference and revision reason | Whether motion is ready to start |
| Motion | Requested move and visible drift | Keep, simplify or replace with a cut |
| Edit | Final duration, text and licensed soundtrack | Whether the promise is understandable |
| Export | Caption, aspect ratio and destination | Whether the film fits its actual placement |
Finish with an invitation the film has earned
Watch the sequence without music and ask a colleague to describe the product’s change in one sentence. If they remember the aesthetic but cannot explain the result, shorten the decorative opening and give the result more time. This is an editorial review, not a claimed conversion test.
Export a short preview and a still from the same visual family. Pair the film with a page that shows the actual product workflow. For a concept like Fieldnote, label the interface and narrative as fictional. For a shipped product, link the viewer to the feature they just saw.
Questions about this workflow
Can I install the creator’s skill inside Ciyo?
The linked repository describes an agent skill and its external tool setup. This article adapts its visual-state planning method to Ciyo; it does not claim Ciyo runs that skill or offers H3 Max.
Should I expect a finished film in five minutes?
No fixed turnaround follows from one demonstration. Source preparation, frame approval, generation retries, audio and editing all affect your project. Use the creator’s timing as context rather than a production promise.
Build the keyframes for your launch story
Create the action and result frames first. Once the transformation reads clearly, extend the film around them.