Magnific builds a romantic short from one generated image

• One generated image became a romantic short: the team's opening post credits GPT-6 Astra with helping shape the love story.
• A 30-second video prompt divides the action into four timed scenes, with continuity rules spelled out in the published instructions.
• Title lettering was generated separately in Magnific One, then animated over the final shot, as described in the credits breakdown.

Nobody licks an ice cream in the video prompt, and boarding the Ferris wheel is skipped entirely. The end-card prompt places “just” on the left and “in time” on the right, keeping the couple’s faces clear.

Magnific began with a woman waiting at a night fair, holding two vanilla ice cream cones. The spare cone is an elegantly economical story hook.

The team took the Magnific One image to GPT-6 Astra with a simple request: “Let’s make a short film about love using this image.”

The published video prompt specifies the supplied still as the exact opening frame, followed by four timed beats. The thread does not name the video-generation model.

1. 0–7 seconds, waiting: The woman searches the crowd, glances at the spare cone and lowers her eyes, believing her date is not coming.
2. 7–14 seconds, running late: A man in an orange satin jacket runs through the nearby night market while the camera tracks backward.
3. 14–22 seconds, reunion: He slows, catches his breath and offers an apologetic smile. She smiles back and hands him the spare cone once.
4. 22–30 seconds, Ferris wheel: A clear cut jumps to the couple already seated in a gondola. She rests her head on his shoulder as the camera pulls back to reveal the fairground below.

The team also specified production constraints in its video instructions:

• Identity: Preserve the woman’s face, shoulder-length dark hair, cream cardigan and pale floral dress throughout.
• Props: After the single handoff, maintain exactly one ice cream per person.
• Running: Keep a steady pace, gentle shoulder rotation, coherent background parallax and realistic motion blur. Avoid camera bouncing, floating and a digitally locked face.
• Boarding: Omit movement through gates, barriers or the ride’s structure by cutting directly to the seated couple.
• Anatomy and expression: Avoid morphing, duplicated objects, extra fingers and exaggerated expressions.
• Sound: Request fairground ambience, footsteps, clothing movement and quiet breathing, without dialogue, voiceover, shouting or music.

Magnific generated the lettering separately in Magnific One, then animated it over the ending. The video prompt explicitly excludes generated typography, subtitles, logos and credits, and reserves music for editing too.

Magnific closes the thread with a link to the image tool. Separately from the film demo, its official product page describes this still-image workflow:

1. Draft: Explore a grid of eight or 16 image options.
2. Final: Render the selected image, with optional Brand Kit guidance and brand checks. Final images support up to 4K.
3. Editing: Split an image into editable layers with Auto Layers, or open it in Designer for vector control.

The October 11 launch post advertised an unlimited-use promotion through October 15.