Cuenca rewrites Candela after early cut resembled San Junipero

• An unused three-minute cut called San Juan centered on an Alzheimer’s flashback to a night of first love, as cuenca’s original-cut post describes.
• The team changed the title, story and central character after cuenca’s explanation said the characters, setting and story elements felt too close to Black Mirror’s “San Junipero.”
• Candela became the Cannes Lions release, and cuenca’s announcement identifies it as the final music video.
• The final film was a materially different production, not a quick rename: cuenca’s Cannes post accompanies reporting of 2,591 generations, 2.5 million credits and a roughly 10-day, three-person build.
• Current Seedance 2.5 makers are publishing multi-stage look-development pipelines, as techhalla’s Magnific workflow demonstrates with character references, generated stills, audio and video references.

Magnific’s Cannes recap names Myriam Mira as Candela’s director and puts the final count at 2,591 generations. Its pre-premiere production note adds the operational detail: character consistency across scenes, work organized in Spaces, and MCP agents holding project context.

Late in May, cuenca’s first post says Luisa sent over a video that the account found emotionally affecting after watching it later that day. The project was called San Juan: a woman with Alzheimer’s has flashbacks to the night she fell in love, set against Mediterranean imagery, Spanish language and a catchy song.

The internal alarm was specificity, not a generic concern about AI originality. cuenca wrote in the rewrite note that the characters, location and parts of the story resembled “San Junipero,” the Black Mirror episode.

The team then changed the story’s name and its main character. The same explanation calls this kind of revision one of the options AI makes easier, while leaving the exact replacement names and beats undisclosed.

The replacement retained the San Juan connection. Magnific’s official Cannes account calls Candela an intimate story from a personal memory tied to that night, and says it received its world premiere during the company’s Debussy Theatre seminar.

Reason Why’s production report identifies the three-minute film’s song as “No Lo Entiendo” by Müxa and describes two timelines joined by the love story of two women. Myriam Mira, the film’s director and a graphic designer, said the aim was for audiences to forget they were watching AI-generated work.

Reason Why’s production report identifies Mira, Carmen Callejón and Cristina Cortés as the three-person team. It says the film took about 10 days, with no shoot, cast or locations.

The published production bill breaks down as:

• 2,591 generated assets.
• 2.5 million credits consumed.
• Nano Banana 2 for images.
• Seedance 2.0 for video.
• AI-generated source material for the soundtrack, then restructuring and mastering by Mira.

The same report says Spaces was used to organize shots, scenes and references in one visual canvas, giving Mira a way to inspect each shot within the larger film.

Magnific’s pre-premiere post describes the project as one connected workflow with three explicit components:

1. Consistent characters across every scene.
2. 2,591 generations inside Spaces.
3. MCP agents keeping context under control.

The company’s MCP documentation says its remote server lets a connected AI assistant generate images and video, train consistent characters, browse generation history and use assets on the creator’s existing Magnific account. The public Candela accounts describe the workflow architecture, not the film’s individual prompts.

Candela’s documented video model was Seedance 2.0. A later Magnific post advertised Seedance 2.5 on its platform, powered by BytePlus, while creator posts show the surrounding production pattern rather than a single generation.

• CharaspowerAI’s PRE-ECHO breakdown describes a two-minute found-footage horror sequence built from a script, character references, a custom playbook, Seedance 2.5 generations, continuity fixes and a final edit in Slate.
• techhalla’s workflow starts with Recraft V4 Pro for the initial face, locks the character, uses Seedream 5 Pro for the scene stills, then animates in Seedance 2.5. The first clip uses audio and references so cuts land on beats; the second uses the preceding video plus the final perfume still as references.
• Two smaller demonstrations show the model in circulation: bennash posted a 72-second Seedance clip, while techhalla’s follow-up carries a 34-second perfume-film demo.