Inside the Candela workflow
Candela is the first short film from Magnific Studios, our in-house team. It’s a romantic drama that holds real feeling, tenderness, memory, and the ache of looking back. Made entirely with Claude connected to the Magnific MCP. Proof that with AI, talented teams can carry real emotion.
Table of contents
How Candela is created
Every frame was generated on Magnific, then edited into a love story with its own rhythm, mood, and point of view.
The setup will be familiar if you work with MCP: the team ran the entire production from Claude, with the Magnific MCP giving access to every tool in the conversation. They directed it one stage at a time, and each asset was saved and edited in the Magnific node-based tool, Spaces.
Rather than break down the look, the lighting, and the pacing in the abstract, we asked the Magnific Studios team who made Candela to walk us through it.
The Magnific MCP flow
It started with a single instruction. From Claude, the team asked the Magnific MCP to create a project called Candela and set up its folders, so every asset had a home before any work began. From there, the production moved through six stages, each building on the last:
- Characters: design the cast and age the same faces across the two timelines.
- Locations: establish the beach and the world around it.
- Still frames: generate the key images that anchor each scene.
- Videos: bring the stills to motion, shot by shot.
- Audio: add the score and the sound of the scene.
- Editing: assemble it all into the finished film.
Because everything stayed together in Spaces, nothing was lost between stages, and the project held a consistent thread from the first folder to the final cut.
The full making of Candela is also on YouTube now, broken into short videos.
In conversation with the team
Where did Candela begin?
It began with the story, fully formed, about a year before we generated a single frame. The most overlooked part of any AI project is pre-production, and for us, it was everything: the story, characters, locations, visual style, tone, and a proper script all came first. The clearer those bases, the less time and frustration you spend later, and the fewer credits.
How would you describe the visual world of Candela?
Two times living in one place: a sun-warmed Mediterranean coast on a summer night in 1978, and the present that remembers it. But the identity came less from the palette than from the people. A character has to feel real. So we designed each one with Nano Banana Pro, called through the Magnific MCP in Claude, down to small details (a scar, a few freckles, a particular way of looking), then used those character sheets to age the same faces across decades. This way, they stay recognizable in both timelines.
How did the film come together on Magnific?
Nearly the whole film, around 90 to 95 percent, runs on Seedance 2.0 multishot. The recurring challenge was spatial continuity: keeping Julia and the group consistent around the bonfire from one cut to the next. The fix was to anchor the cast with full-group reference images, then describe each shot in close detail before generating it. We sharpened the key assets with the Image Upscaler before animating, and ran the finished cut through the Video Upscaler. For the score, we started with a track from the Magnific-licensed library and added lyrics and vocals with Suno, shaping the song to the edit.
What surprised you most about working this way?
How human the work stayed. The biggest surprise was how believable the acting was: most of the craft lies in directing the characters well, just as it does when working with real actors. The other surprise was less welcome. Late in the project, we had to replace one of the main characters. We tried swapping her directly via video-to-video, but the quality wasn’t there, so we recreated from the ground up: a fresh casting, a new design, and most of her sequences reproduced as close as possible to the original prompts. One of the most demanding stretches of the whole production.
What does Candela prove about where AI storytelling is heading?
That authorship is the point. AI is fast as long as you don’t care too much about the result, but telling exactly what’s in your head still takes real work. There’s no magic click, no magic prompt, no tool that replaces creative direction. The fundamentals haven’t moved: a good story needs good planning, a good character needs careful construction. The clearer your vision, the more the tool can help you realize it.
If a director or visual storyteller wanted to start tomorrow, where would you tell them to begin?
Start with pre-production, before you generate anything. Then treat casting like a traditional one: test characters and make sure they work across a whole story, not just a single image, before you commit. Write long, specific prompts (the clearer your intention, the better the model answers), and learn the language of film so you can describe exactly what you want. And keep the team small where you can; it’s the simplest way to protect consistency.
Wrapping up
That was the full path Candela took, from the first prompt to the final cut. If the same tools were in your hands, what would you make with them?
