Magnific shows how to make a three-part animated mini-series
Magnific starts with character sheets in GPT-6 Astra and uses its MCP workflow to retain a six-member band's visual identity across three scenes. Seedance 2.5 generates the episodes set in a garage, sunflower field, and busking performance.

TL;DR
- The series starts with a character sheet and group portrait, which magnific's reference post says set the faces, outfits and visual style for all three chapters.
- Each short is written as a three-shot scene with explicit camera and audio directions, as magnific's rehearsal prompt demonstrates.
- The fixed six-person band moves from a sunflower field in magnific's field episode to a busking finale in magnific's street finale.
- Magnific describes the stack as GPT-6 Astra plus its MCP workflow in magnific's overview.
- The thread ends as a reusable prompt package, with magnific's prompt invitation asking readers to substitute their own characters.
An amp coughs up a sunflower petal in magnific's rehearsal prompt, then the field chapter places the band among full-size flowers. The street finale gives them a pigeon as their first audience member before the tips arrive.
The character bible
The six characters are defined by individual silhouettes and accessories, not a generic group reference.
- Lola: wavy blonde hair, white oval sunglasses, mustard fur coat.
- Noa: braids, cream beanie, black bomber jacket.
- Kofi: tied dreadlocks with gold beads and gold aviators.
- Rio: spiky hair and goggles on his forehead.
- Peque: cream bucket hat and mustard hoodie.
- Bebo: mustard beanie, yellow shades and black puffer.
The post calls the references GPT-6 Astra creations, then ends with a separate “created with GPT2” label that it does not explain.
Three-shot episode scripts
The rehearsal script schedules an 8-second setup, a 10-second first chord and an 8-second fix. Its repeated prompt structure separates:
- FORMAT & STYLE: aspect ratio, chibi proportions, materials, palette, light and comic tone.
- CHARACTER CONTINUITY: the same six faces, proportions, clothes and accessories.
- SHOT 1, 2 and 3: action, composition or camera move, plus a dedicated audio cue.
The story mechanics remain tiny and visual: the band sets up, the amp smokes, then a single kick brings it back to life.
Garage, field, street
Each location gets a compact narrative beat while the cast block stays intact:
- Rehearsal: the band enters a cluttered garage, blows the amp on its first chord, then resumes playing after the repair.
- Field session: they march through sunflowers, play from a dirt mound, then lie in a circle for the closing overhead shot.
- First crowd: they wheel gear to a city corner, perform first for a pigeon, and finish on coins dropping into a flat cap.
The prompts specify the comedy at the action level: the smallest member jumps twice to get into the field-performance frame, while the street-band instruments keep both hands occupied and nobody claps.
MCP and model handoff
Magnific frames the examples as prompts that can carry a replacement cast through new stories. On its product page, the company says its MCP connection works from Claude, ChatGPT, Cursor and other agents, and advertises multi-image references plus still-image motion control.
The 24.224-second clip attached to magnific's rehearsal prompt fits within the 30-second one-pass limit described in Seedance 2.5's launch post, which also describes the model as reference-based audio-video generation.