Skip to content
AI Primer
workflow

Seedance 2.5 workflow produces a 30-second Ali–Tyson fight animation

The workflow uses Midjourney for fighter designs, GPT Image 2.5 for consistent character sheets, and Seedance 2.5 for the arena and 30-second fight. The creator finished the music in Suno and edited the result in CapCut.

3 min read
Seedance 2.5 workflow produces a 30-second Ali–Tyson fight animation
Seedance 2.5 workflow produces a 30-second Ali–Tyson fight animation

TL;DR

Dreamina's Seedance 2.5 product page advertises 30-second videos with up to 50 multimodal inputs. MayorKingAI uses five in his first Seedance prompt, and Dreamina's official video extender page says a continuation draws on a clip's motion, camera flow and visual context.

The image bible

MayorKingAI began with prompt-only Midjourney designs, explaining in a MayorKingAI reply that he used no reference images for either fighter, then used Seedance to give them mannerisms reminiscent of their real-life counterparts.

Ali's source prompt ends with --profile ddpmenj, and Midjourney's parameter guide says such parameters belong after the written description. The GPT Image 2.5 stage in MayorKingAI's character-sheet prompt turns the render into a 16:9 sheet while correcting glove and shorts details.

The sheet locks four things:

  • a close-up, front view and back view
  • facial features, build, silhouette and height-to-width ratio
  • the 2.5D painterly CGI look
  • Everlast branding only on trunks and gloves

A stadium without fighters

The environment is generated separately. MayorKingAI's stadium prompt uses the two character sheets as style references, but explicitly prohibits characters from appearing in the arena image.

The stadium brief fixes a centered ring, tiered crowds, overhead trusses and spotlights cutting through haze. Its job is to carry the sheets' painted texture, sculpted forms and cinematic shading into a shared location.

Five references, one 30-second scene

Seedance receives named assets rather than an undifferentiated pile of images. MayorKingAI's video prompt assigns the five inputs to exact wardrobe states and durations.

  • TYSON1 and ALI1: robe looks for the 0 to 10 second entrances.
  • TYSON2 and ALI2: shirtless fight looks after 10 seconds, with robes prohibited.
  • STADIUM: the ring, crowd and lighting layout for the whole sequence.

The prompt's shot list runs from Tyson's tunnel walk and Ali's ring entrance to the robe drop, face-off and bell. It then gives Tyson bob-and-weave pressure and heavy hooks, gives Ali lateral movement and long jabs, and ends the first clip with Tyson flooring Ali near the blue corner.

The extended ending

The second prompt starts from that knockdown, rather than generating a fresh fight. MayorKingAI's extension prompt has Ali beat the count at eight, evade Tyson's increasingly wild combination, then land a right hand that ends with Tyson on the canvas.

Only the fight-state images, TYSON2 and ALI2, are allowed in the continuation, so the entrance robes cannot reappear. Although the first generation prompt requests diegetic sound only and no score, MayorKingAI's production note says the finished music came from Suno and the final assembly from CapCut.

Further reading

Discussion across the web

Where this story is being discussed, in original context.

On X· 3 threads
TL;DR2 posts
The image bible2 posts
The extended ending1 post
Share on X