SkyProduction Puts Music, MVs, and Soundtracks on One Infinite Canvas

On September 27, Kunlun Wanwei's SkyProduction (Tiangong Workbench) launched three capabilities at once: AI MV creation opened to users, the infinite canvas integrated three music generation models—Mureka V8, V9, and V9.5—and intelligent soundtracking for videos and images went live at the same time.

These three things used to be scattered across different tools. A song is finished in one place, visuals are generated in another, and the soundtrack is found in a third. Moving between tools produces no work in itself, yet it consumes a great deal of time, requires paying for multiple products separately, and involves several rounds of less-than-ideal generation rolls. SkyProduction's approach is to bring them onto a single canvas, where assets, nodes, and results all stay in place.

Three Entry Points Along a Single Pipeline

The three features correspond to three starting points; they can be used separately or together.

If you don't even have a song yet, you can create an audio node directly on the canvas and call Mureka V9.5 to generate music. Write a prompt about genre, mood, and vocal style, and you'll get lyrics and a finished song; you can also choose to skip lyrics and get only the instrumental. Compared with previous generations, Mureka V9.5 improves in four areas: the structural progression of the whole song is more natural, the arrangement adjusts its layers and density with each section, vocal performance is more believable, and execution of genre, mood, vocal, and arrangement requirements is more stable. In the company's words, it advances music generation from “producing every element” to “putting every element where it needs to function.”

If the footage is already edited but the soundtrack isn't decided, you can use the newly launched intelligent soundtracking. Video nodes and image nodes on the infinite canvas both support this feature; the visual content and supplementary description together serve as the basis for generation—the image itself is the creative brief for the music. Users can choose different music models, set the time range for the soundtrack, and keep or discard the original video's audio as they wish. After generation, the soundtrack returns to the canvas as a new audio node, where it can be previewed and used directly as material for subsequent compositing.

If what you want is a complete MV, the AI MV feature first analyzes the song's structure, rhythm, and lyrics, generates a director's plan and storyboards, then uses Seedance 2.5 to generate visuals while maintaining character and scene consistency across shots. This capability was previously made available first in Tiangong Engine and is now open to general users.

What's Worth Noting

Taken individually, none of these capabilities are new. Music generation has Suno, video generation has Kling and Seedance, and for soundtracks there are all kinds of ready-made asset libraries. What's truly different is that SkyProduction puts them into the same context.

Songs generated by music generation, segment audio cut from AI MV storyboards, and finished pieces from intelligent soundtracking are all audio nodes on the canvas. Later nodes can reference them directly, with no need to download, upload, and organize them all over again. For people making AI short films and animations, this saves more time than a single model's image-quality improvement.

There's another easily overlooked change. In the past, music and video were more like two independent production lines, with copyright, format, and rhythm all needing to be manually aligned; now they share the same context. The beat of the soundtrack can directly reference the visuals, and the visuals of an MV are driven by the song's structure. For people creating serialized content, this alignment is itself a consistency tool, saving the effort of repeated rework.

This also continues SkyProduction's direction this year. Kunlun Wanwei's 2026 semi-annual report mentioned that SkyProduction, relying on Agent technology, integrated video models such as SkyReels, Seedance, and Kling to deliver one-stop industrialized AI short-drama production that goes “from script straight to finished film.” In July, the product underwent a major iteration, launching a dual-track creation mode of “Agent intelligent storyboarding + infinite canvas,” with a director Agent responsible for script coordination, character, scene, and camera-position planning, and the infinite canvas supporting 720-degree panoramic image reasoning, multi-angle and light-source editing, and a 3D director's console. On Mureka's side, it connected the full chain from creative ideation to “lyrics, vocals, characters, songs, and MVs,” and also supports dynamic covers and custom AI singers.

Tools Handle Generation, but the Choices Remain with People

Zhou Zhou and Jinhua, the main creators of Wutiaoren's “Earth Love Song” AI MV, said something in an interview that fits well here: they still hope audiences ultimately focus on the work itself, no matter what tools it used.

This SkyProduction update does not answer the question of whether AI-generated works are good. It merely compresses repetitive labor. The tools handle breakdown and generation, but how to frame the shot, where the emotion should go, and which version to keep in the end are still up to the creator.

For independent creators and small teams, this means a process that used to require three or four subscriptions and three or four software programs can now be explored on a single canvas. The cost lies in the same place: when every stage can be compressed into a few minutes, the only difference between works is judgment itself.