Many people are moved by that moment when they first encounter music generation.

Write a sentence, wait a while, and a complete song with vocals, melody, and arrangement comes out. What once required finding a composer, arranger, singer, and mixing engineer can now be started by one person.

But after actually making a few songs, problems quickly begin to appear.

Generating is not difficult; generating a song you are willing to keep is still difficult.

Sometimes the lyrics are written very carefully, but when sung, they sound like they are being read word by word; sometimes the verse has a good atmosphere, but the chorus does not truly lift; and with some works, the first listen feels amazing, but on a second listen, you realize the melody has no memorable hook. So you can only change the prompt, adjust the style, and generate again, continuing to search among a pile of similar versions.

More troublesome still, when we want to put music generation into our own products, the matter is no longer just about “whether it can generate.” Whether the API is stable, how many songs are returned at once, how to handle task failures, and whether the generation cost is affordable—these seemingly unromantic issues often ultimately determine whether an idea can truly go live.

The value of ACE Music begins to become clear precisely at this stage.

It is not simply adding another music model, but providing creators and developers with another option: using a lighter cost to turn one spark of inspiration or a complete set of lyrics into two song candidates that can be previewed and downloaded, then selecting from them the direction truly worth continuing.

When “generating a song” becomes easy, good results are what become truly scarce

A late night when creation gets stuck

Music creation has a very real contradiction.

The easier it becomes to generate, the more likely we are to spend our time on endless attempts.

One prompt can generate a song, which already seems simple enough. But if a creator has a clear theme, lyrics, and emotion in mind, relying only on a broad description is usually not enough. What we want to control is not just whether it is “pop” or “folk,” but how the verse should build up, when the chorus should open up, whether the female vocal should be restrained or bright, and whether the final line should leave space or land firmly.

These details determine whether a song merely sounds like a song, or can truly be remembered.

ACE Music provides two very direct creation methods in Ace Data Cloud.

If there is only a vague idea, you can use Inspiration Mode to describe the song’s theme, emotion, and musical direction. If lyrics are already prepared, you can submit a title, complete lyrics, and Style, allowing the model to complete the singing and arrangement around clear content. Structural labels such as [Verse] and [Chorus] can also be retained in the lyrics, making the relationship between sections clearer.

One complete generation returns two candidates, rather than providing only one result. Creators can directly compare two melody and vocal treatments, then decide which one is more worth keeping.

This design may look like it simply generates one more song, but in reality it fits music creation habits very well.

We are usually not judging whether a song “passes or fails,” but looking between two directions for the version that is closer to our own idea. With a comparison, it becomes easier to judge whether the melody has a memorable hook, whether the chorus works well enough, and whether the vocal suits the lyrics.

What ACE Music wants to do is not just sing text out loud

From lyrics to two song candidates

In Ace Data Cloud, ACE Music currently provides two models, acemusic-v1.1 and acemusic-v1, with acemusic-v1.1 as the default choice. In addition to regular inspiration generation and custom lyric generation, it also supports importing reference audio, covering existing works, and blending two reference songs.

For people who only want to quickly validate a theme, Inspiration Mode is already sufficient.

For example, enter:

A warm and restrained Chinese folk song, written for someone who has been away from their hometown for many years and suddenly thinks of an old street late at night. Mainly featuring acoustic guitar and gentle strings, with a natural female vocal, without excessive sentimentality.

The model will understand the song direction based on this description.

If the creator has already written the lyrics, they can switch to Custom Mode. The title, lyrics, and Style are submitted separately, so the model does not need to reinterpret the story for you, but directly completes the music around the existing text.

This approach is especially suitable for Chinese creation.

Because Chinese lyrics rely heavily on sentence length, pauses, and tone. If the theme, lyrics, and Style are all crammed into one description, the model can easily focus on the style while overlooking the specific expression. After separating the content from the musical direction, creators can carefully revise the lyrics first, then separately adjust the musical style and vocal feel.

ACE Music also provides a separate lyrics generation API. It can provide a title and lyric draft based on the theme and musical style. After generation, it can still be manually modified before being sent to the song API.

One important point here is: lyric generation and song generation are two separate stages.

It does not force creators to accept all content produced in one go. Writing lyrics first, then revising them, and finally making the song—this process is closer to real creation and also makes it easier to preserve human judgment.

ACE Music and Suno are not about one replacing the other

Different workflows of ACE Music and Suno

When AI music is mentioned, many people’s first reaction is still Suno. This is normal. Suno has already established a fairly complete creation pipeline. In addition to basic song generation, it also covers continuation, partial replacement, stitching, cover generation, Remaster, track addition and extraction, stem separation, and timbre-related capabilities. Among the interfaces currently provided by Ace Data Cloud, Suno is still a more complete post-production toolchain.

ACE Music has a different direction.

It is currently more focused on song generation itself: turning ideas or lyrics into two complete candidates, and it can also use reference audio for cover generation or dual-reference fusion. It does not currently provide parameters such as continuation, stem separation, and voice cloning in the Ace Data Cloud API.

So, a more accurate understanding is not whether “ACE Music can replace Suno,” but that they are suitable for different positions in the creative workflow.

If the goal is to quickly find appealing melodies, natural vocals, and complete song structures from a set of lyrics, ACE Music is a very attractive choice. Its workflow is shorter, and its calling method is also more concentrated.

If, after song generation, there is still a need to continue extending it, replace a certain section, separate vocals and accompaniment, or keep multiple works using the same set of timbres, Suno’s toolchain is currently more mature.

For many content teams, the most practical solution may not be choosing one over the other.

It is entirely possible to first use ACE Music to quickly explore song directions, select the version with the most suitable melody and emotion, and then enter a more detailed post-production workflow according to project needs. Models are only tools; what truly matters is placing different tools where they are best suited.

The advantages of ACE Music are first reflected in creation costs

According to Ace Data Cloud’s current billing rules, ACE Music consumes approximately 0.3148 Credits to fully generate two song candidates. Suno’s current mainstream new models usually consume approximately 0.56 Credits to generate two songs.

Compared using the same Credits standard, ACE Music’s basic song generation cost is about 40% lower.

For people who only make a song occasionally, the difference may not be that noticeable. But for teams that need to repeatedly test styles, generate in batches, or integrate music capabilities into products, costs will quickly be amplified.

For example, with the same set of lyrics, we may need to test:

  • The difference between warm folk music and urban pop
  • The difference between a female solo and a male-female duet
  • The difference between a restrained version and a version with a stronger chorus
  • The impact of different arrangements on the emotion of the lyrics

Each generation produces two candidates. As the scale of calls gradually increases, lower per-call costs mean that products can give users more room to try, rather than having to seriously consider the budget every time they press “regenerate.”

ACE Music’s failed-task billing mechanism is also relatively friendly. Fees are charged per generation group only after a complete two-track group is successfully delivered; if a task ultimately fails, or only one song is completed and the overall task enters a partially failed state, no song generation fee is charged for that group, while completed tracks will still be retained.

For asynchronous generation services, this rule is very important.

Users do not need to bear the cost of results that were not fully delivered, and developers can more easily design retries and exception notifications, instead of continuing to handle billing disputes after a task fails.

Reference audio makes creation no longer rely only on text

When using only Style descriptions for music, a problem can sometimes arise: we know what we want, but it is difficult to describe it accurately.

Words such as “warm,” “retro,” and “cinematic” mean different things to different people. What creators truly want to preserve may be the movement of a melody, a certain sense of rhythm, or the feeling of emotion gradually unfolding in the original song.

ACE Music can first import a publicly accessible MP3 file to generate a reference track that can continue to be used. After that, a Cover can be created around this reference, or two different reference tracks can be placed together for a Mashup.

In cover and fusion modes, audio_influence can also be used to control the degree of influence from the reference audio, with a value range from 1 to 100 and a default value of 60.

This leaves a relatively practical space for creation.

With a higher influence level, the result will be closer to the reference; with a lower level, the model will have more room to reinterpret it. It may not replace professional arranging, but it is much more intuitive than describing music using only a few adjectives.

Of course, reference audio is not a copyright shortcut.

Whether for covers, fusion, or commercial projects, it should be ensured that the uploaded content and final use have the necessary authorization. What technology can do and whether content can be legally used are still two different matters.

Integration is not complicated; what truly needs attention is the task workflow

ACE Music API Integration Workflow

ACE Music mainly provides three types of interfaces in Ace Data Cloud: songs, lyrics, and task queries.

Song generation uses:

POST https://api.acedata.cloud/acemusic/audios

Lyric generation uses:

POST https://api.acedata.cloud/acemusic/lyrics

Task queries use:

POST https://api.acedata.cloud/acemusic/tasks

Before making formal calls, you need to first create or copy an API Key in the Ace Data Cloud console, and ensure that the credentials have ACE Music access permissions and available credits. Requests use Bearer Token for authentication.

If you are only starting from a sentence of inspiration, you can submit a request like this:

curl --request POST 'https://api.acedata.cloud/acemusic/audios' \
  --header 'Authorization: Bearer YOUR_API_KEY' \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "acemusic-v1.1",
    "custom": false,
    "prompt": "一首温暖克制的中文民谣,写给离开故乡多年的人,木吉他与轻柔弦乐,女声自然",
    "async": true
  }'

If lyrics have already been prepared, then custom mode can be used:

curl --request POST 'https://api.acedata.cloud/acemusic/audios' \
  --header 'Authorization: Bearer YOUR_API_KEY' \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "acemusic-v1.1",
    "custom": true,
    "title": "老街的灯",
    "style": "Warm Chinese folk pop, intimate female vocal, acoustic guitar, soft strings, restrained emotion",
    "lyric": "[Verse]\n离开以后才发现\n那条街并没有走远\n\n[Chorus]\n老街的灯还亮着\n像有人等我回去",
    "async": true
  }'

After the request is accepted, it usually first returns a task_id, which does not mean that the song has already been generated.

The correct approach is to save this task ID, then query the progress through the task endpoint. After the task succeeds, the result will include two audio tracks, each with its own title, lyrics, Style, duration, audio URL, and a track ID that can be used for subsequent covers or fusion.

For a production product, it is recommended to also pass in Idempotency-Key.

When the network is interrupted or the page is refreshed, the client can use the original request identity to recover the task, rather than submitting a new paid generation again. Using the same idempotency key and the same request will return the original task; if the idempotency key is the same but the request content is different, it will be rejected.

For products that need to run over the long term, you can also configure callback_url. After the task is completed, the service will push the final result to the specified address, and the frontend does not need to keep polling.

From a generation button to a music product that can truly be used

Integrating a music model is not difficult.

What is truly difficult is making users willing to keep the result.

This requires that the product not only provide an input box and a generation button, but also seriously handle lyric editing, version comparison, previewing, task recovery, favorites, and downloads. The two candidates should be placed together for comparison, rather than scattered across the history; failure states should clearly explain what happened, rather than only displaying “Generation exception”; after a network interruption, the original task should also continue to be recovered, rather than making users start over.

The strengths of ACE Music are exactly suited to this more pragmatic product approach.

It does not package its current capabilities as an omnipotent music workstation, but instead makes the most commonly used path direct enough: enter an idea or lyrics, wait for the task to complete, get two complete songs, then choose the better version from them.

If further refinement is needed, the result can be sent into the subsequent production workflow; if you are simply looking for a complete song for short videos, stories, game scenes, or personal creation, the generated result itself can become a starting point.

Music generation today is no longer as simple as “making a model sing a song.”

We have begun to care whether lyrics are expressed with care, whether the chorus can leave a memory, whether the cost can support repeated attempts, and whether a capability is stable enough to truly enter our own products.

What ACE Music provides is another lighter and more direct path.

It does not make all the decisions for creators, but lets inspiration more quickly have a version that can be heard.

And where many works truly begin is often the first time they are heard.


If you want to try ACE Music yourself, you can enter Ace Data Cloud here:

ACE Music Integration Technical Documentation