RapidComposer v5 beta feedback and discussion

[lulukom]

I agree with lovemusic15 in many ways, especially regarding the ability to train or tune a model on custom MIDI datasets.

[lovemusic15]

I don't think ANY "tool" is going to do what you're asking. Why not just learn to write the music yourself and get EXACTLY what you're looking for?

I disagree with the “just learn to write music” take — and frankly, it feels out of place in a forum about music software.

First, the claim that you “don’t think any tool like that would exist” is incorrect. I’ve researched this area: tools such as Udio and Suno already generate riffs, chord progressions, grooves and MIDI ideas from prompts. Saying no tool could do this is misleading given what’s already out there.

To be clear: I’m not asking for tools that recreate entire songs by particular artists — that would be unethical and pointless. I’m asking for idea generation that can be steered toward a genre or style: strong startup ideas (riffs, chord progressions, grooves) that feel close to a style so I can develop them. Some software already offers genre‑specific MIDI generation; others give largely random seeds. It makes sense for RC (especially as it explores neural networks) to add clearer style controls and user‑data training so outputs better match what users want.

Second, saying everyone should “just learn it yourself” misunderstands how creativity and professional workflows work. Artists don’t always wake up with perfect ideas — that’s why they use teams, co‑writers, engineers and specialists. Telling people to do everything solo is narrow‑minded. Software that generates ideas is another legitimate form of creative support: it supplies inspiration and usable starting points so artists can focus their specialized skills where they add most value.

Using software to generate ideas that match what we want is no different from using a sample pack, a preset sound, or picking up a guitar and playing it — all are pre‑made materials musicians use as kickstarts. Nobody expects a guitarist to have built their own instrument before they play, and no one bats an eye when a producer uses loops or presets. AI‑generated riffs or MIDI ideas are simply another form of legitimate starting material: you take it, reshape it, and make it your own. Dismissing that workflow as lazy or “cheating” ignores decades of how music is actually made and unfairly discourages people who just want to create. Telling someone to “just learn it yourself” is the same as insisting they invent every tool or sound before they create — it’s gatekeeping, plain and simple.

The developer of this software is a software developer, not a producer, so they rely on user feedback to improve the product. Many features exist because forum suggestions drove them. A lot of visitors are new to music, hobbyists, or can’t invest in formal training; hearing “just learn it yourself” can discourage them from requesting features or giving feedback. If everyone had that attitude, the product would stagnate.

As a working producer, I use these tools not to copy hits but to get usable starting points in genres I don’t regularly work in or to speed up ideation for client work. If a client wants a rock track in a style I don’t specialize in, a generated idea gives me a head start — that’s efficient, professional practice. I’m asking for better control and closer genre/style steering, not tools that output full replicas of famous songs.

Practical upgrade point: version 5’s idea tool gives random seeds, which is useful for some. But in an AI era where users expect steerable, genre‑aware generation and the option to train models on personal MIDI/sample datasets, an upgrade should offer those capabilities. If the next version can’t provide clearer, genre‑ or user‑tailored idea generation, many users may see little value in upgrading. Clarifying whether RC plans to add these features is a fair question for anyone deciding whether to upgrade.

Everyone is entitled to an opinion, but if you’re going to be dismissive, do your research first. Read the comments on RC’s official channels and social media about AI music — producers consistently ask for more control and the ability to generate ideas that match the music they love. Many people don’t have the time, money, or opportunity to learn everything from scratch; they want tools that get them closer to their goals quickly so they can create and progress. If you’re going to criticize, back it up with facts and specific technical concerns. Throwaway lines like “just learn it yourself” shut down newcomers, discourage useful feedback, and harm the community — think before you post.

[sj1]

Just dropped the other day, IC3:

IC has always been a good program, but I think this v3 really takes it interstellar.

I commend in particular the song structure and generation features as food for thought. This is an area where RC seems poised to also excel, provided it gets the specific attention needed. IMO, this would be an excellent choice for next priority after adding recording and loop-on-touch to the LIVE MODE.

[sj1]

Q. Where is the setting that will determine if RC loops to the beginning when reaching the end of the song (timeline), or simply stops?

[BluGenes]

Q. Where is the setting that will determine if RC loops to the beginning when reaching the end of the song (timeline), or simply stops?

right click the play button

[sj1]

Thanks for the reply. What I see there (still v5) is this:

Snag_31b557.png

None of those options, as written, are "Stop At End of Composition", or "Do Not Loop At End of Composition".

My goal is to have compositions cease playback when they reach the end. IOW, no starting over (in an endless loop) unless I've specifically asked for that.

I am still not sure how to make that happen. Can it happen currently? If so, then I think possibly we need enhanced labeling/help. If not, then adding this as an explicit option would be helpful.

[sj1]

Well, now it is stopping. Before (yesterday) it was looping. Ah well.

[sj1]

Q. Is there a way to type in the Audio Gain value for a track? (e.g. -6 dB)

[Brian Collins]

Question: Why is "Midi" instrument greyed out.

I am trying a workflow with RC stand-alone (rather then in my DAW, where I think it is causing hangs). I have set up a loopMidi device for passing Midi to my DAW, but the track properties do not let me select it:


Any idea why the Midi option might be greyed out?

[BluGenes]

That is just a title, click the one that says "2 other instruments"

[Brian Collins]

That is just a title, click the one that says "2 other instruments"

D'oh! I should have spotted the pattern from the grey VST & SF titles. Thank you.

[sj1]

Yeah, it's a minor thing, but I found the grey color of those titles confusing also.
Maybe a vibrant blue instead?

[sj1]

I have set up a loopMidi device for passing Midi to my DAW, but the track properties do not let me select it:

i confirm that. LoopMIDI ports do not appear here either as destinations. Logically, they should, for most flexible operation.

IOW, both plugins and ports are desirable as potential outputs (instruments) for RC tracks.

[lovemusic15]

I agree with lovemusic15 in many ways, especially regarding the ability to train or tune a model on custom MIDI datasets.

Thanks — I appreciate the support. I’ve used RC for years and I really believe it should prioritise custom datasets.

Over the years I’ve seen companies drawing inspiration from RC and turn into household names or be bought out by bigger players. That’s a shame because RC already offers so much, but it has a real opportunity to grow its user base by giving people what they’re asking for: control.

Custom datasets are the next big step. With tools like Suno showing you can create exactly what you want (and even export MIDI), AI music has moved from gimmick to must-have for producers across genres. People don’t want random, generic results anymore — they want models that reflect their tastes and samples. Imagine a generator that only outputs phrases and sounds you actually like — it’s like having your own personal Spotify playlist that always recommends stuff you’ll enjoy. That’s the real advantage of a neural-network approach: it should work for the individual, not for a one-size-fits-all output.

If RC enables user-trained datasets and finer control, it would help RC stand out from the crowd instead of watching others reap the rewards. Tons of people in forums are already searching for AI that accepts their own samples and inputs — I know I am.

[lulukom]

Yes, that's right. If I'm writing jazz today, I want to load a template I've trained and honed based on my jazz harmonies and grooves. Tomorrow I'll be writing funk and using my trained funk template. A dream!