The quality of AI-generated music depends heavily on the quality of the prompt. A vague description produces vague music. A detailed, well-structured caption with fitting lyrics produces something much closer to what you actually had in mind.
The problem is that writing good music prompts is not obvious. What level of detail does the model need? How should you format lyrics for a 2-minute vs. a 4-minute track? When is a description too specific or not specific enough?
This is what the Creative AI solves.
What the Music Writer does
The Music Writer is a guided writing assistant built into Foundry. Its job is to turn your rough idea into a generation-ready creative brief: a detailed caption with the right structure, and lyrics that fit your chosen duration.
You start with something simple. "I want a mellow jazz track with female vocals." The Music Writer takes that and drafts a full caption: genre, style, instrumentation, vocal character, energy arc. It fills in the technical details that the generation model needs, based on what you described and what tends to produce good results for that kind of music.
If you also want lyrics, it drafts those too, structured with verse/chorus/bridge sections and timed to fit the duration you selected.
Validation and correction
The Music Writer does not just generate text and hand it to you. It validates the output against practical constraints. Is the lyric density too high for the chosen duration? Is the caption missing a critical detail that would cause the generator to guess badly? Are the structural tags misaligned?
If something is off, the system corrects it before you even see the result. This validation loop catches the kinds of prompt problems that would otherwise waste a generation cycle: you generate, listen, realize the prompt was wrong, go back, fix it, regenerate. The Music Writer short-circuits that loop.
Conversational iteration
Once you have a draft, you refine it by talking to the AI. This works like a normal conversation:
- "Make the drums heavier"
- "Add a bigger hook in the chorus"
- "Turn it into a duet"
- "Keep the verse but rewrite the bridge"
- "Make the whole thing darker and more cinematic"
The Music Writer updates the caption and lyrics accordingly, maintaining consistency with what you have already built. It has a back/forward history, so you can navigate between versions the same way you navigate browser tabs.
Multiple intelligence levels
Not every prompt needs the same level of effort. Sometimes you want a quick draft to throw at the generator and see what comes out. Other times you want a carefully refined brief for a specific creative goal.
The Music Writer offers multiple intelligence levels that trade speed for quality. A fast draft runs in seconds with smaller models. A higher-effort refinement produces more detailed, more considered output using larger models.
This is a practical design choice. VRAM is shared between the Creative AI and the other models in Foundry. The intelligence levels let you balance creative assistance with hardware constraints.
Why this matters for output quality
Good prompt engineering is the single highest-impact thing you can do to improve AI music output. A detailed caption with precise genre, instrumentation, vocal description, and energy direction produces noticeably better results than a vague one-liner.
But most people are not prompt engineers. They are musicians, producers, or creators who have a clear idea of what they want to hear and need a way to communicate that idea to the AI precisely. The Music Writer bridges that gap.
It does not replace your creative vision. It translates it. You know what you want. The Music Writer knows how to ask for it in a way the generator understands.
Staying VRAM-aware
The Creative AI runs as a separate quantized language model alongside the music generator and the Planner. On 12 GB cards, all three can coexist. On lower VRAM, the Ultra-VRAM system swaps models in and out as needed.
The writing workflow is designed to feel responsive even when VRAM is tight. The model loads when you open Music Writer and unloads when you switch back to generation, keeping the memory footprint practical.