The Green Anchor · Playbook

AI Music × Virtual Artist

Client Summer Dawn Artist Name TBD Date 2026-07-20 Status Playbook v1 Control surface CLI + AI agents Research /last30days · 86 items · 6 sources
The Promise

One artist, fully synthetic: Summer Dawn writes, the music engine sings, Higgsfield gives it a face - all driven from the terminal.

The Stack
Songwriting + Orchestration

Claude Code or Codex

Lyrics, style prompts, release copy, and the glue that drives every other tool below via MCP, CLI, and REST. The whole pipeline lives in one agent session.

Music Engine - two lanes

Suno + ElevenLabs

Suno for creative quality and vocal identity (Personas). ElevenLabs Music API for anything released commercially: license-clean, fully programmatic. MiniMax via fal.ai as the cheap bulk-iteration lane.

Face + Body

Higgsfield Soul ID

Persistent character identity trained on 20+ images in 3-5 minutes (launched Jul 10, 2026). Same face across every outfit, scene, angle, and video. This is the artist.

Performance Video

Higgsfield Video Models

Kling video + Motion Control: upload a reference performance video and the artist mimics the exact movements while keeping their identity. Lipsync the vocal on top.

Agent Control Layer

Higgsfield MCP + CLI

Official MCP server (OAuth, no API keys) exposes generate_image, generate_video, create_character, get_generation_status, list_characters across 30+ models, straight from Claude Code.

Publish

Cloudflare + Socials

Artist site and press kit on Cloudflare Pages (this page is the proof). Short-form performance clips to TikTok, Reels, Shorts. Streaming distribution decided in Week 4.

The Music Decision

Quality and clean APIs live in different houses

The core tension in July 2026: Suno and Udio lead on output quality but have no official public API and unsettled licensing (training-data lawsuits in flight; a July 15 hack showed Suno scraped YouTube, Deezer, and Genius). ElevenLabs, MiniMax, Stable Audio, and Google Lyria have proper APIs and clean commercial terms, a step behind on vocal quality. So: run two lanes.

ServiceOutput qualityCLI / agent controlCommercial licenseCostVerdict
Suno v5.5 Best vocals + song craft; Personas lock a consistent artist voice across songs No official API yet. Developer API intake form open now (announced Jul 3). Unofficial: gcui-art/suno-api wrapper, 3.1K stars, self-hosted REST Unsettled - lawsuits in flight $10-30/mo; 10 free songs/day Draft lane
ElevenLabs Music v2 Strong, editable (inpainting, section rewrites), multilingual vocals First-class official API + docs; trivially scriptable Clean commercial-use terms ~$0.40-0.80 per generated minute Release lane
MiniMax Music 2.5 Good, improving fast API via fal.ai; simplest per-call pricing for bulk runs Commercial via fal terms ~$0.035 per generation Bulk iteration
Udio High-fidelity instrumentals No API; community reports it became a locked-in system (downloads removed) Unsettled Subscription Skip
Stable Audio Instrumental / texture strength Official API Clean commercial terms API pricing Backup
Lane A - Draft (Suno)

Explore, write, and iterate where quality is highest. Create a Suno Persona for the artist so every track shares one voice. Automate via the self-hosted gcui-art/suno-api wrapper (agent-callable REST), and apply for Suno's official developer API early access today - it is the future of this lane.

Lane B - Release (ElevenLabs)

Anything the artist actually publishes gets generated (or re-generated) in the license-clean lane: ElevenLabs Music API for vocal tracks, MiniMax via fal.ai for cheap variations and beds. If Suno's official API ships with commercial terms, re-evaluate and possibly collapse to one lane.

Agent Control Layer
Higgsfield - official MCP (60-second setup)
claude mcp add --transport http --scope user \ higgsfield https://mcp.higgsfield.ai/mcp

Browser OAuth, no API keys. Tools: generate_image, generate_video, create_character, get_generation_status, list_characters. 30+ models (Soul, Cinema Studio, Kling, Seedream, Hailuo, Veo). Free tier: 150 credits/mo. A separate Higgsfield CLI wraps the same API for scripts.

Higgsfield - Claude Code skills
git clone https://github.com/higgsfield-ai/skills git clone https://github.com/AKCodez/higgsfield-claude-skills

The official skills repo plus a community pack of 19 skills covering image gen, Seedance 2.0 video, and full UGC pipelines. Install into ~/.claude/skills/ and the whole video side becomes prompt-driven.

Suno - unofficial API wrapper (draft lane)
git clone https://github.com/gcui-art/suno-api # self-host, auth with your Suno session cookie # exposes local REST: /api/generate, /api/get

3.1K stars, built "to integrate into agents like GPTs." Cookie-based - treat as fragile and personal-account-scoped. Watch for the official developer API to replace this.

ElevenLabs + fal (release lane)
# ElevenLabs Music API curl -X POST https://api.elevenlabs.io/v1/music \ -H "xi-api-key: $ELEVENLABS_API_KEY" \ -d '{"prompt": "<style + lyrics>"}' # MiniMax via fal.ai (~$0.035/gen) fal run fal-ai/minimax-music --prompt "<style>"

Both key-based and fully scriptable today: cron a nightly batch, have Claude review takes, keep the best. Exact endpoints per current docs at integration time.

The Pipeline - Song to Screen
1

Design the artistGenerate the face with Soul 2.0, train Soul ID on 20+ renders. Lock wardrobe, era, vibe in a style doc.

2

Prep the songStart from Summer Dawn's finished lyrics; Claude structures them for the engine + writes the style prompt from the voice bible.

3

Generate takesSuno Persona (draft) or ElevenLabs API (release). 10-20 takes per song, agent-scored, human-picked.

4

Perform itHiggsfield MCP: Soul ID character + Motion Control reference video + lipsync on the final vocal.

5

PackageCover art (Soul 2.0), canvas loops, 3 short-form cuts per single, press-kit page.

6

Publish + learnSocials first, streaming when licensing lane is settled. Feed metrics back into step 2.

Risks - Eyes Open
Suno legal exposure

Training-data lawsuits are unresolved, and a July 15 hack showed Suno scraped YouTube, Deezer, and Genius. Public sentiment is sharpening ("AI music is theft" content is pulling six figures of views on TikTok). This is exactly why the release lane is ElevenLabs.

Unofficial API fragility

gcui-art/suno-api is cookie-auth against your personal account: it can break on any Suno update and sits in ToS gray space. Keep it draft-only, never load-bearing. The official developer program is the real path.

Platform disclosure

Streaming and social platforms are tightening AI-content labeling. Decide the artist's stance early: openly synthetic is a positioning advantage (see the whole virtual-influencer wave), hiding it is a time bomb.

Voice drift

A virtual artist dies if the voice changes song to song. Suno Personas anchors the vocal identity in the draft lane; matching that timbre in the release lane is the key technical test in Week 2.

Credit burn

Video generation is the expensive step. Storyboard before generating, batch takes, and use Higgsfield's free 150 credits/mo to calibrate before buying a plan.

Name clearance

Before the name lands anywhere public: trademark search, streaming-platform collision check, and handle availability across TikTok, IG, YT, X, plus the .com.

Pricing - verified 2026-07-20
ServicePlanPriceWhat you getRole in stack
SunoFree$050 credits/day (~10 songs), non-commercial onlyTrial before committing
Pro$10/mo ($8/mo annual)2,500 credits (~500 songs), commercial rights, PersonasDraft lane pick
Premier$30/mo ($24/mo annual)10,000 credits (~2,000 songs), Suno StudioScale later if needed
ElevenLabsCreator$22/mo (first month $11)Subscription credits, pro voice cloning, 192 kbps outputRelease lane pick
Music API (self-serve)~$0.15-0.40/min generatedPer-minute metered music generationPer-song release cost; sources conflict on the exact rate, confirm at signup
MiniMax 2.5 (fal.ai)Pay-as-you-go$0.035/generationFull track per call, API/CLI nativeBulk iteration lane
HiggsfieldFree$0~150 credits/moWeek 1-2 calibration
Starter$15/mo (annual billing)200 credits/moLight image work only
Plus$39/mo (annual billing)1,000 credits/moVideo cadence pick
Ultra$99/mo (annual billing)3,000-9,000 credits, 8 parallel generations, cheapest per-creditScale later
Cloudflare PagesFree$0Artist site, press kit, this pagePublishing
DistroKid (Week 4)Musician Plus$44.99/yrUnlimited uploads, custom release dates, pre-saves, 2 artist namesStreaming distribution pick
Real-world add-ons~$150-300/yr all-inContent ID $4.95/single/yr (+20% of CID revenue), Leave a Legacy $29/single if you ever lapseBudget this, not the sticker price
Agent toolingMCP + suno-api + skills$0Higgsfield MCP, gcui-art/suno-api self-host, both skills repos ride the subscriptions aboveControl layer
Pilot month (Weeks 1-4)

Suno Pro $10 + ElevenLabs Creator $22 + Higgsfield free tier $0 + fal.ai pay-as-you-go ~$5 + Cloudflare $0 = ~$37/mo to prove the whole pipeline end to end.

Production cadence (Month 2+)

Swap Higgsfield free for Plus $39 and add DistroKid Musician Plus $44.99/yr at distribution time = ~$76/mo + $45/yr. A 3-minute single with ~15 draft takes and 3 release-lane finals lands around $1.50-4.00 in metered generation on top.

30-Day Plan
Open Decisions
DecisionStatusNotes
Artist nameTBDShortlist in Week 1; clearance checklist above must pass
Genre / laneTBDPick one lane deep rather than three shallow; informs Persona + Soul ID styling
Openly synthetic?Lean yesTransparency reads as a feature in 2026; hiding it is downside-only risk
Suno official APIApplied / waitingIntake form open since Jul 3; would merge the two lanes if terms are commercial
Streaming distributionWeek 4Gate on licensing lane + platform AI policy review