modellix – skill Lập trình cho agent AI

Modellix

Overview

Modellix is a Model-as-a-Service platform for AI image, video, and speech generation or transcription. This skill teaches agents to use the official modellix-cli workflow (doctor → model run --wait → task download).

Upstream package: https://github.com/Modellix/modellix-plugin/tree/main/skills/modellix (Open Plugins layout; skill tree under skills/modellix/).

When to Use This Skill

• Generate images from text prompts
• Generate or edit videos from text or images
• Generate speech from text, transcribe speech, or transform one voice into another
• Call Modellix models through a unified API/CLI
• The user mentions Modellix, Seedream, Seedance, Nano Banana, Whisper, Qwen Audio, CosyVoice, or similar providers via Modellix

How It Works

1. Authenticate with MODELLIX_API_KEY or modellix-cli auth login

2. Run modellix-cli doctor --json

3. Use default models when unspecified (T2I: google/nano-banana-2-lite, T2V: bytedance/seedance-2.0-mini-t2v, I2I: google/nano-banana-2-lite-edit, I2V: bytedance/seedance-2.0-fast-i2v, V2V: bytedance/seedance-2.0-fast-v2v, TTS: alibaba/qwen-audio-3.0-tts-flash, STT: openai/whisper-1, STS: alibaba/cosyvoice-clone)

4. Submit with modellix-cli model run --wait --json

5. Persist outputs with modellix-cli task download

Examples

Text-to-image

modellix-cli model run \
  --model-slug google/nano-banana-2-lite \
  --body '{"prompt":"A cinematic sunset over a futuristic city"}' \
  --wait --timeout 5m --json

Text-to-video

modellix-cli model run \
  --model-slug bytedance/seedance-2.0-mini-t2v \
  --body '{"prompt":"Ocean waves under a cloudy sunset"}' \
  --wait --timeout 10m --json

Text-to-speech

modellix-cli model run \
  --model-slug alibaba/qwen-audio-3.0-tts-flash \
  --body '{"text":"There is a large garden behind my house.","voice":"longanhuan_v3.6"}' \
  --wait --timeout 5m --json

Speech-to-text

modellix-cli model run \
  --model-slug openai/whisper-1 \
  --body '{"url":"https://example.com/meeting.mp3"}' \
  --wait --timeout 5m --json

Speech-to-speech

modellix-cli model run \
  --model-slug alibaba/cosyvoice-clone \
  --body '{"model":"cosyvoice-v3.5-plus","url":"https://example.com/reference.wav","text":"There is a large garden behind my house."}' \
  --wait --timeout 5m --json

Best Practices

• Prefer CLI model run --wait over hand-rolled polling
• Before a paid submission, disclose the provider, model, prompt or source media that will leave the machine, expected cost, and output path; obtain explicit user approval
• Prefer session-scoped API-key use; run modellix-cli auth login only when the user approves persistent local credential storage
• Do not blindly retry paid submissions after unknown outcomes — check task history
• Confirm the destination and overwrite policy before task download; never replace an existing file without explicit approval
• Download results before they expire; hosted result URLs are retained for about 7 days
• Fetch request schemas from model describe docs_url or https://docs.modellix.ai/llms.txt

Security & Safety Notes

• Requires a Modellix API key; never print secrets in logs
• Prompts and uploaded source media leave the machine for api.modellix.ai and Modellix CDN processing
• Paid generation consumes account balance and must not be submitted or retried without the approval described above

Limitations

• Requires a Modellix account, network access, a valid API key, and sufficient account balance.
• Model availability, request schemas, pricing, quotas, moderation, and generation time are controlled by Modellix and may change.
• Generated outputs require human review for quality, rights, privacy, and policy compliance before publication.
• This skill documents the CLI workflow only; it does not define a REST fallback or guarantee that a completed remote task downloads successfully.