Generate AI images and videos via Modellix CLI

Modellix connects image, video, and speech generation or transcription to the official modellix-cli workflow. It guides authentication, a readiness check, asynchronous model submission, waiting for the result, and downloading the output. The supported task types include text-to-image, image-to-image, text-to-video, image-to-video, video-to-video, text-to-speech, speech-to-text, and speech-to-speech voice transformation. These operations share a CLI workflow while retaining the request schema and input requirements of the selected model.

The source supplies default model choices when a user has not selected one: google/nano-banana-2-lite for image generation, google/nano-banana-2-lite-edit for image editing, bytedance/seedance-2.0-mini-t2v for text-to-video, and the Seedance 2.0 fast i2v and v2v variants for media-based video work. Its speech examples use alibaba/qwen-audio-3.0-tts-flash, openai/whisper-1, and alibaba/cosyvoice-clone. These configured defaults depend on provider availability, quotas, pricing, moderation, schemas, and generation time, which can change.

Use this skill when a request calls for generating or editing visual media, producing spoken audio, transcribing a recording, or transforming a voice through Modellix. It also fits requests that name providers such as Seedance, Nano Banana, Whisper, Qwen Audio, or CosyVoice through this platform. It requires a Modellix account, network access, a valid API key, and sufficient balance.

Before a paid submission, the workflow discloses the provider, model, prompt or source media leaving the machine, expected cost, and output path, then obtains explicit approval. An uncertain submission outcome is checked through task history before any retry. Generated media needs human review for quality, rights, privacy, and policy compliance before publication. The skill documents the CLI path; it provides no REST fallback and makes no guarantee that a completed remote task will download successfully.

Inputs include a prompt, text to speak, or source image, video, or audio appropriate to the chosen operation. The examples pass request bodies containing prompt text, a voice selection, or a media URL; the voice transformation example also selects a CosyVoice model. Request details come from model describe and its documentation URL or the provider's schema documentation.

Authentication uses a session-scoped key where possible. Persistent credential storage through auth login requires approval. The readiness check is:

modellix-cli doctor --json

Submission uses model run with --wait and JSON output, favoring the CLI's waiting behavior over custom polling. Successful remote results are then downloaded as local files. The destination and overwrite policy must be confirmed first, and an existing file is never replaced without explicit approval. Hosted result URLs are retained for about seven days; download outputs before expiry.

The integration is the official Modellix CLI and platform, with model-specific documentation supplying request schemas. Prompts and uploaded media leave the machine for the Modellix API and CDN processing. Credentials must stay out of logs throughout authentication, submission, history checks, and downloads.