Pixcli Creative Production Toolkit

The Problem

Creative agent workflows often split into many model-specific steps: image generation, editing, background removal, SVG vectorization, video generation, voiceover, music, sound effects, and podcast assembly. Calling these models directly forces the agent to choose providers, craft prompts, poll jobs, download files, and avoid common pitfalls such as face drift, long video timeouts, and inconsistent audio stitching. pixcli collapses that surface into one command-level entry point.

How It Works

• Images and assets: image handles text-to-image, reference images, ratios, transparency, and batches, with --vectorize for SVG logos and icons.
• Editing and vector output: edit covers image edits, enhancement, background removal, and upscaling; vectorize converts raster graphics into scalable SVG.
• Video and audio: video supports text-to-video, image-to-video, extension, transitions, and lipsync; voice, music, sfx, and dialogue cover narration, scoring, effects, and multi-speaker dialogue.
• Podcasts: podcast or the sipcast alias turns a topic into a finished show, including script, voices, music, cover art, and a shareable page.
• Final assembly: generated assets can be assembled into a finished video with Remotion templates.

It behaves like a creative task router: the agent describes the desired result, while the CLI/API classifies the job, enriches the prompt, selects a model, and returns parseable output.

Limits and Notes

• It expects a PIXCLI_API_KEY and a Node.js runtime.
• Video jobs can run long; submission plus polling is safer than blocking execution.
• Identity-preserving edits should avoid drift-prone models; the docs recommend identity-preserving routes for people.
• Podcast requests should use the full podcast command rather than manually stitching audio assets.