WaveSpeed CLI

Generate images, videos, SVGs, and edit images using 36+ WaveSpeed AI models from the command line.

References

Load these files when needed — do not load all of them upfront:

• references/model-guide.md — decision tree for picking the right model. Load this before any generation or editing task. It maps user intent ("I need a logo", "restore this old photo") to the best model.
• references/cli.md — full CLI command reference. Load before running any wavespeed command.
• references/models.md — full model catalog (36+ models with IDs, types, speeds). Load only when the model-guide doesn't cover the use case or the user asks to browse all models.
• references/installation.md — installation and API key setup. Load if wavespeed is not installed or WAVESPEED_API_KEY is not set.

Environment Check

Before running any workflow, verify the CLI is available and authenticated:

wavespeed --help
echo $WAVESPEED_API_KEY

If wavespeed is not installed or the API key is missing, load references/installation.md and stop until setup is complete.

Workflow Routing

---

workflows/generate-asset.md — Creating something new

• "Generate an image / video / logo / icon for my project"
• "Create a hero image / thumbnail / placeholder / social graphic"
• "Text to image / text to video / text to SVG"
• "I need a visual asset for..."
• "Make me a logo" / "Generate an icon"
• Any single image, video, or SVG generation request

---

workflows/iterate-design.md — Changing something existing

• "Refine this image" / "Try a different style"
• "Change the background / colors / lighting of this image"
• "Edit this photo to..." / "Make it more [adjective]"
• "Relight this" / "Restore this old photo" / "Colorize this"
• "Face swap" / "Change the season" / "Blend these together"
• Any request to modify, improve, or transform an existing image

---

workflows/batch-generate.md — Creating many things at once

• "Generate images for all these pages / sections"
• "I need 5 variations of..." / "Batch generate thumbnails"
• "Make multiple versions with different styles"
• Any request involving multiple generations from a list of prompts

---

Once you identify the right workflow, load that file and follow its instructions.

If the user's intent matches more than one workflow, ask one clarifying question before routing. If it matches none, ask what they are trying to accomplish.

Direct CLI operations — these skip workflows entirely. Load references/cli.md and run directly:

• Upload a file — wavespeed upload <file>
• List models — wavespeed models
• Check status — wavespeed status <request-id>
• Quick one-off where the user already specified all parameters