Generate2DMedia is the layer the other Forge sprite and map skills call when they need raw media. Its route_media.py command picks the first working route: a provider API whose key you configured (OpenAI, Google Gemini, xAI, BytePlus ModelArk, fal.ai), then your own image tool, then a signed-in Codex or Grok command line tool. It prints one JSON line with the route, the file, a checksum and a cost estimate.
It also runs the capability check, helps configure API keys, keeps a spend ledger, resumes interrupted jobs and adopts images made in a Codex session.
What You Get
- A single command for still images, reference edits and image-to-video clips
- A capability report showing which keys are configured (yes or no) and which local tools are ready
- A ledger file listing every call with its cost estimate, plus optional budget and call limits
- Dry-run mode that shows the plan and estimate without sending anything
Common Use Cases
- Generate a master still for a game character from a prompt file and reference images
- Animate an approved still with an image-to-video model
- Check which generation routes work on this machine before starting a project
- Review spend after a long generation session
What Makes It Unique
Routing is explicit and ordered: configured APIs first, local CLIs next, code art last. It supports several image and video providers behind one interface, records cost in a ledger, snaps video lengths to what each model renders and reports the route that ran. Keys are read only from the environment or a user config file, never from a project.
Under The Hood
The agent runs Python scripts from the skill folder through the shell. They send your prompt file and reference images over the network to the provider you configured, or call your local Codex or Grok command line tool, and save the returned image or video into a new output folder. Keys come from environment variables or a config file in your home config folder, and the skill says never to print or commit them. Paid calls are written to a ledger file in the project. Read the scripts before running them on a sensitive machine; the automated scan findings are shown on the page.