Generate, inspect, and refine AI images on Fal.ai through a Python CLI with cost caps.
npx skills add https://github.com/line7works/tony-skills --skill forge-line7works
Please help me install this Agent Skill. Skill: forge Source: https://github.com/line7works/tony-skills/tree/main/plugins/forge/skills/forge Command: npx skills add https://github.com/line7works/tony-skills --skill forge-line7works
--cap circuit breaker, writes a manifest.json per run, and supports resume after interruptions without double-paying. - Brand consistency: per-project .forge/brand.json profiles set default models, palettes, and styles folded into every prompt. - Use Case: Ask for a set of marketing hero images for a project; the Skill estimates the cost, drafts on a cheap model, shows you the survivors, and finishes your picks at high resolution with optional transparent cut-outs and export size presets. ## Quick Start Ask the assistant to generate two draft mascot images for your project on the nano model with a $0.50 cap, then finish the one you pick at high quality.High-intent search queries and answers about installing and using this skill.
Run the forge Python CLI with a prompt, model, size, and cost cap, such as gen with --model nano --size 1024 --cap 0.50. Claude writes the prompt, executes the CLI, inspects the output images, and iterates on failures before presenting keepers.
Use nano (nano-banana-2) for flat-vector logos, icons, and mascots since it produces clean linework and on-brand color cheaply. Avoid flux for crisp vector marks because it is a photo-diffusion model that yields soft, painterly output.
Pass --cap with a dollar amount on every run. On batch and compare commands the cap acts as a circuit breaker that skips shots exceeding the budget without charging, and resume finishes them later once the cap is raised.
Yes, use the style command with --refs pointing to a folder of reference images, supported on the nano and gpt models. A per-project .forge/brand.json profile also folds palette and style rules into every prompt automatically.
Run resume with the run-id to continue. It re-polls in-flight jobs without paying twice and re-submits only generations that truly failed, reading state from the run's manifest.json.
Do not use it for pixel-art game sprites and animations; that work belongs to PixelLab. This workflow targets illustration, photoreal, marketing, and concept imagery rather than pixel art.
labs21-devExtract brand design systems and handoff tokens from images, mood boards, or briefs.
OMIXECGenerate App Store screenshots from a Next.js workflow with html-to-image.
AutoByteusReview kids coloring assets for story correspondence, character consistency, and print readiness.
GOODDAYDAYGenerate pixel-art sprites and tilesets with MCP-based AI tools and canvas rendering.