@tpmjs/tools-image-gen

Generate and describe images with whichever AI provider you have a key for — OpenAI, Gemini/Imagen, fal.ai or Replicate — and deliver results as URLs (Discord) instead of megabytes of base64.

Part of the ajax weapons set: task-level, provider-agnostic tools that an agent can pick up without learning a vendor API. Credentials come from the environment; on tpmjs add them as collection env vars (the collection owner's calls get them injected, everyone else supplies their own).

Installation

npm install @tpmjs/tools-image-gen

Environment

| Variable | Required | Purpose | | --- | --- | --- | | OPENAI_API_KEY | no | OpenAI key (gpt-image-1 / dall-e-3 and vision) | | GEMINI_API_KEY | no | Google AI Studio key (Imagen 3 and Gemini vision) | | FAL_KEY | no | fal.ai key (FLUX models; returns hosted URLs) | | REPLICATE_API_TOKEN | no | Replicate token (FLUX models; returns hosted URLs) | | DISCORD_WEBHOOK_URL | no | If set, generated images are posted there and returned as durable Discord CDN URLs | | IMAGE_PROVIDER | no | Force a provider: openai | gemini | fal | replicate (default: first configured, preferring URL-returning providers) |

Tools

| Export | What it does | | --- | --- | | generateImage | Generate one or more images from a prompt using the first configured provider, returning URLs (via Discord delivery or the provider) or inline data. | | describeImage | Describe an image or answer a question about it using a vision model (OpenAI or Gemini). |

Usage

import { generateImage, describeImage } from '@tpmjs/tools-image-gen';

const ctx = { toolCallId: 'c1', messages: [] };
const { urls } = await generateImage.execute({ prompt: 'a pig in a spacesuit, film still', aspect: 'landscape' }, ctx);
console.log(await describeImage.execute({ imageUrl: urls[0] }, ctx));

Every tool throws a readable error on provider failures (status code + provider message), so agents can react instead of guessing.

License

MIT