atelier/sidecar/catalogue/shipped.ts

import { CATALOGUE_REVISION, type Catalogue } from './types';

/**
 * The catalogue as it ships with the plugin — the floor a refresh builds on, and the whole of it when
 * no refresh has ever run or every refresh has failed.
 *
 * Written by hand and reviewed in git, because none of it is fetchable: `GET /v1/models` returns ids,
 * an owner and a timestamp, and says nothing about price, sizes, transparency, or whether a model is
 * one you would still choose. That is the entire reason this file exists rather than a nightly fetch.
 *
 * The prices are USD per million tokens, from OpenAI's published pricing, and three of the five models
 * have none — deliberately. `PRICING` in providers/openai.ts had the same gap and the same reason: a
 * guessed price produces a number that looks authoritative and sums into a wrong total.
 */
export const SHIPPED_CATALOGUE: Catalogue = {
  revision: CATALOGUE_REVISION,
  providers: [
    {
      name: 'openai',
      label: 'OpenAI',
      envVar: 'OPENAI_API_KEY',
      defaultModel: 'gpt-image-1',
      models: [
        {
          id: 'gpt-image-1.5',
          label: 'GPT Image 1.5',
          kind: 'image',
          sizes: ['auto', '1024x1024', '1536x1024', '1024x1536'],
          qualities: ['auto', 'low', 'medium', 'high'],
          transparency: true,
          edit: true,
          pricing: { unit: 'token', textInput: 5, imageInput: 8, imageOutput: 32 },
          status: 'current',
          source: 'shipped',
        },
        {
          id: 'gpt-image-1',
          label: 'GPT Image 1',
          kind: 'image',
          sizes: ['auto', '1024x1024', '1536x1024', '1024x1536'],
          qualities: ['auto', 'low', 'medium', 'high'],
          transparency: true,
          edit: true,
          pricing: { unit: 'token', textInput: 5, imageInput: 10, imageOutput: 40 },
          status: 'current',
          source: 'shipped',
        },
        {
          id: 'gpt-image-1-mini',
          label: 'GPT Image 1 mini',
          kind: 'image',
          sizes: ['auto', '1024x1024', '1536x1024', '1024x1536'],
          qualities: ['auto', 'low', 'medium', 'high'],
          transparency: true,
          edit: true,
          pricing: { unit: 'token', textInput: 2, imageInput: 2.5, imageOutput: 8 },
          status: 'current',
          source: 'shipped',
        },
        {
          // No transparency and a different size set — which is exactly what the form was getting
          // wrong by offering one hard-coded list of sizes and qualities for every model.
          id: 'dall-e-3',
          label: 'DALL·E 3',
          kind: 'image',
          sizes: ['1024x1024', '1792x1024', '1024x1792'],
          qualities: ['standard', 'hd'],
          transparency: false,
          edit: false,
          status: 'current',
          source: 'shipped',
        },
        {
          id: 'dall-e-2',
          label: 'DALL·E 2',
          kind: 'image',
          sizes: ['256x256', '512x512', '1024x1024'],
          qualities: ['standard'],
          transparency: false,
          edit: true,
          status: 'current',
          source: 'shipped',
        },
      ],
    },
    {
      /**
       * The first LOCAL provider: sd-server on this machine's GPU, not a vendor.
       *
       * No `envVar`, because there is no key and never will be — see `keyless` on the adapter. It is
       * `configured` by existing.
       *
       * The id is `sd-cpp-local` and not `qwen-image-2.1` because that is what `GET /v1/models`
       * returns, hardcoded in sd-server with no override. The catalogue is the only side that can
       * move, so it moves: the id agrees with the server, and the LABEL carries what the thing
       * actually is.
       */
      name: 'qwen-local',
      label: 'Qwen-Image (local)',
      defaultModel: 'sd-cpp-local',
      models: [
        {
          id: 'sd-cpp-local',
          /**
           * The timings are in the label because there is nowhere else for them, and a user choosing
           * 1024 should know before they wait rather than after. Measured from the server's own log on
           * this machine: 512 completes in ~185s; 1024 samples at ~41.4s/step over 20 steps.
           *
           * A `note` field on CatalogueModel would be the cleaner home for this. Not added on a guess —
           * it needs the UI to render it, and a field nothing displays is worse than a long label.
           */
          label: 'Qwen-Image 2.1 (local) — 512 takes ~3 min, 1024 ~14 min',
          kind: 'image',
          /**
           * 512 first, and the adapter substitutes it for `auto` or for nothing. 1024 is offered
           * because it is genuinely better and sometimes worth the wait — but a default nobody chose
           * should not spend a quarter of an hour of the only GPU there is.
           */
          sizes: ['512x512', '768x768', '1024x1024'],
          transparency: false,
          /** No `edit()` in this version — same GPU, same queue, and the generate path is unproven. */
          edit: false,
          /** Zero is the true answer, not a missing one: it runs on hardware we already own. */
          pricing: { unit: 'image', perImage: 0 },
          status: 'current',
          source: 'shipped',
        },
      ],
    },
  ],
};