@wavespeed/cli 0.4.27 → 0.4.29

package/README.md

@@ -120,7 +120,7 @@ wavespeed extend video "the camera keeps rising" --video @./clip.mp4 --json
 
 | | Default | Alternatives (`-m`) |
 |---|---|---|
-| Image | `seedream-5-pro` | `gpt-image-2.5`, `nano-banana-2.1`, `nano-banana-pro`, `nano-banana-2`, `gpt-image-2` |
+| Image | `seedream-5-pro` | `gpt-image-2.6`, `gpt-image-2.5`, `nano-banana-2.1`, `nano-banana-pro`, `nano-banana-2`, `gpt-image-2` |
 | Video | `seedance-2.5` | `seedance-2.5-turbo`, `kling-4.0`, `wan-3.0`, `wan-3.0-prime`, `minimax-h3` (cheap and fast, native stereo audio, 768p by default), Seedance 2.0 tiers |
 
 For everything else (audio, 3D, upscaling, a specific endpoint), find the model with `wavespeed models <query>` and call it with `wavespeed run <model-id>`.
@@ -137,7 +137,7 @@ The skill teaches the agent the shortcuts for images and videos (`generate`, `ed
 
 **OpenCode** loads the same file with no extra steps — `.claude/skills/` is on its skill search path. For the MCP tools as well, add the plugin to your OpenCode config — `"plugin": ["@wavespeed/opencode"]` — and run `opencode mcp auth wavespeed` ([WaveSpeedAI/wavespeed-opencode](https://github.com/WaveSpeedAI/wavespeed-opencode)).
 
-**Codex** reads `.agents/skills/`, which the same command writes. For a one-line install that also wires up the MCP server: `codex plugin marketplace add WaveSpeedAI/codex-plugin-wavespeed-cli` ([WaveSpeedAI/codex-plugin-wavespeed-cli](https://github.com/WaveSpeedAI/codex-plugin-wavespeed-cli)).
+**Codex** reads `.agents/skills/`, which the same command writes. For a one-line install that also wires up the MCP server: `codex plugin marketplace add WaveSpeedAI/codex-plugins` ([WaveSpeedAI/codex-plugins](https://github.com/WaveSpeedAI/codex-plugins)).
 
 **DeepSeek Harness (dsh)**: the same skill is packaged for dsh's `.dsh/skills` layout at [WaveSpeedAI/wavespeed-dsh-skill](https://github.com/WaveSpeedAI/wavespeed-dsh-skill).
 

package/dist/lib/api.js

@@ -150,7 +150,7 @@ export function modelPlaygroundUrl(modelId) {
 export async function fetchPricing(modelId, inputs) {
     // /model/price is the documented pricing endpoint; /model/pricing is a
     // legacy contract kept for old clients and absent from the public docs.
-    return apiPost('/api/v3/model/price', { model_id: modelId, inputs });
+    return apiPost('/api/v3/model/price', { model_id: modelId, input: inputs });
 }
 /** The run time in ms the API reports (`timings.inference`), or undefined when absent or zero. */
 export function inferenceMs(item) {

package/dist/lib/media-models.js

@@ -1,6 +1,6 @@
 // The image and video models the dedicated generate/edit tools offer: the
 // same list, in the same order, as the WaveSpeed /chat page (api-dashboard
-// lib/media-models.ts, checked 2026-10-26). The first of each is the default.
+// lib/media-models.ts, checked 2026-11-05). The first of each is the default.
 // This is a vendored copy: when /chat changes its models, change this file and
 // its test (src/lib/media-models.test.ts) with it. Capabilities mirror the
 // endpoints' input schemas; the tools still fit every request to the live
@@ -18,6 +18,7 @@ export const IMAGE_MODELS = {
     'nano-banana-2.1': image('nano-banana-2.1', 'Nano Banana 2.1', 'google/nano-banana-2.1', 'text-to-image', ['1k', '2k', '4k'], NANO_BANANA_2_RATIOS, 14),
     'nano-banana-2': image('nano-banana-2', 'Nano Banana 2', 'google/nano-banana-2', 'text-to-image', ['0.5k', '1k', '2k', '4k'], NANO_BANANA_2_RATIOS, 14),
     'nano-banana-pro': image('nano-banana-pro', 'Nano Banana Pro', 'google/nano-banana-pro', 'text-to-image', ['1k', '2k', '4k'], NANO_BANANA_RATIOS, 14),
+    'gpt-image-2.6': image('gpt-image-2.6', 'GPT Image 2.6', 'openai/gpt-image-2.6-sunburst', 'text-to-image', ['1k', '2k', '4k'], WIDE_RATIOS, 16),
     'gpt-image-2.5': image('gpt-image-2.5', 'GPT Image 2.5', 'openai/gpt-image-2.5-sunburst', 'text-to-image', ['1k', '2k', '4k'], WIDE_RATIOS, 16),
     'gpt-image-2': image('gpt-image-2', 'GPT Image 2', 'openai/gpt-image-2', 'text-to-image', ['1k', '2k', '4k'], WIDE_RATIOS, 16),
 };
@@ -76,16 +77,16 @@ export function videoModelForEndpoint(modelId) {
     return VIDEO_MODEL_KEYS.map((k) => VIDEO_MODELS[k]).find((m) => m.turbo === turbo && modelId.startsWith(`${m.base}/`));
 }
 /**
- * A model's endpoints. Wan, Kling 4.0 and H3 take references on a dedicated
- * reference-to-video endpoint; Seedance takes them on text-to-video. Turbo
- * endpoints have no extend.
+ * A model's endpoints. Every family takes references on a dedicated
+ * reference-to-video endpoint. Seedance has turbo ones; the others do not.
+ * Turbo endpoints have no extend.
  */
 export function videoEndpoints(m) {
     const suffix = m.turbo ? '-turbo' : '';
     return {
         text: `${m.base}/text-to-video${suffix}`,
         image: `${m.base}/image-to-video${suffix}`,
-        ...(m.family === 'wan-3.0' || m.family === 'kling-4.0' || m.family === 'minimax-h3' ? { reference: `${m.base}/reference-to-video` } : {}),
+        reference: `${m.base}/reference-to-video${m.family === 'seedance-2.5' || m.family === 'seedance-2.0' ? suffix : ''}`,
         edit: `${m.base}/video-edit${suffix}`,
         ...(m.modes.includes('extend') ? { extend: `${m.base}/video-extend` } : {}),
     };

package/dist/lib/media-plan.js

@@ -5,16 +5,73 @@
 // the same words give the same model, endpoint and input on both.
 import { DEFAULT_IMAGE_MODEL, DEFAULT_VIDEO_MODEL, IMAGE_MODELS, IMAGE_MODEL_KEYS, VIDEO_MODELS, VIDEO_MODEL_KEYS, referenceFields, videoEndpoints, } from './media-models.js';
 import { renameRefs } from './prompt-refs.js';
-/** Keep only fields the model accepts, snapping values onto its enums and ranges. */
+// Size tiers: 480p, 720p, 1080p (rank = N) and 0.5k, 1k, 2k, 4k (rank = X*1000),
+// so 480p < 720p < 768p < 1080p < 2k < 4k on one scale.
+function tierRank(v) {
+    if (typeof v !== 'string')
+        return undefined;
+    const m = /^(\d+(?:\.\d+)?)([pk])$/i.exec(v.trim());
+    if (!m)
+        return undefined;
+    const n = Number(m[1]);
+    return m[2].toLowerCase() === 'k' ? n * 1000 : n;
+}
+// ln(W/H) of a W:H ratio, so 2:1 and 1:2 are equally far from 1:1.
+function ratioLog(v) {
+    if (typeof v !== 'string')
+        return undefined;
+    const m = /^(\d+(?:\.\d+)?):(\d+(?:\.\d+)?)$/.exec(v.trim());
+    if (!m)
+        return undefined;
+    const w = Number(m[1]);
+    const h = Number(m[2]);
+    return w > 0 && h > 0 ? Math.log(w / h) : undefined;
+}
+// Distances within this count as a tie (4:3 and 3:4 around 1:1 differ only in the last bit).
+const TIE_EPSILON = 1e-9;
+/** The member whose score is closest to target; on a tie, the one `better` prefers (default: the first). */
+function closestMember(members, score, target, better = () => false) {
+    let best;
+    for (const v of members) {
+        const s = score(v);
+        if (s === undefined)
+            continue;
+        const d = Math.abs(s - target);
+        if (!best || d < best.d - TIE_EPSILON || (Math.abs(d - best.d) <= TIE_EPSILON && better({ v, s }, best)))
+            best = { v, s, d };
+    }
+    return best?.v;
+}
+/**
+ * Keep only fields the model accepts and fit each value to its schema. The
+ * rule is shared with the remote MCP (api-server) and the dashboard /chat:
+ * - a size tier (resolution, or an enum of Np / Nk values) the model does not
+ *   offer becomes the closest tier it does; a tie goes to the lower one;
+ * - a W:H aspect ratio becomes the closest ratio (in ln(W/H)) among the
+ *   members that are W:H; a tie goes to the first in enum order;
+ * - any other value outside an enum is left out, so the model default applies;
+ * - numbers go to the nearest numeric enum value, or are rounded (integers)
+ *   and clamped to the range.
+ * Every value not sent as requested gets a note.
+ */
 export async function fitInput(modelId, wanted, propsOf, showDefaults = []) {
     const properties = await propsOf(modelId);
     if (!properties)
         throw new Error(`No schema for ${modelId} in the model catalog.`);
     const input = {};
+    const notes = [];
+    // Defaults filled in only to show the price are not the caller's request: no note for them.
+    const fromDefault = new Set();
+    const used = (key, from, to) => {
+        input[key] = to;
+        if (to !== from && !fromDefault.has(key))
+            notes.push(`${key} ${String(from)} is not offered by ${modelId}; used ${String(to)}.`);
+    };
     // Spell out the defaults that set the price, so the estimate shows them.
     for (const key of showDefaults) {
         if (properties[key]?.default !== undefined && (wanted[key] === undefined || wanted[key] === null || wanted[key] === '')) {
             wanted = { ...wanted, [key]: properties[key].default };
+            fromDefault.add(key);
         }
     }
     for (const [key, value] of Object.entries(wanted)) {
@@ -28,11 +85,28 @@ export async function fitInput(modelId, wanted, propsOf, showDefaults = []) {
             continue;
         }
         if (prop.enum?.length) {
-            if (prop.enum.includes(value))
+            const members = prop.enum;
+            if (members.includes(value)) {
                 input[key] = value;
-            else if (typeof value === 'number' && prop.enum.every((v) => typeof v === 'number')) {
-                input[key] = prop.enum.reduce((a, b) => (Math.abs(b - value) < Math.abs(a - value) ? b : a));
+                continue;
+            }
+            if (typeof value === 'number' && members.every((v) => typeof v === 'number')) {
+                used(key, value, members.reduce((a, b) => (Math.abs(b - value) < Math.abs(a - value) ? b : a)));
+                continue;
+            }
+            const rank = tierRank(value);
+            const tierMembers = members.filter((v) => tierRank(v) !== undefined);
+            if (rank !== undefined && tierMembers.length && (key === 'resolution' || tierMembers.length === members.length)) {
+                used(key, value, closestMember(tierMembers, tierRank, rank, (a, b) => a.s < b.s));
+                continue;
+            }
+            const ratio = ratioLog(value);
+            if (ratio !== undefined && members.some((v) => ratioLog(v) !== undefined)) {
+                used(key, value, closestMember(members, ratioLog, ratio));
+                continue;
             }
+            if (!fromDefault.has(key))
+                notes.push(`${key} ${String(value)} is not accepted by ${modelId}; the model default is used.`);
             continue;
         }
         if (typeof value === 'number' && (prop.type === 'integer' || prop.type === 'number')) {
@@ -41,12 +115,12 @@ export async function fitInput(modelId, wanted, propsOf, showDefaults = []) {
                 v = Math.max(prop.minimum, v);
             if (typeof prop.maximum === 'number')
                 v = Math.min(prop.maximum, v);
-            input[key] = v;
+            used(key, value, v);
             continue;
         }
         input[key] = value;
     }
-    return input;
+    return { input, notes };
 }
 function imageModelKey(named) {
     if (named === undefined || named === '')
@@ -64,10 +138,9 @@ export async function planImage(args, propsOf) {
     const wanted = { prompt: args.prompt, aspect_ratio: args.aspect_ratio, resolution: args.resolution };
     if (edit)
         wanted.images = images;
-    const input = await fitInput(modelId, wanted, propsOf);
+    const { input, notes } = await fitInput(modelId, wanted, propsOf);
     if (typeof input.prompt === 'string')
         input.prompt = renameRefs(input.prompt, input, modelId);
-    const notes = [];
     if (edit && Array.isArray(input.images) && input.images.length < images.length) {
         notes.push(`${model.label} takes at most ${input.images.length} images, so only the first ${input.images.length} were used.`);
     }
@@ -167,7 +240,8 @@ export async function planVideo(args, action, propsOf) {
     const model = VIDEO_MODELS[chosen.key];
     const modelId = endpoints(chosen.key)[chosen.mode];
     // Image-to-video follows the first frame's shape: no aspect_ratio, not even the schema default.
-    const input = await fitInput(modelId, chosen.wanted, propsOf, chosen.mode === 'image' ? ['duration', 'resolution'] : ['duration', 'resolution', 'aspect_ratio']);
+    const fitted = await fitInput(modelId, chosen.wanted, propsOf, chosen.mode === 'image' ? ['duration', 'resolution'] : ['duration', 'resolution', 'aspect_ratio']);
+    const input = fitted.input;
     if (chosen.mode === 'image')
         delete input.aspect_ratio;
     if (typeof input.prompt === 'string')
@@ -184,5 +258,6 @@ export async function planVideo(args, action, propsOf) {
             ? `${model.label} does not take ${dropped.join(', ')} in any mode, so those were left out; the model you named was not switched. ${await whoTakes(dropped, propsOf)}`
             : `No available video model takes ${dropped.join(', ')} for this request, so those were left out.`);
     }
+    notes.push(...fitted.notes);
     return { modelId, label: model.label, mode: MODE_LABEL[chosen.mode], input, notes };
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@wavespeed/cli",
-  "version": "0.4.27",
+  "version": "0.4.29",
   "description": "WaveSpeed CLI — one command to run any WaveSpeed AI model from your terminal.",
   "publishConfig": {
     "access": "public"

package/skills/SKILL.md

@@ -33,7 +33,7 @@ They use the same models as the WaveSpeed /chat page and pick the endpoint (text
 
 | | Default | Alternatives (`-m` / `model`) |
 |---|---|---|
-| Image | `seedream-5-pro` (Seedream 5 Pro) | `gpt-image-2.5` (GPT Image 2.5, the model behind ChatGPT Images; strong at text and photorealism), `nano-banana-2.1`, `nano-banana-pro`, `nano-banana-2`, `gpt-image-2` |
+| Image | `seedream-5-pro` (Seedream 5 Pro) | `gpt-image-2.6` (GPT Image 2.6, the model behind ChatGPT Images 2.6; strong at text and photorealism), `gpt-image-2.5`, `nano-banana-2.1`, `nano-banana-pro`, `nano-banana-2`, `gpt-image-2` |
 | Video | `seedance-2.5` (Seedance 2.5) | `seedance-2.5-turbo`, `kling-4.0`, `wan-3.0`, `wan-3.0-prime`, `minimax-h3` (cheap and fast, native stereo audio; defaults to 768p, its counterpart of 720p), older Seedance 2.0 tiers |
 
 Video resolution is 720p unless the user asks for another (MiniMax H3: 768p). Offer 480p for a cheaper draft.