feat(media): route Open Design Cloud generation through Vela

Add Vela image and video adapters behind od media generate with trusted workspace attribution, multi-image support, video polling, and fail-closed behavior.

Expose AMR-specific media defaults and pin the packaged Vela CLI used by the integration.

Author: Cheems

apps/daemon/src/cli.ts

@@ -1474,6 +1474,11 @@ async function runMediaGenerate(rawArgs) {
     console.error('--model required (see http://<daemon>/api/media/models)');
     process.exit(2);
   }
+  const images = repeatableFlagValues(rawArgs, 'image');
+  if (flags.model.startsWith('vela/') && images.length > 5) {
+    console.error(`Vela media accepts at most 5 --image values; received ${images.length}`);
+    process.exit(2);
+  }
 
   // Long-form media prompts (detailed image/video descriptions, program-
   // generated prompts) arrive via --prompt-file <path|-> (stdin) per the CLI
@@ -1490,7 +1495,8 @@ async function runMediaGenerate(rawArgs) {
     voice: flags.voice,
     audioKind: flags['audio-kind'],
     compositionDir: flags['composition-dir'],
-    image: flags.image,
+    image: images[0],
+    images,
     language: flags.language,
   };
   if (flags.length != null) body.length = Number(flags.length);
@@ -1832,11 +1838,11 @@ Common options:
                             to the dir containing hyperframes.json /
                             meta.json / index.html. The daemon runs
                             \`npx hyperframes render\` against it.
-  --image <path>            Project-relative path to a reference image
-                            (image-to-video for Seedance i2v models, or
-                            future image-edit endpoints). Daemon reads
-                            the file from the project, base64-encodes
-                            it, and forwards it to the upstream API.
+  --image <path>            Project-relative reference image; repeat up to 5
+                            times for Vela image editing or video references.
+                            The first video image is the first frame; the rest
+                            are references. Existing providers still receive
+                            the first image through the legacy single-image field.
   --daemon-url <url>
 
 Output: a single line of JSON: {"file": { name, size, kind, mime, ... }}

apps/daemon/src/media/index.ts

@@ -80,6 +80,7 @@ import {
   fetchImageGenerationWithResponseRetry,
   type ImageGenerationRequestSummary,
 } from './image-generation-retry.js';
+import { renderVelaImage, renderVelaVideo } from './vela.js';
 import { codexNeedsDangerFullAccessSandbox } from '../runtimes/defs/codex.js';
 import {
   ensureProject,
@@ -144,6 +145,7 @@ type MediaContext = {
   /** Additional reference images for multi-image i2v / style reference flows. */
   imageRefs: ImageRef[];
   projectRoot: string;
+  workspaceId: string | undefined;
   onProviderRequestSettled:
     | ((summary: ImageGenerationRequestSummary & { providerId: string }) => void)
     | undefined;
@@ -326,6 +328,7 @@ export async function generateMedia(args: {
   prompt?: string; output?: string; aspect?: string; length?: number; duration?: number; voice?: string;
   audioKind?: AudioKind; language?: string; loop?: boolean; promptInfluence?: number;
   compositionDir?: string; image?: string; images?: string[]; onProgress?: ProgressFn; requestInit?: MediaRequestInit;
+  workspaceId?: string;
   onProviderRequestSettled?: (summary: ImageGenerationRequestSummary & { providerId: string }) => void;
 }) {
   const {
@@ -347,6 +350,7 @@ export async function generateMedia(args: {
     compositionDir,
     image,
     requestInit,
+    workspaceId,
     onProviderRequestSettled,
   } = args;
 
@@ -431,7 +435,15 @@ export async function generateMedia(args: {
   // when stubs are swapped for paid integrations.
   const lengthClamp =
     surface === 'video'
-      ? clampWithWarning(length, VIDEO_LENGTHS_SEC, 'length')
+      ? def.provider === 'vela'
+        ? {
+            value:
+              typeof length === 'number' && Number.isFinite(length)
+                ? length
+                : undefined,
+            warning: null,
+          }
+        : clampWithWarning(length, VIDEO_LENGTHS_SEC, 'length')
       : { value: undefined, warning: null };
   const usesProviderSpecificAudioDuration =
     def.provider === 'elevenlabs'
@@ -510,6 +522,7 @@ export async function generateMedia(args: {
     requestInit: requestInit || {},
     imageRefs,
     projectRoot,
+    workspaceId,
     onProviderRequestSettled,
   };
 
@@ -572,6 +585,19 @@ export async function generateMedia(args: {
       bytes = result.bytes;
       providerNote = result.providerNote;
       suggestedExt = result.suggestedExt;
+    } else if (def.provider === 'vela' && surface === 'image') {
+      const result = await renderVelaImage(ctx);
+      bytes = result.bytes;
+      providerNote = result.providerNote;
+      suggestedExt = result.suggestedExt;
+    } else if (def.provider === 'vela' && surface === 'video') {
+      const result = await renderVelaVideo({
+        ...ctx,
+        onProgress: args.onProgress,
+      });
+      bytes = result.bytes;
+      providerNote = result.providerNote;
+      suggestedExt = result.suggestedExt;
     } else if (def.provider === 'openai' && surface === 'image') {
       const result = await renderOpenAIImage(ctx, credentials);
       bytes = result.bytes;
@@ -765,7 +791,7 @@ export async function generateMedia(args: {
     // HyperFrames is a local render, not a remote provider. Falling back
     // to a stub here hides actionable composition/preflight failures and
     // can make the agent retry or narrate a fake MP4 as success.
-    if (def.provider === 'hyperframes') {
+    if (def.provider === 'hyperframes' || def.provider === 'vela') {
       throw err;
     }
     // A real provider failed (network blip, 4xx, missing key, …). We

apps/daemon/src/media/models.ts

@@ -33,6 +33,7 @@ export type MediaModel = {
 export const MEDIA_PROVIDERS: MediaProvider[] = [
   { id: 'openai', label: 'OpenAI', hint: 'gpt-image-2 / dall-e-3', integrated: true, defaultBaseUrl: 'https://api.openai.com/v1' },
   { id: 'codex', label: 'Codex Subscription', hint: 'gpt-image-2 via local Codex CLI login', integrated: true, credentialsRequired: false, docsUrl: 'https://developers.openai.com/codex' },
+  { id: 'vela', label: 'Open Design Cloud', hint: 'Managed image and video generation through Vela', integrated: true, credentialsRequired: false, settingsVisible: false },
   { id: 'volcengine', label: 'Volcengine Ark (Doubao)', hint: 'Seedance 2.0 / Seedream', integrated: true, defaultBaseUrl: 'https://ark.cn-beijing.volces.com/api/v3' },
   { id: 'grok', label: 'xAI Grok Imagine', hint: 'grok-imagine — image + video with native audio', integrated: true, defaultBaseUrl: 'https://api.x.ai/v1' },
   { id: 'hyperframes', label: 'HyperFrames', hint: 'Local HTML -> MP4 renderer', integrated: true, credentialsRequired: false, settingsVisible: false },
@@ -85,6 +86,11 @@ export const MEDIA_PROVIDERS: MediaProvider[] = [
 ];
 
 export const IMAGE_MODELS: MediaModel[] = [
+  { id: 'vela/gpt-image-2', label: 'gpt-image-2 (Cloud)', hint: 'Open Design Cloud · managed image generation and editing', provider: 'vela', caps: ['t2i', 'i2i'] },
+  { id: 'vela/nano-banana-2', label: 'nano-banana-2 (Cloud)', hint: 'Open Design Cloud · managed image generation and editing', provider: 'vela', caps: ['t2i', 'i2i'] },
+  { id: 'vela/nano-banana-2-lite', label: 'nano-banana-2-lite (Cloud)', hint: 'Open Design Cloud · fast managed image generation and editing', provider: 'vela', caps: ['t2i', 'i2i'] },
+  { id: 'vela/seedream-5.0', label: 'seedream-5.0 (Cloud)', hint: 'Open Design Cloud · managed image generation and editing', provider: 'vela', caps: ['t2i', 'i2i'] },
+  { id: 'vela/seedream-5.0-pro', label: 'seedream-5.0-pro (Cloud)', hint: 'Open Design Cloud · high-quality managed image generation and editing', provider: 'vela', caps: ['t2i', 'i2i'] },
   { id: 'gpt-image-2', label: 'gpt-image-2', hint: 'OpenAI · 4K, native multimodal', provider: 'openai', caps: ['t2i', 'i2i', 'inpaint'], default: true },
   { id: 'gpt-image-1.5', label: 'gpt-image-1.5', hint: 'OpenAI · 4× faster than gpt-image-1', provider: 'openai', caps: ['t2i', 'i2i', 'inpaint'] },
   { id: 'gpt-image-1', label: 'gpt-image-1', hint: 'OpenAI · ChatGPT native', provider: 'openai', caps: ['t2i', 'i2i', 'inpaint'] },
@@ -152,6 +158,7 @@ export const IMAGE_MODELS: MediaModel[] = [
 ];
 
 export const VIDEO_MODELS: MediaModel[] = [
+  { id: 'vela/doubao-seedance-2-0-260128', label: 'seedance-2.0 (Cloud)', hint: 'Open Design Cloud · managed text/image-to-video · 720p default', provider: 'vela', caps: ['t2v', 'i2v'] },
   { id: 'doubao-seedance-2-0-260128', label: 'seedance-2.0', hint: 'ByteDance · t2v + i2v + audio', provider: 'volcengine', caps: ['t2v', 'i2v', 'audio'], default: true },
   { id: 'doubao-seedance-2-0-fast-260128', label: 'seedance-2.0-fast', hint: 'ByteDance · faster, cheaper', provider: 'volcengine', caps: ['t2v', 'i2v', 'audio'] },
   { id: 'doubao-seedance-1-0-pro-250528', label: 'seedance-1.0-pro', hint: 'ByteDance · 1.0', provider: 'volcengine', caps: ['t2v', 'i2v'] },