atomantic · atomantic · Sep 3, 2026 · Sep 3, 2026
diff --git a/client/src/hooks/README.md b/client/src/hooks/README.md
@@ -206,7 +206,7 @@ grep -i "what you want to do" client/src/hooks/README.md
 | `useYoutubeIngest` | One YouTube brain-ingest job slot: start/cancel + SSE progress + terminal-frame handling via `POST /api/brain/youtube/ingest`. Returns `{ active, percent, stage, start(body), cancel }` — `start` takes the whole payload (`{ url, captureTranscript, downloadVideo, ingestAudio, note, agentPrompt, tags }`), not a bare URL. `onComplete(ingest)` fires with the stored ingest record; non-fatal `warnings[]` on the terminal frame are toasted automatically. | Quick Capture's YouTube path, and any other surface that ingests a video into the brain. |
 | `useYoutubeTrackImport` | One YouTube-audio-import job slot (#1945): start/cancel + SSE progress + terminal-frame handling via `POST /api/tracks/import/youtube`. Returns `{ active, percent, start(url, context), cancel }`; `onComplete(track, context)` fires with the finished Track. Call once per UI surface that can independently kick off an import — a shared slot would let one surface's kickoff orphan another's in-flight job. | Any picker that lets a user paste a YouTube URL to create a Track (Music Video's create form and track-change row). |
 | `useVideoGenFieldState` | Mutable VideoGen field values, setters, and lifecycle refs initialized from URL handoff values. | Internal state slice composed by `useVideoGenForm`; keep model reconciliation and submission behavior in their focused units. |
-| `useVideoGenForm` | `useVideoGenForm({ models, status, availableLoras, grokEnabled })` → every VideoGen form field + setter, the URL-param prefill paths (ImageGen handoff, Continue, Remix, `?lora=`), the mode/backend transitions that clear stale inputs, the derived model/keyframe/IC gates (`extendModeBlocked`, `a2vModeBlocked`, `icLoraModeBlocked`, `keyframesError`), `applyRemix` / `applyResumedParams`, and `buildGeneratePayload()`. | The /media/video form — the single client-side source of truth for the payload `server/routes/videoGen.js` validates. Reuse it for any second entry point that generates video. |
+| `useVideoGenForm` | `useVideoGenForm({ models, modelContext, availableLoras, grokEnabled })` → every VideoGen form field + setter, the URL-param prefill paths (ImageGen handoff, Continue, Remix, `?lora=`), the mode/backend transitions that clear stale inputs, the derived model/keyframe/IC gates (`extendModeBlocked`, `a2vModeBlocked`, `icLoraModeBlocked`, `keyframesError`), `applyRemix` / `applyResumedParams`, and `buildGeneratePayload()`. | The /media/video form — the single client-side source of truth for the payload `server/routes/videoGen.js` validates. Reuse it for any second entry point that generates video. |
 | `useVideoGenSubmitFlow` | Builds the current VideoGen wire payload and enveloped prompt from a form-state snapshot. | Internal submission slice composed by `useVideoGenForm`; all local, Grok, and federated request shaping remains in `videoGenSubmission`. |
 | `useVideoGenValidation` | Derives keyframe errors and all VideoGen submit-blocking predicates from the current fields and model capabilities. | Internal validation slice composed by `useVideoGenForm`; also exports `validateVideoKeyframes` for focused tests. |
 | `useVideoDownload` | One full-video-download job slot (#1946): start/cancel + SSE progress + terminal-frame handling via `POST /api/devtools/video-download`. Returns `{ active, percent, stage, context, start(url, context), cancel }`; `onComplete(video)` fires with the finished video-history entry. | The Dev Tools Video Downloader page. |

diff --git a/client/src/hooks/useVideoGenForm.js b/client/src/hooks/useVideoGenForm.js
@@ -53,12 +53,14 @@ const editableRemixModel = (models, defaultModelId) => {
  * that clear now-irrelevant inputs, the derived model/keyframe/IC gates, and
  * `buildGeneratePayload()` — the single client-side source of truth for the
  * shape `server/routes/videoGen.js` validates. `VideoGen.jsx` keeps the
- * fetching (status/models/history/gallery), the SSE run pipeline, the batch
- * queue, and the rendering.
+ * fetching (status/model-context/history/gallery), the SSE run pipeline, the
+ * batch queue, and the rendering.
  *
  * The caller supplies the fetched context the form has to react to:
- *   - `models` / `status` — from `getVideoGenStatus()`; drive the model
- *     dropdown, the default-model seed, and the mode-compatibility fallback.
+ *   - `models` / `modelContext` — from `getVideoGenModelContext()`; drive the
+ *     model dropdown, the default-model seed, and the mode-compatibility
+ *     fallback. Deliberately NOT `getVideoGenStatus()`: that route shells out
+ *     to python, and the picker must not wait on the interpreter probe.
  *   - `availableLoras` — the installed LoRA library, for name resolution.
  *   - `grokEnabled` — the Settings → Image Gen toggle that reveals the
  *     Local/Grok backend switch.
@@ -69,7 +71,7 @@ const editableRemixModel = (models, defaultModelId) => {
  *     wire accepts — kept here rather than in the page so there stays exactly
  *     one builder for what `server/routes/videoGen.js` validates.
  */
-export function useVideoGenForm({ models, status, availableLoras, grokEnabled, remoteSubmissionFields = null }) {
+export function useVideoGenForm({ models, modelContext, availableLoras, grokEnabled, remoteSubmissionFields = null }) {
   const [searchParams, setSearchParams] = useSearchParams();
   const incomingSourceImage = searchParams.get('sourceImageFile');
   const incomingAudioFilename = searchParams.get('audioFilename');
@@ -163,11 +165,11 @@ export function useVideoGenForm({ models, status, availableLoras, grokEnabled, r
       });
     return () => { cancelled = true; };
   }, [incomingAudioFilename, setSearchParams]);
-  // Seed the model dropdown from the server's default once /status lands,
-  // without clobbering a Remix/deep-link/user pick that already set it.
+  // Seed the model dropdown from the server's default once the model context
+  // lands, without clobbering a Remix/deep-link/user pick that already set it.
   useEffect(() => {
-    if (status?.defaultModel) setModelId((prev) => prev || status.defaultModel);
-  }, [status?.defaultModel]);
+    if (modelContext?.defaultModel) setModelId((prev) => prev || modelContext.defaultModel);
+  }, [modelContext?.defaultModel]);
 
   // Re-sync when ImageGen pipes a new image via ?sourceImageFile=...
   useEffect(() => {
@@ -370,7 +372,8 @@ export function useVideoGenForm({ models, status, availableLoras, grokEnabled, r
   //     server would 400 on submit; we proactively swap to a compatible model.
   // a2v fallback preference: highest-memory model that fits this machine
   // (leaving headroom for the OS + text encoder) > the largest if none fit.
-  // Other modes: status.defaultModel (if compatible) > first compatible model.
+  // Other modes: the context's defaultModel (if compatible) > first
+  // compatible model.
   useEffect(() => {
     if (!modelId || models.length === 0) return;
     const current = models.find((m) => m.id === modelId);
@@ -385,40 +388,40 @@ export function useVideoGenForm({ models, status, availableLoras, grokEnabled, r
       // so the user can at least try, and the install banner / OOM surfaces
       // the real constraint instead of a silent dropdown change.
       const reserveGb = 16;
-      // typeof === 'number' (not `status?.systemMemoryGb ? ...`) so a server
-      // legitimately reporting a tiny number (0 GB after rounding on a
-      // sub-GB box) flows through the `fits` check and lands on the
-      // smallest model. The truthiness shortcut would collapse 0 with
-      // "absent" and pick the LARGEST model on a tiny machine.
-      const budget = typeof status?.systemMemoryGb === 'number'
-        ? Math.max(0, status.systemMemoryGb - reserveGb)
+      // typeof === 'number' (not `modelContext?.systemMemoryGb ? ...`) so a
+      // server legitimately reporting a tiny number (0 GB after rounding on a
+      // sub-GB box) flows through the `fits` check and lands on the smallest
+      // model. The truthiness shortcut would collapse 0 with "absent" and pick
+      // the LARGEST model on a tiny machine.
+      const budget = typeof modelContext?.systemMemoryGb === 'number'
+        ? Math.max(0, modelContext.systemMemoryGb - reserveGb)
         : Number.POSITIVE_INFINITY;
       const sortedDesc = [...visibleModels].sort(
         (a, b) => videoModelMemoryGb(b) - videoModelMemoryGb(a),
       );
       const fits = sortedDesc.find((m) => videoModelMemoryGb(m) <= budget);
       fallback = (fits || sortedDesc[sortedDesc.length - 1])?.id || '';
     } else {
-      const defaultModel = models.find((m) => m.id === status?.defaultModel);
+      const defaultModel = models.find((m) => m.id === modelContext?.defaultModel);
       if (defaultModel && isModelAllowedForMode(defaultModel, mode)) {
         fallback = defaultModel.id;
       } else {
-        fallback = visibleModels[0]?.id || status?.defaultModel || models[0]?.id || '';
+        fallback = visibleModels[0]?.id || modelContext?.defaultModel || models[0]?.id || '';
       }
     }
     if (!fallback || fallback === modelId) return;
     // Toast only for the stale-id case (model removed from catalog). The
     // mode-incompatibility swap is expected behavior after a mode change —
     // no need to surface it. Name the destination model so users on a2v
-    // don't think they landed on `status.defaultModel` (they may not have —
+    // don't think they landed on `modelContext.defaultModel` (they may not have —
     // a2v picks the largest-fits model, which is often a dgrauet entry).
     if (!current && staleModelToastRef.current !== modelId) {
       staleModelToastRef.current = modelId;
       const fallbackName = models.find((m) => m.id === fallback)?.name || fallback;
       toast(`Original model "${modelId}" is no longer available — switched to "${fallbackName}"`);
     }
     applyModelSelection(fallback);
-  }, [modelId, models, status?.defaultModel, status?.systemMemoryGb, mode, visibleModels, applyModelSelection]);
+  }, [modelId, models, modelContext?.defaultModel, modelContext?.systemMemoryGb, mode, visibleModels, applyModelSelection]);
 
   const currentModel = models.find((m) => m.id === modelId);
 
@@ -435,17 +438,17 @@ export function useVideoGenForm({ models, status, availableLoras, grokEnabled, r
     setNumFrames(frames);
   }, [audioDurationSec, currentModel, fps, mode]);
 
-  // A source model can reach this hook either through a URL handoff before
-  // /status has populated `models`, or from the in-page gallery after it has.
-  // Resolve both cases here. The fallback is deliberately limited to models
+  // A source model can reach this hook either through a URL handoff before the
+  // model context has populated `models`, or from the in-page gallery after it
+  // has. Resolve both cases here. The fallback is deliberately limited to models
   // that can run a text remix and expose all restored prompt/sampler controls;
   // if no such model is installed we leave the source selected rather than
   // silently changing a faithful re-render.
   useEffect(() => {
     if (!remixSourceModel || models.length === 0) return;
     const source = models.find((model) => model.id === remixSourceModel.id);
     if (source && !remixSourceModel.preserveConditioning && !hasEditableRemixControls(source)) {
-      const target = editableRemixModel(models, status?.defaultModel);
+      const target = editableRemixModel(models, modelContext?.defaultModel);
       if (target) {
         setModelId(target.id);
         setRemixModelFallback({
@@ -459,7 +462,7 @@ export function useVideoGenForm({ models, status, availableLoras, grokEnabled, r
       setRemixModelFallback(null);
     }
     setRemixSourceModel(null);
-  }, [remixSourceModel, models, status?.defaultModel]);
+  }, [remixSourceModel, models, modelContext?.defaultModel]);
 
   // Until the user deliberately chooses a size, model changes carry their own
   // native default canvas. This is material for H3: the shared 768x512 default
@@ -712,7 +715,7 @@ export function useVideoGenForm({ models, status, availableLoras, grokEnabled, r
     keyframesActive,
     mode,
     numFrames,
-    pixelBudget: status?.fflfLtx2PixelBudget,
+    pixelBudget: modelContext?.fflfLtx2PixelBudget,
     sourceImageFile,
     sourceImageUpload,
     width,

diff --git a/client/src/hooks/useVideoGenForm.test.jsx b/client/src/hooks/useVideoGenForm.test.jsx
@@ -67,13 +67,13 @@ const H3_REF2VA = {
   defaultFrames: 124,
 };
 const MODELS = [MLX, LTX2];
-const STATUS = { connected: true, defaultModel: MLX.id };
+const MODEL_CONTEXT = { defaultModel: MLX.id };
 
-const render = ({ models = MODELS, status = STATUS, availableLoras = [], grokEnabled = false, url = '/media/video' } = {}) => {
+const render = ({ models = MODELS, modelContext = MODEL_CONTEXT, availableLoras = [], grokEnabled = false, url = '/media/video' } = {}) => {
   const wrapper = ({ children }) => <MemoryRouter initialEntries={[url]}>{children}</MemoryRouter>;
   return renderHook(
     (props) => useVideoGenForm(props),
-    { wrapper, initialProps: { models, status, availableLoras, grokEnabled } },
+    { wrapper, initialProps: { models, modelContext, availableLoras, grokEnabled } },
   );
 };
 
@@ -94,7 +94,7 @@ describe('useVideoGenForm', () => {
     vi.unstubAllGlobals();
   });
 
-  it('seeds the model from status.defaultModel without clobbering a URL pick', async () => {
+  it('seeds the model from modelContext.defaultModel without clobbering a URL pick', async () => {
     const { result } = render();
     await waitFor(() => expect(result.current.modelId).toBe(MLX.id));
 
@@ -324,7 +324,7 @@ describe('useVideoGenForm', () => {
   it('clears sampler overrides on an automatic mode-compatible model fallback', async () => {
     const { result } = render({
       models: [WAN_T2V, WAN_TI2V],
-      status: { connected: true, defaultModel: WAN_T2V.id },
+      modelContext: { defaultModel: WAN_T2V.id },
     });
     await waitFor(() => expect(result.current.modelId).toBe(WAN_T2V.id));
     act(() => {
@@ -340,7 +340,7 @@ describe('useVideoGenForm', () => {
   it('does not submit chunks for a T2V-only Wan profile', async () => {
     const { result } = render({
       models: [WAN_T2V],
-      status: { connected: true, defaultModel: WAN_T2V.id },
+      modelContext: { defaultModel: WAN_T2V.id },
     });
     await waitFor(() => expect(result.current.modelId).toBe(WAN_T2V.id));
     act(() => {
@@ -353,7 +353,7 @@ describe('useVideoGenForm', () => {
   it('normalizes MiniMax H3 to its fixed temporal and sampler contract', async () => {
     const { result } = render({
       models: [MLX, H3],
-      status: { connected: true, defaultModel: MLX.id },
+      modelContext: { defaultModel: MLX.id },
     });
     await waitFor(() => expect(result.current.modelId).toBe(MLX.id));
     act(() => {
@@ -388,7 +388,7 @@ describe('useVideoGenForm', () => {
   // Substitutable prompt conditioner (#4081).
   describe('text encoder selection', () => {
     const renderWithH3 = async () => {
-      const rendered = render({ models: [MLX, H3], status: { connected: true, defaultModel: MLX.id } });
+      const rendered = render({ models: [MLX, H3], modelContext: { defaultModel: MLX.id } });
       await waitFor(() => expect(rendered.result.current.modelId).toBe(MLX.id));
       act(() => rendered.result.current.handleModelChange(H3.id));
       await waitFor(() => expect(rendered.result.current.modelId).toBe(H3.id));
@@ -398,7 +398,7 @@ describe('useVideoGenForm', () => {
     // An empty list is what hides the picker; a model with substitutions
     // exposes them straight off the server-decorated entry.
     it('exposes only the selected model’s options', async () => {
-      const { result } = render({ models: [MLX, H3], status: { connected: true, defaultModel: MLX.id } });
+      const { result } = render({ models: [MLX, H3], modelContext: { defaultModel: MLX.id } });
       await waitFor(() => expect(result.current.modelId).toBe(MLX.id));
       expect(result.current.textEncoderOptions).toEqual([]);
       act(() => result.current.handleModelChange(H3.id));
@@ -465,7 +465,7 @@ describe('useVideoGenForm', () => {
   it('preserves H3 native 32px-grid geometry in the submitted payload', async () => {
     const { result } = render({
       models: [MLX, H3],
-      status: { connected: true, defaultModel: MLX.id },
+      modelContext: { defaultModel: MLX.id },
     });
     await waitFor(() => expect(result.current.modelId).toBe(MLX.id));
     act(() => result.current.handleModelChange(H3.id));
@@ -482,7 +482,7 @@ describe('useVideoGenForm', () => {
   it('offers MiniMax H3 image mode, chaining and a non-advisory last frame', async () => {
     const { result } = render({
       models: [MLX, H3],
-      status: { connected: true, defaultModel: MLX.id },
+      modelContext: { defaultModel: MLX.id },
     });
     await waitFor(() => expect(result.current.modelId).toBe(MLX.id));
     act(() => {
@@ -791,7 +791,7 @@ describe('useVideoGenForm', () => {
   it('moves a fixed-profile remix to an editable model while preserving its restored controls', async () => {
     const { result } = render({
       models: [MLX, H3],
-      status: { connected: true, defaultModel: MLX.id },
+      modelContext: { defaultModel: MLX.id },
     });
     await waitFor(() => expect(result.current.modelId).toBe(MLX.id));
 
@@ -818,7 +818,7 @@ describe('useVideoGenForm', () => {
   it('uses the same editable-model fallback for a cross-page Remix handoff', async () => {
     const { result } = render({
       models: [MLX, H3],
-      status: { connected: true, defaultModel: MLX.id },
+      modelContext: { defaultModel: MLX.id },
       url: `/media/video?modelId=${H3.id}&numFrames=124&steps=9&guidanceScale=0`,
     });
 
@@ -938,10 +938,10 @@ describe('useVideoGenForm — i2v reference mode (#4874)', () => {
     lastFrameAnchored: true, supportedModes: RUNTIME_MODES,
   };
   const LTX25_MODELS = [LTX25, LTX2];
-  const LTX25_STATUS = { connected: true, defaultModel: LTX25.id };
+  const LTX25_MODEL_CONTEXT = { defaultModel: LTX25.id };
 
   const inImageMode = async (opts = {}) => {
-    const { result } = render({ models: LTX25_MODELS, status: LTX25_STATUS, ...opts });
+    const { result } = render({ models: LTX25_MODELS, modelContext: LTX25_MODEL_CONTEXT, ...opts });
     await act(async () => { result.current.handleModeChange('image'); });
     return result;
   };
@@ -982,15 +982,15 @@ describe('useVideoGenForm — i2v reference mode (#4874)', () => {
       (props) => useVideoGenForm(props),
       {
         wrapper: ({ children }) => <MemoryRouter initialEntries={['/media/video']}>{children}</MemoryRouter>,
-        initialProps: { models: [], status: LTX25_STATUS, availableLoras: [], grokEnabled: false },
+        initialProps: { models: [], modelContext: LTX25_MODEL_CONTEXT, availableLoras: [], grokEnabled: false },
       },
     );
     await act(async () => {
       result.current.applyResumedParams({ modelId: LTX25.id, mode: 'image', i2vReferenceMode: 'inspire' });
     });
     expect(result.current.i2vReferenceMode).toBe('inspire');
 
-    rerender({ models: LTX25_MODELS, status: LTX25_STATUS, availableLoras: [], grokEnabled: false });
+    rerender({ models: LTX25_MODELS, modelContext: LTX25_MODEL_CONTEXT, availableLoras: [], grokEnabled: false });
     await waitFor(() => expect(result.current.currentModel?.id).toBe(LTX25.id));
     expect(result.current.i2vReferenceMode).toBe('inspire');
   });