diff --git a/src/tools/imagegen.ts b/src/tools/imagegen.ts
index 99efb08..4aabb5a 100644
--- a/src/tools/imagegen.ts
+++ b/src/tools/imagegen.ts
@@ -21,6 +21,8 @@ import type { ContentLibrary } from '../content/library.js';
 import { checkImageBudget, recordImageAsset } from '../content/record-image.js';
 import { ModelClient } from '../agent/llm.js';
 import { analyzeMediaRequest, renderProposalForAskUser } from '../agent/media-router.js';
+import { recordUsage } from '../stats/tracker.js';
+import { findModel, estimateCostUsd } from '../gateway-models.js';
 
 interface ImageGenInput {
   prompt: string;
@@ -220,6 +222,20 @@ function buildExecute(deps: ImageGenDeps) {
     const sizeKB = (fileSize / 1024).toFixed(1);
     const revisedPrompt = imageData.revised_prompt ? `\nRevised prompt: ${imageData.revised_prompt}` : '';
 
+    // Stats: record this generation so it shows up in `franklin insights`
+    // alongside chat spend. Before this, media generations bypassed
+    // recordUsage entirely (only LLM chat calls were tracked), so the
+    // insights panel under-reported total spend and never surfaced
+    // image-generation models in its "top models" list. Fire-and-forget —
+    // stats write must not fail a user-visible generation.
+    void (async () => {
+      try {
+        const m = await findModel(imageModel);
+        const estCost = m ? estimateCostUsd(m, { quantity: 1 }) : 0;
+        recordUsage(imageModel, 0, 0, estCost, 0);
+      } catch { /* ignore stats errors */ }
+    })();
+
     let contentSummary = '';
     if (contentId && deps.library) {
       const rec = recordImageAsset(deps.library, {
diff --git a/src/tools/videogen.ts b/src/tools/videogen.ts
index 2fba8b7..d6c22ce 100644
--- a/src/tools/videogen.ts
+++ b/src/tools/videogen.ts
@@ -36,6 +36,8 @@ import { loadChain, API_URLS, VERSION } from '../config.js';
 import type { ContentLibrary } from '../content/library.js';
 import { ModelClient } from '../agent/llm.js';
 import { analyzeMediaRequest, renderProposalForAskUser } from '../agent/media-router.js';
+import { recordUsage } from '../stats/tracker.js';
+import { findModel, estimateCostUsd } from '../gateway-models.js';
 
 interface VideoGenInput {
   prompt: string;
@@ -290,6 +292,21 @@ function buildExecute(deps: VideoGenDeps) {
       const sizeMB = (fileSize / 1_048_576).toFixed(1);
       const dur = videoData.duration_seconds ?? duration;
 
+      // Stats: record this generation so it shows up in `franklin insights`
+      // alongside chat spend. Before this, media generations bypassed
+      // recordUsage entirely, so the insights panel under-reported total
+      // spend and never surfaced video models in its "top models" list.
+      // Prefer the live gateway price when the model is in the catalog;
+      // fall back to the legacy $0.05/s estimate otherwise. Fire-and-
+      // forget — stats write must not fail a user-visible generation.
+      void (async () => {
+        try {
+          const m = await findModel(videoModel);
+          const estCost = m ? estimateCostUsd(m, { duration_seconds: dur }) : estimateVideoCostUsd(dur);
+          recordUsage(videoModel, 0, 0, estCost, 0);
+        } catch { /* ignore stats errors */ }
+      })();
+
       let contentSummary = '';
       if (contentId && deps.library) {
         const rec = deps.library.addAsset(contentId, {