diff --git a/open-sse/handlers/chatCore.js b/open-sse/handlers/chatCore.js
index cb6ce961..88356471 100644
--- a/open-sse/handlers/chatCore.js
+++ b/open-sse/handlers/chatCore.js
@@ -57,7 +57,7 @@ export function stripContinuityFields(body) {
   return body;
 }
 
-export async function handleChatCore({ body, modelInfo, credentials, log, onCredentialsRefreshed, onRequestSuccess, onDisconnect, clientRawRequest, connectionId, userAgent, apiKey, ccFilterNaming, rtkEnabled, headroomEnabled, headroomUrl, headroomCompressUserMessages, cavemanEnabled, cavemanLevel, ponytailEnabled, ponytailLevel, pxpipeEnabled, pxpipeMinChars, pxpipeTimeoutMs, pxpipeTransform, onPxpipeEvent, sourceFormatOverride, providerThinking }) {
+export async function handleChatCore({ body, modelInfo, credentials, log, onCredentialsRefreshed, onRequestSuccess, onDisconnect, clientRawRequest, connectionId, userAgent, apiKey, ccFilterNaming, rtkEnabled, headroomEnabled, headroomUrl, headroomCompressUserMessages, cavemanEnabled, cavemanLevel, ponytailEnabled, ponytailLevel, pxpipeEnabled, pxpipeMinChars, pxpipeTimeoutMs, pxpipeTransform, onPxpipeEvent, sourceFormatOverride, providerThinking, capsOverride }) {
   const { provider, model } = modelInfo;
   const requestStartTime = Date.now();
   // Stable per-session color so all lines of one CLI conversation share a tag
@@ -149,8 +149,10 @@ export async function handleChatCore({ body, modelInfo, credentials, log, onCred
   if (credentials) credentials.rawHeaders = clientRawRequest?.headers || {};
 
   // Auto-strip media blocks the model can't read (vision/audio/pdf) before translation.
+  // capsOverride lets the app layer assert per-model capabilities (e.g. user-registered
+  // models) on top of the static tables.
   if (!passthrough) {
-    const caps = getCapabilitiesForModel(provider, model);
+    const caps = { ...getCapabilitiesForModel(provider, model), ...(capsOverride || {}) };
     if (stripUnsupportedModalities(body, sourceFormat, caps)) {
       log?.debug?.("MODALITY", `stripped unsupported media for ${provider}/${model}`);
     }
diff --git a/open-sse/providers/capabilities.js b/open-sse/providers/capabilities.js
index 24a03b04..024eda65 100644
--- a/open-sse/providers/capabilities.js
+++ b/open-sse/providers/capabilities.js
@@ -317,6 +317,11 @@ export const PATTERN_CAPABILITIES = [
   { pattern: "*ling-*",         caps: { reasoning: true, contextWindow: 128000 } },
 ];
 
+// OpenRouter-style gateways validate modalities upstream — a text-only model
+// sent an image gets a clear upstream error instead of silent corruption. So for
+// unknown models on these providers, trust vision instead of stripping images.
+const TRUST_UPSTREAM_VISION = new Set(["openrouter"]);
+
 /**
  * Resolve capabilities for a model using the 4-step fallback chain,
  * merged over DEFAULT_CAPABILITIES so the result is always complete.
@@ -349,6 +354,9 @@ export function getCapabilitiesForModel(provider, model) {
     }
   }
 
-  // 4. Floor
+  // 4. Floor (upstream-validated gateways keep vision on for unknown models)
+  if (provider && TRUST_UPSTREAM_VISION.has(provider)) {
+    return { ...DEFAULT_CAPABILITIES, vision: true };
+  }
   return { ...DEFAULT_CAPABILITIES };
 }
diff --git a/src/sse/handlers/chat.js b/src/sse/handlers/chat.js
index 88353c50..8b7a89dd 100644
--- a/src/sse/handlers/chat.js
+++ b/src/sse/handlers/chat.js
@@ -7,7 +7,8 @@ import {
   extractApiKey,
   isValidApiKey,
 } from "../services/auth.js";
-import { getSettings } from "@/lib/localDb";
+import { getSettings, getCustomModels } from "@/lib/localDb";
+import { capabilitiesFromServiceKind } from "open-sse/providers/capabilities.js";
 import { getModelInfo, getComboModels } from "../services/model.js";
 import { handleChatCore } from "open-sse/handlers/chatCore.js";
 import { DEFAULT_HEADROOM_URL } from "@/lib/headroom/detect";
@@ -259,6 +260,13 @@ async function handleSingleModelChat(body, modelStr, clientRawRequest = null, re
     // Use shared chatCore
     const chatSettings = await getSettings();
     const providerThinking = (chatSettings.providerThinking || {})[provider] || null;
+    // User-registered models (Provider page → "Add Model") trust upstream: keep
+    // images/media in the request instead of stripping by static-capability guess.
+    let capsOverride = null;
+    try {
+      const cm = (await getCustomModels()).find((m) => m.providerAlias === provider && m.id === model);
+      if (cm) capsOverride = { ...capabilitiesFromServiceKind(cm.type), vision: true };
+    } catch { /* fail-open to static caps */ }
     const result = await handleChatCore({
       body: { ...body, model: `${provider}/${model}` },
       modelInfo: { provider, model },
@@ -284,6 +292,7 @@ async function handleSingleModelChat(body, modelStr, clientRawRequest = null, re
       pxpipeTransform: chatSettings.pxpipeEnabled ? await getPxpipeTransform() : null,
       onPxpipeEvent: appendPxpipeEvent,
       providerThinking,
+      capsOverride,
       // Detect source format by endpoint + body
       sourceFormatOverride: request?.url ? detectFormatByEndpoint(new URL(request.url).pathname, body) : null,
       onCredentialsRefreshed: async (newCreds) => {