diff --git a/.gitignore b/.gitignore
index a89d7e9..6c989a6 100644
--- a/.gitignore
+++ b/.gitignore
@@ -60,3 +60,6 @@ __pycache__/
 Thumbs.db
 .idea/
 .vscode/
+
+# fast-deploy build artifact
+/frontend/dist.tar.gz
diff --git a/frontend/src/components/ModelFormModal.vue b/frontend/src/components/ModelFormModal.vue
index c9e18ac..b689058 100644
--- a/frontend/src/components/ModelFormModal.vue
+++ b/frontend/src/components/ModelFormModal.vue
@@ -241,8 +241,8 @@ async function save() {
                   <span v-for="r in resolutions" :key="r" class="ro-chip">{{ r }}</span>
                   <span v-if="!resolutions.length" class="text-white/30">—</span>
                 </div>
-                <span class="text-white/40">图生图</span>
-                <div class="text-white/70">{{ entry.image_to_image ? '支持' : '不支持' }}</div>
+                <span class="text-white/40">参考图</span>
+                <div class="text-white/70">{{ entry.max_reference_images > 0 ? `${entry.max_reference_images} 张` : (entry.image_to_image ? '支持' : '不支持') }}</div>
               </template>
             </div>
           </div>
diff --git a/frontend/src/components/TestModal.vue b/frontend/src/components/TestModal.vue
index 522d089..7d61730 100644
--- a/frontend/src/components/TestModal.vue
+++ b/frontend/src/components/TestModal.vue
@@ -56,8 +56,8 @@ const refMode = computed(() => familyPreset.value?.reference_mode || props.model
 // text2video; frame refs (首帧/末帧) only enhance the result when supplied.
 const refsRequired = computed(() => false)
 const refsLabel = computed(() => {
-  if (!isVideo) return maxRefs.value > 0 ? '参考图 (可选, 图生图)' : ''
-  if (refMode.value === 'asset') return `参考图 (asset 模式, 最多 ${maxRefs.value} 张)`
+  if (!isVideo) return maxRefs.value > 0 ? '参考图 (可选)' : ''
+  if (refMode.value === 'asset') return `参考图 (最多 ${maxRefs.value} 张)`
   if (refMode.value === 'frame') {
     if (maxRefs.value >= 2) return `首帧 / 末帧 (1=首帧, 2=首尾帧, 最多 ${maxRefs.value} 张)`
     return `首帧 (可选, ${maxRefs.value} 张)`
diff --git a/frontend/src/views/ModelsView.vue b/frontend/src/views/ModelsView.vue
index 2948f88..cae4de6 100644
--- a/frontend/src/views/ModelsView.vue
+++ b/frontend/src/views/ModelsView.vue
@@ -211,12 +211,12 @@ onMounted(loadModels)
                  video frame mode, supported resolutions for video. -->
             <td class="px-3 py-3.5 align-middle">
               <div class="flex flex-wrap items-center gap-1 text-[11px]">
-                <span v-if="m.type === 'image' && m.image_to_image"
-                      class="cap-chip cap-emerald">图生图</span>
-                <!-- 参考图 is a capability like 图生图 — same emerald style, shown first -->
-                <span v-if="m.type === 'video' && m.max_reference_images > 0"
+                <!-- reference capability with count: frame mode = 首尾帧, else 参考图 -->
+                <span v-if="m.max_reference_images > 0"
                       class="cap-chip cap-emerald"
-                      :title="REF_MODE_LABEL[m.reference_mode]">参考图 {{ m.max_reference_images }}</span>
+                      :title="REF_MODE_LABEL[m.reference_mode]">{{ m.reference_mode === 'frame' ? '首尾帧' : '参考图' }} {{ m.max_reference_images }}</span>
+                <span v-else-if="m.type === 'image' && m.image_to_image"
+                      class="cap-chip cap-emerald">参考图</span>
                 <span v-if="m.type === 'video'" v-for="r in (m.resolutions || [])" :key="'vr'+r"
                       class="cap-chip cap-slate">{{ r }}</span>
                 <span v-if="(m.type === 'image' && !(m.ratios || []).length && !m.image_to_image) ||