From d2809855539bb3fb6841297fe901c91c1d86e8de Mon Sep 17 00:00:00 2001
From: Towsty <ianjmcarr@gmail.com>
Date: Wed, 26 Aug 2026 23:08:43 -0500
Subject: [PATCH] Send V2 identity prompts to Reference to Video instead of
 Image to Video.

Co-authored-by: Cursor <cursoragent@cursor.com>
---
 pages/index.vue          | 22 ++++++++++++++--------
 server/utils/comfy.ts    | 10 ++++++----
 server/utils/workflow.ts | 18 +++++++++++++-----
 3 files changed, 33 insertions(+), 17 deletions(-)

diff --git a/pages/index.vue b/pages/index.vue
index 2f2f0c4..1747b58 100644
--- a/pages/index.vue
+++ b/pages/index.vue
@@ -109,10 +109,10 @@
           </div>
           <p v-if="shotScriptMode && studioMode === 'video'" class="mb-2 text-xs text-zinc-500">
             Paste a block. Text before the first marker is shot 1. Put <span class="font-medium text-zinc-300">shot 2</span>, <span class="font-medium text-zinc-300">shot 3</span> on their own lines to queue extensions — a line like <span class="font-medium text-zinc-300">Shot Type</span> will not start a new shot. Every shot uses the duration slider.
-            <span v-if="identityPrompting"> Identity boilerplate is prepended to the first shot only.</span>
+            <span v-if="identityPrompting"> Identity boilerplate is prepended to the first shot only, then concatenated into Reference to Video.</span>
           </p>
           <p v-else-if="identityPrompting" class="mb-2 text-xs text-zinc-500">
-            Write camera, action, and audio. We prepend the <span class="font-medium text-zinc-300">&lt;Picture 1&gt;</span> identity lock automatically.
+            Write camera, action, and audio. That is the dynamic action prompt. We concatenate the <span class="font-medium text-zinc-300">&lt;Picture 1&gt;</span> identity prefix in front of it and send the result to <span class="font-medium text-zinc-300">Reference to Video</span>, not Image to Video.
           </p>
           <textarea
             v-model="prompt"
@@ -121,8 +121,13 @@
             :placeholder="promptPlaceholder"
           />
           <details v-if="identityPrompting && prompt.trim()" class="mt-2 rounded-xl border border-white/10 bg-zinc-950/40 px-3 py-2">
-            <summary class="cursor-pointer text-xs text-zinc-400">Prompt sent to MiniMax</summary>
-            <p class="mt-2 whitespace-pre-wrap text-xs text-zinc-300">{{ composedIdentityPrompt }}</p>
+            <summary class="cursor-pointer text-xs text-zinc-400">Prompt sent to Reference to Video</summary>
+            <p class="mt-2 text-[10px] uppercase tracking-wide text-zinc-500">Identity prefix</p>
+            <p class="mt-1 whitespace-pre-wrap text-xs text-zinc-400">{{ identityBoilerplate(extraIdentityPictures) }}</p>
+            <p class="mt-3 text-[10px] uppercase tracking-wide text-zinc-500">Dynamic action prompt</p>
+            <p class="mt-1 whitespace-pre-wrap text-xs text-zinc-300">{{ identityActionPrompt }}</p>
+            <p class="mt-3 text-[10px] uppercase tracking-wide text-zinc-500">Concatenated result</p>
+            <p class="mt-1 whitespace-pre-wrap text-xs text-zinc-300">{{ composedIdentityPrompt }}</p>
           </details>
           <div v-if="shotScriptMode && studioMode === 'video'" class="mt-3 space-y-2">
             <p v-if="!parsedShots.length" class="text-xs text-zinc-500">Add a first shot, then <span class="text-zinc-300">shot 2</span> for an extension.</p>
@@ -1296,11 +1301,12 @@ const generateLabel = computed(() => {
 const parsedShots = computed(() => parseShotScript(prompt.value))
 const identityPrompting = computed(() => studioMode.value === 'video' && videoWorkflow.value === 'v2' && useIdentityRefs.value)
 const extraIdentityPictures = computed(() => identityRefs.value.flatMap((item, index) => item ? [index + 2] : []))
+const identityActionPrompt = computed(() => {
+  if (shotScriptMode.value) return parsedShots.value[0]?.prompt || prompt.value
+  return prompt.value
+})
 const composedIdentityPrompt = computed(() => {
-  const action = shotScriptMode.value
-    ? (parsedShots.value[0]?.prompt || prompt.value)
-    : prompt.value
-  return buildIdentityPrompt(action, extraIdentityPictures.value)
+  return buildIdentityPrompt(identityActionPrompt.value, extraIdentityPictures.value)
 })
 const promptPlaceholder = computed(() => {
   if (studioMode.value === 'edit') {
diff --git a/server/utils/comfy.ts b/server/utils/comfy.ts
index 15a5ef0..2c3c1cf 100644
--- a/server/utils/comfy.ts
+++ b/server/utils/comfy.ts
@@ -163,12 +163,14 @@ export function extractPromptFromHistory(entry: unknown) {
   const prompt = (entry as { prompt?: unknown[] })?.prompt
   const graph = Array.isArray(prompt) ? prompt[2] : null
   if (!graph || typeof graph !== 'object') return ''
+  let imageToVideo = ''
   for (const node of Object.values(graph as Record<string, { class_type?: string; inputs?: Record<string, unknown> }>)) {
-    if (node?.class_type === 'MiniMaxH3ImageToVideo') {
-      return String(node.inputs?.prompt || node.inputs?.text || '').trim()
-    }
+    const text = String(node.inputs?.prompt || node.inputs?.text || '').trim()
+    if (!text) continue
+    if (node?.class_type === 'MiniMaxH3ReferenceToVideo') return text
+    if (node?.class_type === 'MiniMaxH3ImageToVideo') imageToVideo = text
   }
-  return ''
+  return imageToVideo
 }
 
 export function extractClipMetaFromHistory(entry: unknown) {
diff --git a/server/utils/workflow.ts b/server/utils/workflow.ts
index 0b43345..37e0d87 100644
--- a/server/utils/workflow.ts
+++ b/server/utils/workflow.ts
@@ -73,7 +73,8 @@ export function buildWorkflow(params: GenerateParams) {
   const refs = Array.from({ length: 4 }, (_, index) => String(params.referenceImageNames?.[index] || '').trim())
   const extraPictures = refs.flatMap((name, index) => name ? [index + 2] : [])
   const useRefs = version === 'v2' && params.useIdentityRefs === true
-  const graphPrompt = useRefs ? buildIdentityPrompt(params.prompt, extraPictures) : params.prompt
+  const actionPrompt = params.prompt
+  const graphPrompt = useRefs ? buildIdentityPrompt(actionPrompt, extraPictures) : actionPrompt
 
   for (const [id, node] of Object.entries(graph)) {
     if (IMAGE_CLASSES.has(node.class_type) && 'image' in node.inputs && id === '114') {
@@ -81,19 +82,20 @@ export function buildWorkflow(params: GenerateParams) {
     }
 
     if (PROMPT_CLASSES.has(node.class_type) || node.class_type.includes('Qwen3')) {
-      if ('prompt' in node.inputs) node.inputs.prompt = graphPrompt
-      if ('text' in node.inputs) node.inputs.text = graphPrompt
+      const nodePrompt = promptForNode(node.class_type, useRefs, actionPrompt, graphPrompt)
+      if ('prompt' in node.inputs) node.inputs.prompt = nodePrompt
+      if ('text' in node.inputs) node.inputs.text = nodePrompt
     }
 
     if (node.class_type === 'MiniMaxH3ImageToVideo') {
-      node.inputs.prompt = graphPrompt
+      node.inputs.prompt = useRefs ? '' : actionPrompt
       if (typeof node.inputs.width === 'number') node.inputs.width = width
       if (typeof node.inputs.height === 'number') node.inputs.height = height
       if (typeof node.inputs.length === 'number') node.inputs.length = params.length
     }
 
     if (node.class_type === 'MiniMaxH3ReferenceToVideo') {
-      node.inputs.prompt = graphPrompt
+      node.inputs.prompt = useRefs ? graphPrompt : actionPrompt
       if (!Array.isArray(node.inputs.length)) node.inputs.length = ['174', 1]
     }
 
@@ -206,6 +208,12 @@ export function buildWorkflow(params: GenerateParams) {
   return graph
 }
 
+function promptForNode(classType: string, useRefs: boolean, actionPrompt: string, graphPrompt: string) {
+  if (classType === 'MiniMaxH3ImageToVideo') return useRefs ? '' : actionPrompt
+  if (classType === 'MiniMaxH3ReferenceToVideo') return useRefs ? graphPrompt : actionPrompt
+  return graphPrompt
+}
+
 function applyV2IdentityPath(
   graph: WorkflowGraph,
   params: GenerateParams,