Add MiniMax and PinkCherry LTX as pickable video engines, with still or text start.

Shot 1 can run text-to-video; later shots still extend from the last frame. Recommend output is also normalized into shot-script cards.

Co-authored-by: Cursor <cursoragent@cursor.com>

pages/index.vue

@@ -6,7 +6,7 @@
           <div class="flex h-10 w-10 items-center justify-center rounded-xl bg-amber-400 text-zinc-950 font-display font-extrabold">A</div>
           <div>
             <h1 class="font-display text-xl font-bold leading-none">{{ instanceName }}</h1>
-            <p class="text-[11px] uppercase tracking-[0.22em] text-zinc-500">MiniMax H3 · Flux.2 Klein</p>
+            <p class="text-[11px] uppercase tracking-[0.22em] text-zinc-500">MiniMax H3 · LTX-2.3 · Flux.2 Klein</p>
           </div>
         </div>
         <div class="flex items-center gap-3 text-sm">
@@ -80,8 +80,8 @@
             <button type="button" class="relative mt-3 rounded-full bg-zinc-950/80 px-3 py-1 text-xs" @click.stop="resetImage">Reset</button>
           </div>
           <div v-else class="flex min-h-44 flex-col items-center justify-center text-center">
-            <p class="font-medium">{{ studioMode === 'edit' ? 'Image 1' : 'Choose an image' }}</p>
+            <p class="font-medium">{{ studioMode === 'edit' ? 'Image 1' : (textToVideo ? 'Start still · optional' : 'Choose an image') }}</p>
-            <p class="mt-1 text-sm text-zinc-500">Browse the library or upload a new PNG, JPG, or WEBP</p>
+            <p class="mt-1 text-sm text-zinc-500">{{ textToVideo ? 'Text-to-video does not need a still. Shot 2+ will still extend from the last frame.' : 'Browse the library or upload a new PNG, JPG, or WEBP' }}</p>
           </div>
         </div>
 
@@ -183,6 +183,16 @@
             <p class="text-xs font-medium uppercase tracking-wide text-amber-200/80">Recommended prompt</p>
             <p v-if="recommendBusy" class="mt-2 text-sm text-zinc-400">{{ recommendMessage || 'Sidecar Qwen VL is reading the still…' }}</p>
             <p v-else-if="recommendError" class="mt-2 text-sm text-red-300">{{ recommendError }}</p>
+            <div v-else-if="recommendedShots.length > 1" class="mt-2 space-y-2">
+              <div
+                v-for="shot in recommendedShots"
+                :key="shot.n"
+                class="rounded-xl border border-amber-300/15 bg-zinc-950/50 px-3 py-2"
+              >
+                <p class="text-[11px] font-medium uppercase tracking-wide text-amber-200/80">shot {{ shot.n }}</p>
+                <p class="mt-1 whitespace-pre-wrap text-sm text-zinc-200">{{ shot.prompt }}</p>
+              </div>
+            </div>
             <p v-else class="mt-2 whitespace-pre-wrap text-sm text-zinc-200">{{ recommendText }}</p>
             <div v-if="recommendText && !recommendBusy" class="mt-3 flex flex-wrap gap-2">
               <button
@@ -343,7 +353,7 @@
                 <span class="absolute left-0.5 top-0.5 h-5 w-5 rounded-full bg-white transition peer-checked:translate-x-5" />
               </span>
             </label>
-            <label v-if="studioMode === 'video'" class="flex cursor-pointer items-center justify-between gap-3 rounded-2xl border border-white/10 bg-zinc-950/40 px-4 py-3 text-sm">
+            <label v-if="studioMode === 'video' && !ltxVideo" class="flex cursor-pointer items-center justify-between gap-3 rounded-2xl border border-white/10 bg-zinc-950/40 px-4 py-3 text-sm">
               <span>
                 <span class="block font-medium text-zinc-200">Generate sound</span>
                 <span class="mt-0.5 block text-xs text-zinc-500">Off skips the audio VAE so Comfy encodes a silent clip.</span>
@@ -371,13 +381,60 @@
 
         <div v-if="studioMode === 'video'" class="space-y-3">
           <div>
+            <p class="mb-2 text-xs uppercase tracking-wider text-zinc-500">Video model</p>
+            <div class="grid grid-cols-2 gap-2">
+              <button
+                type="button"
+                class="rounded-xl border px-3 py-2 text-left text-sm"
+                :class="videoEngine === 'minimax' ? 'border-amber-300 bg-amber-400/10 text-amber-100' : 'border-white/10 hover:border-white/20'"
+                @click="selectVideoEngine('minimax')"
+              >
+                <span class="block font-semibold">MiniMax H3</span>
+                <span class="text-xs text-zinc-400">Native stereo audio. Identity refs on v2.</span>
+              </button>
+              <button
+                type="button"
+                class="rounded-xl border px-3 py-2 text-left text-sm"
+                :class="videoEngine === 'ltx' ? 'border-amber-300 bg-amber-400/10 text-amber-100' : 'border-white/10 hover:border-white/20'"
+                @click="selectVideoEngine('ltx')"
+              >
+                <span class="block font-semibold">LTX-2.3 PinkCherry</span>
+                <span class="text-xs text-zinc-400">PinkCherry + distilled LoRA. I2V and T2V.</span>
+              </button>
+            </div>
+          </div>
+          <div>
+            <p class="mb-2 text-xs uppercase tracking-wider text-zinc-500">Start from</p>
+            <div class="grid grid-cols-2 gap-2">
+              <button
+                type="button"
+                class="rounded-xl border px-3 py-2 text-left text-sm"
+                :class="videoStart === 'still' ? 'border-amber-300 bg-amber-400/10 text-amber-100' : 'border-white/10 hover:border-white/20'"
+                @click="videoStart = 'still'"
+              >
+                <span class="block font-semibold">A still</span>
+                <span class="text-xs text-zinc-400">Image to video. Extensions use the last frame.</span>
+              </button>
+              <button
+                type="button"
+                class="rounded-xl border px-3 py-2 text-left text-sm"
+                :class="videoStart === 'text' ? 'border-amber-300 bg-amber-400/10 text-amber-100' : 'border-white/10 hover:border-white/20'"
+                @click="videoStart = 'text'"
+              >
+                <span class="block font-semibold">Text only</span>
+                <span class="text-xs text-zinc-400">No start still. Shot 2+ still continues from last frame.</span>
+              </button>
+            </div>
+            <p class="mt-2 text-xs text-zinc-500">{{ videoModelHint }}</p>
+          </div>
+          <div v-if="videoEngine === 'minimax' && videoStart === 'still'">
             <p class="mb-2 text-xs uppercase tracking-wider text-zinc-500">MiniMax graph</p>
             <div class="grid grid-cols-2 gap-2">
               <button
                 type="button"
                 class="rounded-xl border px-3 py-2 text-left text-sm"
-                :class="videoWorkflow === 'v1' ? 'border-amber-300 bg-amber-400/10 text-amber-100' : 'border-white/10 hover:border-white/20'"
+                :class="minimaxGraph === 'v1' ? 'border-amber-300 bg-amber-400/10 text-amber-100' : 'border-white/10 hover:border-white/20'"
-                @click="videoWorkflow = 'v1'"
+                @click="minimaxGraph = 'v1'"
               >
                 <span class="block font-semibold">v1 · validated</span>
                 <span class="text-xs text-zinc-400">First-frame image to video</span>
@@ -385,8 +442,8 @@
               <button
                 type="button"
                 class="rounded-xl border px-3 py-2 text-left text-sm"
-                :class="videoWorkflow === 'v2' ? 'border-amber-300 bg-amber-400/10 text-amber-100' : 'border-white/10 hover:border-white/20'"
+                :class="minimaxGraph === 'v2' ? 'border-amber-300 bg-amber-400/10 text-amber-100' : 'border-white/10 hover:border-white/20'"
-                @click="videoWorkflow = 'v2'"
+                @click="minimaxGraph = 'v2'"
               >
                 <span class="block font-semibold">v2 · identity refs</span>
                 <span class="text-xs text-zinc-400">Optional lock with up to 4 stills</span>
@@ -394,7 +451,7 @@
             </div>
           </div>
 
-          <div v-if="videoWorkflow === 'v2'" class="rounded-2xl border border-white/10 bg-zinc-950/40 px-4 py-3 space-y-3">
+          <div v-if="videoEngine === 'minimax' && videoStart === 'still' && minimaxGraph === 'v2'" class="rounded-2xl border border-white/10 bg-zinc-950/40 px-4 py-3 space-y-3">
             <label class="flex cursor-pointer items-center justify-between gap-3 text-sm">
               <span>
                 <span class="block font-medium text-zinc-200">Use identity references</span>
@@ -466,13 +523,13 @@
                   <span class="text-xs text-zinc-400">{{ option.id === 'auto' ? autoHint : option.hint }}</span>
                 </button>
               </div>
-              <p class="mt-2 text-xs text-zinc-500">Output {{ outputSize.width }} × {{ outputSize.height }}{{ aspect === 'auto' ? ' · matching the still' : '' }}</p>
+              <p class="mt-2 text-xs text-zinc-500">Output {{ outputSize.width }} × {{ outputSize.height }}{{ aspect === 'auto' ? (file ? ' · matching the still' : ' · 16:9 until a still is loaded') : '' }}</p>
             </div>
             <div>
               <p class="mb-2 text-xs uppercase tracking-wider text-zinc-500">Steps</p>
               <div class="grid grid-cols-2 gap-2">
-                <button type="button" class="rounded-xl border px-3 py-2 text-sm" :class="turbo ? 'border-amber-300 bg-amber-400/10' : 'border-white/10'" @click="turbo = true">{{ videoWorkflow === 'v2' ? 6 : 8 }} Turbo · LoRA on</button>
+                <button type="button" class="rounded-xl border px-3 py-2 text-sm" :class="turbo ? 'border-amber-300 bg-amber-400/10' : 'border-white/10'" @click="turbo = true">{{ ltxVideo ? '8 Distilled LoRA' : (videoWorkflow === 'v2' ? '6 Turbo · LoRA on' : '8 Turbo · LoRA on') }}</button>
-                <button type="button" class="rounded-xl border px-3 py-2 text-sm" :class="!turbo ? 'border-amber-300 bg-amber-400/10' : 'border-white/10'" @click="turbo = false">20 High-Fidelity · LoRA bypass</button>
+                <button type="button" class="rounded-xl border px-3 py-2 text-sm" :class="!turbo ? 'border-amber-300 bg-amber-400/10' : 'border-white/10'" @click="turbo = false">{{ ltxVideo ? '12 Quality · LoRA on' : '20 High-Fidelity · LoRA bypass' }}</button>
               </div>
             </div>
             <div class="text-sm">
@@ -499,7 +556,7 @@
                   @change="onCfgNumber"
                 >
               </div>
-              <p class="mt-1 text-[11px] text-zinc-500">{{ turbo ? 'Default 1.5 with Turbo LoRA' : 'Default 4.0 with High-Fidelity' }}</p>
+              <p class="mt-1 text-[11px] text-zinc-500">{{ ltxVideo ? 'Default 3.5 for PinkCherry' : (turbo ? 'Default 1.5 with Turbo LoRA' : 'Default 4.0 with High-Fidelity') }}</p>
             </div>
             <div class="grid grid-cols-2 gap-3">
               <label class="text-sm">
@@ -1256,6 +1313,19 @@
 
 <script setup lang="ts">
 import { parseEditPasses, looksLikeShotScript } from '~/utils/parseRecommend'
+import { parseShotScript } from '~/utils/parseShots'
+import {
+  composeVideoWorkflow,
+  isLtxWorkflow,
+  isTextToVideo,
+  minimaxGraphOf,
+  parseVideoWorkflow,
+  videoEngineOf,
+  videoStartOf,
+  type MiniMaxGraphId,
+  type VideoEngineId,
+  type VideoStartId
+} from '~/utils/videoModels'
 
 const runtimeConfig = useRuntimeConfig()
 
@@ -1311,7 +1381,7 @@ interface LibraryClip {
   familyId?: string
   parentClipId?: string
   chainIndex?: number
-  workflow?: 'v1' | 'v2'
+  workflow?: string
   stillId?: string
   sound?: boolean
   hideInput?: boolean
@@ -1340,7 +1410,7 @@ interface RetryDraft {
   samplerName?: string
   scheduler?: string
   extensions?: { prompt: string; duration: number }[]
-  workflow?: 'v1' | 'v2'
+  workflow?: string
 }
 
 interface QueuedExtension {
@@ -1517,7 +1587,12 @@ const pickerSlot = ref<'main' | 'editRef' | number>('main')
 const libraryFilter = ref<'all' | 'images' | 'videos'>('all')
 const selectedKeys = ref<string[]>([])
 const lastSelectedKey = ref('')
-const videoWorkflow = ref('v1')
+const videoEngine = ref<VideoEngineId>('minimax')
+const videoStart = ref<VideoStartId>('still')
+const minimaxGraph = ref<MiniMaxGraphId>('v1')
+const videoWorkflow = computed(() => composeVideoWorkflow(videoEngine.value, videoStart.value, minimaxGraph.value))
+const textToVideo = computed(() => isTextToVideo(videoWorkflow.value))
+const ltxVideo = computed(() => isLtxWorkflow(videoWorkflow.value))
 const useIdentityRefs = ref(false)
 const identityRefs = ref<(File | null)[]>([null, null, null, null])
 const identityPreviews = ref(['', '', '', ''])
@@ -1653,10 +1728,22 @@ const outputSize = computed(() => {
 })
 
 const autoHint = computed(() => {
-  if (!imageWidth.value || !imageHeight.value) return 'Match the still'
+  if (!imageWidth.value || !imageHeight.value) {
+    return textToVideo.value ? '16:9 without a still' : 'Match the still'
+  }
   const size = autoResolution(imageWidth.value, imageHeight.value)
   return `${size.width} × ${size.height} · from still`
 })
+const videoModelHint = computed(() => {
+  if (ltxVideo.value) {
+    return textToVideo.value
+      ? 'PinkCherry text-to-video. Distilled LoRA stays on. Silent MP4 — LTX has no audio VAE here.'
+      : 'PinkCherry image-to-video. Distilled LoRA stays on. Silent MP4 — LTX has no audio VAE here.'
+  }
+  if (textToVideo.value) return 'MiniMax H3 text-to-video with native stereo audio. Shot 2+ switches to last-frame image-to-video.'
+  if (videoWorkflow.value === 'v2') return 'MiniMax H3 image-to-video. Identity refs stay on v2 with a start still.'
+  return 'MiniMax H3 image-to-video with native stereo audio.'
+})
 const frameCount = computed(() => Math.max(5, Math.floor(duration.value * fps.value)))
 const cfgPct = computed(() => ((cfg.value - CFG_MIN) / (CFG_MAX - CFG_MIN)) * 100)
 const currentClip = computed(() => clips.value.find(clip => clip.id === currentClipId.value))
@@ -1671,7 +1758,7 @@ const queueReady = computed(() => {
   return extensionQueue.value.every(item => item.prompt.trim())
 })
 const videoGenerateDisabled = computed(() => {
-  return videoBusy.value || !file.value || !prompt.value.trim() || !folderId.value || !queueReady.value
+  return videoBusy.value || (!textToVideo.value && !file.value) || !prompt.value.trim() || !folderId.value || !queueReady.value
 })
 const recommendDisabled = computed(() => {
   if (recommendBusy.value) return true
@@ -1715,6 +1802,10 @@ const editLabel = computed(() => {
   return n === 1 ? 'Edit image + 1 pass' : `Edit image + ${n} passes`
 })
 const parsedShots = computed(() => parseShotScript(prompt.value))
+const recommendedShots = computed(() => {
+  if (!recommendText.value || recommendBusy.value) return []
+  return parseShotScript(recommendText.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(() => {
@@ -2101,7 +2192,20 @@ onMounted(async () => {
     hideInputPreview.value = localStorage.getItem('aigen-hide-input-preview') === 'true'
     withSound.value = localStorage.getItem('aigen-generate-sound') !== 'false'
     autoplayEnabled.value = localStorage.getItem('aigen-autoplay') === 'true'
-    videoWorkflow.value = localStorage.getItem('aigen-video-workflow') === 'v2' ? 'v2' : 'v1'
+    videoEngine.value = 'minimax'
+    videoStart.value = 'still'
+    minimaxGraph.value = 'v1'
+    try {
+      const saved = parseVideoWorkflow(localStorage.getItem('aigen-video-workflow') || '')
+      videoEngine.value = videoEngineOf(saved)
+      videoStart.value = videoStartOf(saved)
+      minimaxGraph.value = minimaxGraphOf(saved)
+      const engine = localStorage.getItem('aigen-video-engine')
+      const start = localStorage.getItem('aigen-video-start')
+      if (engine === 'ltx' || engine === 'minimax') videoEngine.value = engine
+      if (start === 'text' || start === 'still') videoStart.value = start
+      if (videoEngine.value === 'ltx') applyEngineDefaults('ltx')
+    } catch { /* ignore */ }
     useIdentityRefs.value = localStorage.getItem('aigen-use-identity-refs') === 'true'
     activeQueueId.value = localStorage.getItem(QUEUE_STORE) || ''
     const savedPreset = localStorage.getItem('aigen-prompt-preset')
@@ -2190,6 +2294,18 @@ watch(videoWorkflow, (value) => {
   } catch { /* ignore */ }
 })
 
+watch(videoEngine, (value) => {
+  try {
+    localStorage.setItem('aigen-video-engine', value)
+  } catch { /* ignore */ }
+})
+
+watch(videoStart, (value) => {
+  try {
+    localStorage.setItem('aigen-video-start', value)
+  } catch { /* ignore */ }
+})
+
 watch(useIdentityRefs, (value) => {
   try {
     localStorage.setItem('aigen-use-identity-refs', String(value))
@@ -2197,7 +2313,8 @@ watch(useIdentityRefs, (value) => {
 })
 
 watch(turbo, (on) => {
-  if (!cfgTouched.value) cfg.value = on ? CFG_TURBO : CFG_QUALITY
+  if (cfgTouched.value) return
+  cfg.value = ltxVideo.value ? 3.5 : (on ? CFG_TURBO : CFG_QUALITY)
 })
 
 function clampDuration(value: number) {
@@ -2336,7 +2453,7 @@ function applyRecommend() {
     }
   } else {
     prompt.value = text
-    if (looksLikeShotScript(text)) shotScriptMode.value = true
+    if (looksLikeShotScript(text) || parseShotScript(text).length > 1) shotScriptMode.value = true
   }
   recommendApplied.value = true
   toast('Recommended prompt applied')
@@ -2709,7 +2826,7 @@ async function rerun(item: LibraryClip, collection = false) {
   if (showPrivacyToggles.value) hideInputPreview.value = target.hideInput === true
   folderId.value = target.folderId
   browseFolderId.value = target.folderId
-  videoWorkflow.value = target.workflow === 'v2' ? 'v2' : 'v1'
+  applyVideoWorkflow(target.workflow)
   if (typeof target.sound === 'boolean') withSound.value = target.sound
   shotScriptMode.value = restoreAll
   useIdentityRefs.value = false
@@ -2722,7 +2839,8 @@ async function rerun(item: LibraryClip, collection = false) {
     }))
     : []
   try {
-    await loadRerunImage(target, parts)
+    const initialTextToVideo = textToVideo.value && !target.parentClipId && !(target.chainIndex)
+    if (!(initialTextToVideo && !target.stillId)) await loadRerunImage(target, parts)
     statusMessage.value = restoreAll
       ? 'Collection loaded into the form. Generate when you are ready.'
       : 'Clip loaded into the form. Generate when you are ready.'
@@ -2730,8 +2848,12 @@ async function rerun(item: LibraryClip, collection = false) {
       ? 'Shot script and settings restored. Make any changes, then generate.'
       : 'Inputs restored. Make any changes, then generate.')
   } catch {
-    statusMessage.value = 'Clip settings loaded. Re-drop the start still if it is missing.'
+    statusMessage.value = textToVideo.value
-    toast('Settings restored, but the start still could not be loaded. Pick it again.')
+      ? 'Clip settings loaded. Add a still only if you want image-to-video.'
+      : 'Clip settings loaded. Re-drop the start still if it is missing.'
+    toast(textToVideo.value
+      ? 'Settings restored. Text-to-video does not need a start still.'
+      : 'Settings restored, but the start still could not be loaded. Pick it again.')
   }
 }
 
@@ -2746,7 +2868,7 @@ async function loadDraft(draft: RetryDraft) {
   withSound.value = draft.sound !== false
   if (draft.hideInput && showPrivacyToggles.value) hideInputPreview.value = true
   duration.value = clampDuration(Number(draft.duration))
-  if (draft.workflow === 'v2') videoWorkflow.value = 'v2'
+  applyVideoWorkflow(draft.workflow)
   extensionQueue.value = (draft.extensions || []).map(item => ({
     id: crypto.randomUUID(),
     prompt: String(item.prompt || ''),
@@ -2758,9 +2880,13 @@ async function loadDraft(draft: RetryDraft) {
     readFile(new File([blob], filename, { type: blob.type || 'image/png' }), { keepThumbnailPref: true })
     statusMessage.value = 'Held job restored. Generate when ComfyUI is free.'
   } catch {
+    if (textToVideo.value) {
+      statusMessage.value = 'Held job restored. Generate when ComfyUI is free.'
+    } else {
       statusMessage.value = 'Held job settings restored. Re-drop the still if the image is missing.'
       toast('Settings restored. Unlock the folder or re-drop the still if the image is missing.')
     }
+  }
 }
 
 async function dismissDraft(draft: RetryDraft) {
@@ -3363,8 +3489,32 @@ async function useEditAsInput() {
   }
 }
 
+function applyEngineDefaults(engine: VideoEngineId) {
+  if (engine === 'ltx') {
+    samplerName.value = 'euler'
+    if (!cfgTouched.value) cfg.value = 3.5
+    return
+  }
+  if (samplerName.value === 'euler') samplerName.value = 'res_multistep'
+  if (!cfgTouched.value) cfg.value = turbo.value ? CFG_TURBO : CFG_QUALITY
+}
+
+function selectVideoEngine(engine: VideoEngineId) {
+  if (videoEngine.value === engine) return
+  cfgTouched.value = false
+  videoEngine.value = engine
+  applyEngineDefaults(engine)
+}
+
+function applyVideoWorkflow(raw?: string) {
+  const parsed = parseVideoWorkflow(raw)
+  videoEngine.value = videoEngineOf(parsed)
+  videoStart.value = videoStartOf(parsed)
+  minimaxGraph.value = minimaxGraphOf(parsed)
+}
+
 async function generate() {
-  if (!file.value || !prompt.value.trim()) return
+  if ((!file.value && !textToVideo.value) || !prompt.value.trim()) return
   if (!queueReady.value) {
     toast(shotScriptMode.value ? 'Each shot in the script needs a prompt.' : 'Each queued extension needs a prompt.')
     return
@@ -3401,7 +3551,7 @@ async function generate() {
   startTimer('video')
   try {
     const body = new FormData()
-    body.append('image', file.value)
+    if (file.value) body.append('image', file.value)
     body.append('prompt', initialPrompt)
     body.append('aspect', aspect.value)
     body.append('width', String(outputSize.value.width))

server/api/generate.post.ts

@@ -1,5 +1,6 @@
 import { frameLength, runGeneration } from '~/server/utils/videoChain'
 import { createShotQueue, finishQueueBurst, setQueueJob } from '~/server/utils/shotQueue'
+import { defaultVideoSteps, isLtxWorkflow, isTextToVideo, parseVideoWorkflow } from '~/utils/videoModels'
 
 function isComfyBusyTimeout(error: unknown) {
   const err = error as { statusCode?: number; status?: number; data?: { code?: string }; message?: string; statusMessage?: string }
@@ -42,16 +43,16 @@ function parseFps(raw: string | undefined) {
   return fps === 12 || fps === 30 || fps === 24 ? fps : 24
 }
 
-function parseCfg(raw: string | undefined, turbo: boolean) {
+function parseCfg(raw: string | undefined, turbo: boolean, ltx = false) {
-  const fallback = turbo ? 1.5 : 4
+  const fallback = ltx ? 3.5 : (turbo ? 1.5 : 4)
   const value = Number(raw)
   if (!Number.isFinite(value)) return fallback
   const snapped = Math.round(value * 2) / 2
   return Math.min(10, Math.max(1, snapped))
 }
 
-function parseSampler(raw: string | undefined) {
+function parseSampler(raw: string | undefined, ltx = false) {
-  return SAMPLERS.has(raw || '') ? raw! : 'res_multistep'
+  return SAMPLERS.has(raw || '') ? raw! : (ltx ? 'euler' : 'res_multistep')
 }
 
 function parseScheduler(raw: string | undefined) {
@@ -86,7 +87,9 @@ export default defineEventHandler(async (event) => {
     }
   }
 
-  if (!image) {
+  const workflow = parseVideoWorkflow(fields.workflow)
+  const textToVideo = isTextToVideo(workflow)
+  if (!image && !textToVideo) {
     throw createError({ statusCode: 400, statusMessage: 'An initial image is required' })
   }
   const prompt = (fields.prompt || '').trim()
@@ -94,10 +97,13 @@ export default defineEventHandler(async (event) => {
     throw createError({ statusCode: 400, statusMessage: 'A motion prompt is required' })
   }
   const extensions = parseExtensions(fields.extensions)
-  const workflow = parseVideoWorkflow(fields.workflow)
+  const useIdentityRefs = workflow === 'v2' && !textToVideo && !isLtxWorkflow(workflow) && fields.useIdentityRefs === 'true'
-  const useIdentityRefs = workflow === 'v2' && fields.useIdentityRefs === 'true'
+  const { width, height } = resolveOutputSize(
-
+    fields.aspect,
-  const { width, height } = resolveOutputSize(fields.aspect, image.data, Number(fields.width), Number(fields.height))
+    image?.data || Buffer.alloc(0),
+    Number(fields.width),
+    Number(fields.height)
+  )
   const turbo = fields.turbo !== 'false'
   const steps = Number(fields.steps || defaultVideoSteps(turbo, workflow))
   const seed = fields.seed && fields.seed !== 'random'
@@ -105,13 +111,13 @@ export default defineEventHandler(async (event) => {
     : Math.floor(Math.random() * 2_147_483_647)
   const durationSeconds = parseDuration(fields.duration || '5')
   const fps = parseFps(fields.fps)
-  const cfg = parseCfg(fields.cfg, turbo)
+  const cfg = parseCfg(fields.cfg, turbo, isLtxWorkflow(workflow))
-  const samplerName = parseSampler(fields.sampler_name)
+  const samplerName = parseSampler(fields.sampler_name, isLtxWorkflow(workflow))
   const scheduler = parseScheduler(fields.scheduler)
   const length = frameLength(durationSeconds, fps)
   const hideThumbnail = fields.hideThumbnail === 'true'
   const hideInput = fields.hideInput === 'true'
-  const sound = fields.sound !== 'false'
+  const sound = !isLtxWorkflow(workflow) && fields.sound !== 'false'
   const clipName = (fields.name || '').trim().slice(0, 80)
   const ownerKey = libraryOwnerKey(event)
   const library = publicLibrary(event)
@@ -124,7 +130,8 @@ export default defineEventHandler(async (event) => {
   const destFolder = library.folders.find(folder => folder.id === folderId)
   const folderLocked = Boolean(destFolder?.protected && !destFolder.unlocked)
   assertFolderExists(event, folderId)
-  const still = await rememberInputStill({
+  const still = image
+    ? await rememberInputStill({
       ownerKey,
       folderId,
       filename: image.filename,
@@ -133,6 +140,7 @@ export default defineEventHandler(async (event) => {
       height,
       hideInput
     })
+    : null
   const referenceStillIds: Array<string | null> = [null, null, null, null]
   if (useIdentityRefs) {
     for (const [index, ref] of referenceImages.entries()) {
@@ -175,7 +183,7 @@ export default defineEventHandler(async (event) => {
     fps,
     samplerName,
     scheduler,
-    thumb: isPipelineFrameFilename(image.filename) ? undefined : image.data,
+    thumb: image && !isPipelineFrameFilename(image.filename) ? image.data : undefined,
     stillId: still?.id,
     stillFilename: still?.filename,
     referenceStillIds,

server/assets/workflow_ltx_video.json

@@ -0,0 +1,168 @@
+{
+  "1": {
+    "inputs": {
+      "unet_name": "PinkCherry_FineTune_int8_v1_8_LTX23.safetensors",
+      "weight_dtype": "default"
+    },
+    "class_type": "UNETLoader",
+    "_meta": { "title": "Load PinkCherry LTX-2.3" }
+  },
+  "21": {
+    "inputs": {
+      "lora_name": "ltx-2.3-22b-distilled-lora-384-1.1.safetensors",
+      "strength_model": 1,
+      "model": ["1", 0]
+    },
+    "class_type": "LoraLoaderModelOnly",
+    "_meta": { "title": "LTX distilled LoRA" }
+  },
+  "3": {
+    "inputs": {
+      "vae_name": "LTX23_video_vae_bf16.safetensors"
+    },
+    "class_type": "VAELoader",
+    "_meta": { "title": "Load VAE" }
+  },
+  "4": {
+    "inputs": {
+      "image": "input_image.png",
+      "upload": "image"
+    },
+    "class_type": "LoadImage",
+    "_meta": { "title": "Load Image" }
+  },
+  "5": {
+    "inputs": {
+      "width": 1280,
+      "height": 1280,
+      "interpolation": "lanczos",
+      "method": "keep proportion",
+      "condition": "always",
+      "multiple_of": 32,
+      "image": ["4", 0]
+    },
+    "class_type": "ImageResize+",
+    "_meta": { "title": "Resize still" }
+  },
+  "6": {
+    "inputs": {
+      "text": "",
+      "clip": ["20", 0]
+    },
+    "class_type": "CLIPTextEncode",
+    "_meta": { "title": "Positive prompt" }
+  },
+  "7": {
+    "inputs": {
+      "text": "blurry, low quality, jitter, deformed, static frame, freeze, watermark, bad anatomy, text, subtitles",
+      "clip": ["20", 0]
+    },
+    "class_type": "CLIPTextEncode",
+    "_meta": { "title": "Negative prompt" }
+  },
+  "8": {
+    "inputs": {
+      "width": ["5", 1],
+      "height": ["5", 2],
+      "length": 97,
+      "batch_size": 1,
+      "strength": 1,
+      "positive": ["6", 0],
+      "negative": ["7", 0],
+      "vae": ["3", 0],
+      "image": ["5", 0]
+    },
+    "class_type": "LTXVImgToVideo",
+    "_meta": { "title": "LTX image to video" }
+  },
+  "9": {
+    "inputs": {
+      "frame_rate": 24,
+      "positive": ["8", 0],
+      "negative": ["8", 1]
+    },
+    "class_type": "LTXVConditioning",
+    "_meta": { "title": "LTX conditioning" }
+  },
+  "10": {
+    "inputs": {
+      "cfg": 3.5,
+      "model": ["21", 0],
+      "positive": ["9", 0],
+      "negative": ["9", 1]
+    },
+    "class_type": "CFGGuider",
+    "_meta": { "title": "CFG Guider" }
+  },
+  "11": {
+    "inputs": {
+      "sampler_name": "euler"
+    },
+    "class_type": "KSamplerSelect",
+    "_meta": { "title": "KSamplerSelect" }
+  },
+  "12": {
+    "inputs": {
+      "steps": 8,
+      "max_shift": 2.05,
+      "base_shift": 0.95,
+      "stretch": true,
+      "terminal": 0.1,
+      "latent": ["8", 2]
+    },
+    "class_type": "LTXVScheduler",
+    "_meta": { "title": "LTX scheduler" }
+  },
+  "13": {
+    "inputs": {
+      "noise_seed": 1
+    },
+    "class_type": "RandomNoise",
+    "_meta": { "title": "RandomNoise" }
+  },
+  "14": {
+    "inputs": {
+      "noise": ["13", 0],
+      "guider": ["10", 0],
+      "sampler": ["11", 0],
+      "sigmas": ["12", 0],
+      "latent_image": ["8", 2]
+    },
+    "class_type": "SamplerCustomAdvanced",
+    "_meta": { "title": "SamplerCustomAdvanced" }
+  },
+  "15": {
+    "inputs": {
+      "samples": ["14", 0],
+      "vae": ["3", 0]
+    },
+    "class_type": "VAEDecode",
+    "_meta": { "title": "VAE Decode" }
+  },
+  "16": {
+    "inputs": {
+      "frame_rate": 24,
+      "loop_count": 0,
+      "filename_prefix": "video/LTX23",
+      "format": "video/h264-mp4",
+      "pix_fmt": "yuv420p",
+      "crf": 19,
+      "save_metadata": true,
+      "trim_to_audio": false,
+      "pingpong": false,
+      "save_output": true,
+      "images": ["15", 0]
+    },
+    "class_type": "VHS_VideoCombine",
+    "_meta": { "title": "Save MP4" }
+  },
+  "20": {
+    "inputs": {
+      "clip_name1": "gemma-3-12b-it-qat-UD-Q4_K_XL.gguf",
+      "clip_name2": "text_encoders\\ltx-2.3-22b-dev_embeddings_connectors.safetensors",
+      "type": "ltxv"
+    },
+    "class_type": "DualCLIPLoaderGGUF",
+    "_meta": { "title": "DualCLIPLoader (GGUF)" }
+  }
+}

server/utils/comfy.ts

@@ -1,3 +1,5 @@
+import { LTX_NEGATIVE } from '~/utils/videoModels'
+
 let comfyHostOverride = ''
 
 export function setComfyHostOverride(url: string) {
@@ -188,7 +190,12 @@ export function extractPromptFromHistory(entry: unknown) {
     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
+    if (node?.class_type === 'MiniMaxH3ImageToVideo' || node?.class_type === 'MiniMaxH3TextToVideo') imageToVideo = text
+    if (node?.class_type === 'CLIPTextEncode') {
+      const title = String((node as { _meta?: { title?: string } })._meta?.title || '')
+      if (/negative/i.test(title) || text === LTX_NEGATIVE) continue
+      if (!imageToVideo) imageToVideo = text
+    }
   }
   return imageToVideo
 }
@@ -200,14 +207,18 @@ export function extractClipMetaFromHistory(entry: unknown) {
   if (!graph || typeof graph !== 'object') return meta
   for (const node of Object.values(graph as Record<string, { class_type?: string; inputs?: Record<string, unknown> }>)) {
     const inputs = node?.inputs || {}
-    if (node?.class_type === 'MiniMaxH3ImageToVideo') {
+    if (node?.class_type === 'MiniMaxH3ImageToVideo' || node?.class_type === 'MiniMaxH3TextToVideo' || node?.class_type === 'EmptyLTXVLatentVideo') {
       meta.width = Number(inputs.width || meta.width)
       meta.height = Number(inputs.height || meta.height)
     }
-    if (node?.class_type === 'BasicScheduler' || node?.class_type === 'PrimitiveInt') {
+    if (node?.class_type === 'BasicScheduler' || node?.class_type === 'PrimitiveInt' || node?.class_type === 'LTXVScheduler') {
       const steps = Number(inputs.steps || 0)
       if (steps > 0) meta.steps = steps
     }
+    if (node?.class_type === 'ImageResize+' && typeof inputs.width === 'number' && typeof inputs.height === 'number') {
+      meta.width = Number(inputs.width)
+      meta.height = Number(inputs.height)
+    }
     if (node?.class_type === 'RandomNoise' || node?.class_type === 'SamplerMiniMax' || node?.class_type === 'KSampler') {
       const seed = Number(inputs.noise_seed ?? inputs.seed ?? 0)
       if (seed) meta.seed = seed
@@ -276,13 +287,16 @@ export function comfyFilenamePrefix() {
 }
 
 export function isOurComfyVideo(video: { filename: string; subfolder: string }) {
-  const prefix = comfyFilenamePrefix().replace(/\/$/, '')
+  const prefixes = [comfyFilenamePrefix(), 'video/LTX23']
+  return prefixes.some((raw) => {
+    const prefix = raw.replace(/\/$/, '')
     const parts = prefix.split('/')
     const namePrefix = parts[parts.length - 1]
     const sub = parts.length > 1 ? parts.slice(0, -1).join('/') : 'video'
-  const nameOk = video.filename.startsWith(namePrefix)
+    const nameOk = video.filename.startsWith(namePrefix) || video.filename.startsWith('LTX23')
     const subOk = !video.subfolder || video.subfolder === sub
     return nameOk && subOk
+  })
 }
 
 export function purgeComfyEnabled() {

server/utils/jobs.ts

@@ -85,7 +85,7 @@ export interface Job {
     familyId?: string
     parentClipId?: string
     parentStillId?: string
-    workflow?: 'v1' | 'v2'
+    workflow?: import('~/utils/videoModels').VideoWorkflowId
     chainContinuing?: boolean
     passes?: { prompt: string }[]
     queueId?: string

server/utils/library.ts

@@ -38,7 +38,7 @@ export interface LibraryClip {
   familyId?: string
   parentClipId?: string
   chainIndex?: number
-  workflow?: 'v1' | 'v2'
+  workflow?: import('~/utils/videoModels').VideoWorkflowId
   stillId?: string
   sound?: boolean
   hideInput?: boolean
@@ -102,7 +102,7 @@ export interface RetryDraft {
   samplerName?: string
   scheduler?: string
   extensions?: QueuedExtension[]
-  workflow?: 'v1' | 'v2'
+  workflow?: import('~/utils/videoModels').VideoWorkflowId
 }
 
 interface Catalog {
@@ -969,7 +969,7 @@ export async function saveRetryDraft(params: {
   samplerName?: string
   scheduler?: string
   extensions?: QueuedExtension[]
-  workflow?: 'v1' | 'v2'
+  workflow?: import('~/utils/videoModels').VideoWorkflowId
 }) {
   return mutate(params.ownerKey, (catalog) => {
     const existing = params.id ? catalog.drafts.find(item => item.id === params.id) : undefined
@@ -1134,7 +1134,7 @@ export async function saveClip(params: {
   familyId?: string
   parentClipId?: string
   chainIndex?: number
-  workflow?: 'v1' | 'v2'
+  workflow?: import('~/utils/videoModels').VideoWorkflowId
   stillId?: string
   sound?: boolean
 }) {

server/utils/pending.ts

@@ -35,7 +35,7 @@ export interface PendingJob {
   referenceImageNames?: string[]
   referenceStillIds?: Array<string | null>
   useIdentityRefs?: boolean
-  workflow?: 'v1' | 'v2'
+  workflow?: import('~/utils/videoModels').VideoWorkflowId
   duration?: number
   cfg?: number
   fps?: number

server/utils/promptComfy.ts

@@ -1,5 +1,6 @@
 import { getSidecarImageHost } from '~/server/utils/imageComfy'
 import { buildVisionPromptWorkflow } from '~/server/utils/promptWorkflow'
+import { normalizeShotScript } from '~/utils/parseRecommend'
 
 export type PromptJobStatus = 'queued' | 'running' | 'complete' | 'error'
 
@@ -206,7 +207,10 @@ function draftUserPrompt(draft: string, mode: string) {
     : mode === 'shot-script'
       ? 'video-shot-script'
       : 'video'
-  return `Studio mode: ${studio}\n\nDraft Prompt:\n${draft}`
+  const format = studio === 'image-edit'
+    ? 'If this is a single edit, output only the refined prompt. If it needs phases, use ---pass 1--- blocks.'
+    : 'Output a MiniMax shot list. Each shot starts with "shot N" on its own line, then [SHOT CONFIGURATION], [SUBJECT DIRECTION & ACTION], and [AUDIO CUES], each on its own line. Blank line between shots. No markdown fences. No one-paragraph dump.'
+  return `Studio mode: ${studio}\n${format}\n\nDraft Prompt:\n${draft}`
 }
 
 export async function runPromptRecommend(job: PromptJob, params: {
@@ -227,7 +231,8 @@ export async function runPromptRecommend(job: PromptJob, params: {
   const queued = await queuePromptGraph(graph, job.id)
   job.message = 'Writing the recommended prompt…'
   const text = await waitForPromptText(queued.prompt_id)
-  job.text = text.trim()
+  const trimmed = text.trim()
+  job.text = params.mode === 'edit' ? trimmed : normalizeShotScript(trimmed)
   job.status = 'complete'
   job.message = 'Recommended prompt ready'
 }

server/utils/promptPresets.ts

@@ -9,6 +9,49 @@ export interface PromptPreset {
   builtin?: boolean
 }
 
+const SHOT_SCRIPT_FORMAT = `Output ONLY a MiniMax shot list the user can paste into Shot script mode.
+
+Each shot starts with "shot N" on its own line (shot 1, shot 2, shot 3…). Then these three headers, each on its own line:
+
+[SHOT CONFIGURATION]
+[SUBJECT DIRECTION & ACTION]
+[AUDIO CUES]
+
+Blank line between shots. No preamble, no markdown fences, no commentary.
+Do not glue shots into one paragraph. "shot 2" must be a line by itself, never "shot 2:" mid-sentence.
+
+One beat per shot. Later shots: same people, same clothes as now, plus the one new action.
+Do not repeat a long identity paragraph on every shot.
+Never censor raw or explicit terminology.
+
+Example:
+
+shot 1
+[SHOT CONFIGURATION]
+- Shot Type: Medium-wide, 50mm, eye-level
+- Camera Movement: Slow track from a slight side-front angle
+- Lighting: Bright daylight through the trees
+- Environment/Scenery: The forest from the still
+
+[SUBJECT DIRECTION & ACTION]
+The woman from the still walks the path, taking in the trees.
+
+[AUDIO CUES]
+Footsteps on dirt, birds, breeze.
+
+shot 2
+[SHOT CONFIGURATION]
+- Shot Type: Medium, 50mm, eye-level
+- Camera Movement: Hold, then a small pan to her hand
+- Lighting: Same daylight
+- Environment/Scenery: Same path, same trees
+
+[SUBJECT DIRECTION & ACTION]
+She stops and rests her hand on the bark.
+
+[AUDIO CUES]
+Soft fabric, breeze, birds.`
+
 const ENHANCE_PROMPT = `You are a prompt-engineering assistant for Flux.2 Klein image edit and MiniMax H3 image-to-video.
 
 Look at the attached still. The person (or subject) in that still is the identity lock. The user draft is the requested change.
@@ -16,37 +59,27 @@ Look at the attached still. The person (or subject) in that still is the identit
 Write production-ready prompt text the user can paste into the generator.
 
 Rules:
-- Preserve identity from the still: face, age, body type, hair, skin, and distinctive features. Use explicit keep-identity language such as "same person as the reference still", "do not change facial identity", "keep original face, hair, and body".
+- Preserve identity from the still: face, age, body type, hair, skin, and distinctive features.
 - Describe wardrobe, setting, pose, lighting, camera, and materials concretely.
 - Never censor raw or explicit terminology from the draft.
 - Do not add a preamble, title, markdown fences, or commentary.
 
-Output format — pick one:
+If Studio mode is image-edit:
 - Single still/edit: output ONLY the refined prompt.
 - Ambitious still edit that should be done in phases (for example wardrobe first, then location): output:
 ---pass 1---
 <prompt for pass 1>
 ---pass 2---
 <prompt for pass 2>
-- Motion, multi-beat, or video sequence: first-shot prompt, then later shots with a marker on its own line:
+
-shot 2
+If Studio mode is video or video-shot-script:
-<extension prompt>
+${SHOT_SCRIPT_FORMAT}`
-shot 3
-<extension prompt>`
 
 const SHOT_SCRIPT_PROMPT = `You are a MiniMax H3 image-to-video shot-script writer.
 
-Analyze the attached still for identity, wardrobe, lighting, and camera. Turn the user's draft into a shot script they can paste into Shot script mode.
+Analyze the attached still for wardrobe, lighting, and camera. Turn the user's draft into a shot script.
 
-Format:
+${SHOT_SCRIPT_FORMAT}`
-- Shot 1 is the text BEFORE any marker (camera, action, audio for the first clip).
-- Later shots start with a line that is only: shot 2
-- Then shot 3, shot 4, and so on, each on its own line.
-- Every shot must keep the same person as the still unless the draft explicitly changes wardrobe or identity.
-- Include an [AUDIO CUES] block on shots where sound would help.
-- Never censor raw or explicit terminology.
-
-Output ONLY the shot script. No preamble, no markdown fences.`
 
 export const BUILTIN_PRESETS: PromptPreset[] = [
   {
@@ -60,7 +93,7 @@ export const BUILTIN_PRESETS: PromptPreset[] = [
     id: 'shot-script',
     name: 'Shot script',
     builtin: true,
-    description: 'Turn a draft into MiniMax shot 2 / shot 3 extension copy.',
+    description: 'Turn a draft into a MiniMax shot list with configuration, action, and audio blocks.',
     systemPrompt: SHOT_SCRIPT_PROMPT
   }
 ]

server/utils/shotQueue.ts

@@ -38,7 +38,7 @@ export interface ShotQueue {
   fps?: number
   samplerName?: string
   scheduler?: string
-  workflow?: 'v1' | 'v2'
+  workflow?: import('~/utils/videoModels').VideoWorkflowId
   sound?: boolean
   useIdentityRefs?: boolean
   referenceStillIds?: Array<string | null>
@@ -153,7 +153,7 @@ export async function createShotQueue(params: {
   fps?: number
   samplerName?: string
   scheduler?: string
-  workflow?: 'v1' | 'v2'
+  workflow?: import('~/utils/videoModels').VideoWorkflowId
   sound?: boolean
   useIdentityRefs?: boolean
   referenceStillIds?: Array<string | null>

server/utils/videoChain.ts

@@ -5,6 +5,7 @@ import { pendingFromJob, remainingAfterCurrentShot, writePendingJob } from '~/se
 import { emitChainJob, waitForComfySocket, watchComfyJob } from '~/server/utils/watch'
 import { comfyFilenamePrefix, queuePrompt, uploadImage } from '~/server/utils/comfy'
 import { buildWorkflow } from '~/server/utils/workflow'
+import { isLtxWorkflow, isTextToVideo, parseVideoWorkflow, workflowForExtension, type VideoWorkflowId } from '~/utils/videoModels'
 import { clipVideoPath, clipTitle, deleteRetryDraft, extendTempDir, getClip, nextClipPartName, removeExtendTemp, stillPath } from '~/server/utils/library'
 import { extractLastFrame, probeHasAudio } from '~/server/utils/ffmpeg'
 import { ensureComfyReady } from '~/server/utils/comfyLifecycle'
@@ -23,7 +24,7 @@ export type ChainImage = { filename: string; data: Buffer; type?: string }
 
 type VideoChainParams = {
   prompt: string
-  image: ChainImage
+  image?: ChainImage | null
   width: number
   height: number
   steps: number
@@ -36,7 +37,7 @@ type VideoChainParams = {
   samplerName: string
   scheduler: string
   extensions: { prompt: string; duration: number }[]
-  workflow: 'v1' | 'v2'
+  workflow: VideoWorkflowId
   duration: number
   useIdentityRefs: boolean
   referenceImages: Array<ChainImage | null>
@@ -76,7 +77,8 @@ function paramsFromJob(job: Job): VideoChainParams {
     throw new Error('Cannot continue the shot chain: job metadata is missing')
   }
   const image = loadStillImage(library.ownerKey, library.stillId, library.stillFilename)
-  if (!image?.data.length) {
+  const workflow = parseVideoWorkflow(library.workflow)
+  if (!image?.data.length && !isTextToVideo(workflow)) {
     throw new Error('Cannot continue the shot chain: the start still is missing from the library')
   }
   const referenceImages: Array<ChainImage | null> = [null, null, null, null]
@@ -101,7 +103,7 @@ function paramsFromJob(job: Job): VideoChainParams {
     samplerName: library.samplerName || 'res_multistep',
     scheduler: library.scheduler || 'simple',
     extensions: library.extensions || [],
-    workflow: library.workflow || 'v1',
+    workflow,
     duration,
     useIdentityRefs: library.useIdentityRefs === true,
     referenceImages
@@ -121,14 +123,21 @@ export async function queueMiniMax(
   const done = watchComfyJob(job, { persist: params.persist })
   job.status = 'uploading'
   const chainIndex = job.library?.chainIndex || 0
-  const uploading = params.useIdentityRefs
+  const graphId = chainIndex > 0 ? workflowForExtension(params.workflow) : params.workflow
+  const engineName = isLtxWorkflow(graphId) ? 'LTX-2.3' : 'MiniMax H3'
+  const hasImage = Boolean(params.image?.data?.length)
+  const uploading = !hasImage
+    ? `Queueing ${engineName} text-to-video…`
+    : params.useIdentityRefs
       ? (chainIndex > 0 ? 'Uploading identity stills for next shot...' : 'Uploading image to ComfyUI...')
       : (chainIndex > 0 ? 'Uploading last frame to ComfyUI...' : 'Uploading image to ComfyUI...')
-  const queueing = (job.library?.chainIndex || 0) > 0
+  const queueing = chainIndex > 0
-    ? 'Queueing extension on MiniMax H3...'
+    ? `Queueing extension on ${engineName}...`
-    : 'Queueing MiniMax H3 job...'
+    : `Queueing ${engineName} job...`
   emitChainJob(job, { type: 'status', message: uploading, progress: 4 })
-  const uploaded = await uploadImage(params.image, job.id)
+  const uploaded = hasImage && params.image
+    ? await uploadImage(params.image, job.id)
+    : { name: '', subfolder: '' }
   const referenceNames: string[] = ['', '', '', '']
   if (params.useIdentityRefs) {
     for (const [index, ref] of (params.referenceImages || []).entries()) {
@@ -161,9 +170,9 @@ export async function queueMiniMax(
     fps: params.fps,
     samplerName: params.samplerName,
     scheduler: params.scheduler,
-    filenamePrefix: comfyFilenamePrefix(),
+    filenamePrefix: isLtxWorkflow(graphId) ? 'video/LTX23' : comfyFilenamePrefix(),
-    sound: params.sound,
+    sound: params.sound && !isLtxWorkflow(graphId),
-    workflow: params.workflow,
+    workflow: graphId,
     duration: params.duration,
     useIdentityRefs: params.useIdentityRefs,
     referenceImageNames: params.useIdentityRefs ? referenceNames : []
@@ -317,7 +326,7 @@ export async function continueQueuedExtensions(
         samplerName: params.samplerName,
         scheduler: params.scheduler,
         persist,
-        workflow: params.workflow,
+        workflow: workflowForExtension(params.workflow),
         duration: ext.duration,
         useIdentityRefs: params.useIdentityRefs,
         referenceImages: params.useIdentityRefs ? params.referenceImages : []

server/utils/workflow.ts

@@ -1,9 +1,18 @@
 // Nitro bundles these JSON graphs into the production server output.
 import workflowV1 from '../assets/workflow_minimax_video.json'
 import workflowV2 from '../assets/workflow_minimax_video_v2.json'
+import workflowLtx from '../assets/workflow_ltx_video.json'
 import { buildIdentityPrompt } from '~/utils/identityPrompt'
+import {
+  isLtxWorkflow,
+  isTextToVideo,
+  LTX_NEGATIVE,
+  parseVideoWorkflow,
+  type VideoWorkflowId
+} from '~/utils/videoModels'
 
-export type VideoWorkflowId = 'v1' | 'v2'
+export type { VideoWorkflowId }
+export { parseVideoWorkflow }
 
 export interface GenerateParams {
   prompt: string
@@ -31,6 +40,7 @@ type WorkflowGraph = Record<string, WorkflowNode>
 
 const PROMPT_CLASSES = new Set([
   'MiniMaxH3ImageToVideo',
+  'MiniMaxH3TextToVideo',
   'MiniMaxH3ReferenceToVideo',
   'CLIPTextEncode',
   'CLIPTextEncodeQwen3VL',
@@ -58,16 +68,15 @@ function titleOf(node: WorkflowNode) {
 }
 
 function templateFor(id: VideoWorkflowId) {
+  if (isLtxWorkflow(id)) return workflowLtx as WorkflowGraph
   return (id === 'v2' ? workflowV2 : workflowV1) as WorkflowGraph
 }
 
-export function parseVideoWorkflow(raw: unknown): VideoWorkflowId {
-  return String(raw || '').trim() === 'v2' ? 'v2' : 'v1'
-}
-
 export function buildWorkflow(params: GenerateParams) {
-  const version = params.workflow === 'v2' ? 'v2' : 'v1'
+  const version = parseVideoWorkflow(params.workflow)
-  const graph = structuredClone(templateFor(version))
+  if (isLtxWorkflow(version)) return buildLtxWorkflow(params, version)
+
+  const graph = structuredClone(templateFor(version === 't2v' ? 'v1' : version === 'v2' ? 'v2' : 'v1'))
   const width = snap32(params.width)
   const height = snap32(params.height)
   const refs = Array.from({ length: 4 }, (_, index) => String(params.referenceImageNames?.[index] || '').trim())
@@ -155,7 +164,7 @@ export function buildWorkflow(params: GenerateParams) {
     }
   }
 
-  if (version === 'v1' && graph['128']?.class_type === 'ImageScaleToTotalPixels') {
+  if (version === 'v1' && !isTextToVideo(version) && graph['128']?.class_type === 'ImageScaleToTotalPixels') {
     graph['128'] = {
       class_type: 'ImageScale',
       inputs: {
@@ -169,7 +178,7 @@ export function buildWorkflow(params: GenerateParams) {
     }
   }
 
-  if (version === 'v1') {
+  if (version === 'v1' || version === 't2v') {
     graph['105:cfg'] = {
       class_type: 'FluxGuidance',
       inputs: {
@@ -191,10 +200,11 @@ export function buildWorkflow(params: GenerateParams) {
       graph[qualityId].inputs.value = params.turbo ? 20 : params.steps
     }
     if (turboId && graph[turboId]?.class_type === 'PrimitiveInt') {
-      graph[turboId].inputs.value = params.turbo ? params.steps : (params.workflow === 'v2' ? 6 : 8)
+      graph[turboId].inputs.value = params.turbo ? params.steps : (version === 'v2' ? 6 : 8)
     }
   }
 
+  if (version === 't2v') applyMinimaxTextToVideo(graph, params, width, height)
   if (version === 'v2') applyV2IdentityPath(graph, params, useRefs, refs, graphPrompt)
 
   if (params.sound === false) {
@@ -203,6 +213,7 @@ export function buildWorkflow(params: GenerateParams) {
     if (graph['105:91']?.inputs) delete graph['105:91'].inputs.audio
     delete graph['163']
     if (graph['172']?.inputs) delete graph['172'].inputs.audio
+    if (graph['105:104']?.inputs) delete graph['105:104'].inputs.audio_vae
   }
 
   return graph
@@ -210,10 +221,76 @@ export function buildWorkflow(params: GenerateParams) {
 
 function promptForNode(classType: string, useRefs: boolean, actionPrompt: string, graphPrompt: string) {
   if (classType === 'MiniMaxH3ImageToVideo') return useRefs ? '' : actionPrompt
+  if (classType === 'MiniMaxH3TextToVideo') return actionPrompt
   if (classType === 'MiniMaxH3ReferenceToVideo') return useRefs ? graphPrompt : actionPrompt
   return graphPrompt
 }
 
+function applyMinimaxTextToVideo(graph: WorkflowGraph, params: GenerateParams, width: number, height: number) {
+  const node = graph['105:104']
+  if (node) {
+    node.class_type = 'MiniMaxH3TextToVideo'
+    node._meta = { title: 'MiniMax H3 Text to Video' }
+    delete node.inputs.first_frame
+    node.inputs.prompt = params.prompt
+    node.inputs.width = width
+    node.inputs.height = height
+    if (params.sound !== false && graph['105:24']) node.inputs.audio_vae = ['105:24', 0]
+    else delete node.inputs.audio_vae
+  }
+  delete graph['114']
+  delete graph['119']
+  delete graph['120']
+  delete graph['127']
+  delete graph['128']
+}
+
+function buildLtxWorkflow(params: GenerateParams, version: VideoWorkflowId) {
+  const graph = structuredClone(workflowLtx as WorkflowGraph)
+  const width = snap32(params.width)
+  const height = snap32(params.height)
+  const textToVideo = isTextToVideo(version)
+  if (graph['4']?.inputs) graph['4'].inputs.image = params.imageName
+  if (graph['5']?.inputs) {
+    graph['5'].inputs.width = width
+    graph['5'].inputs.height = height
+  }
+  if (graph['6']?.inputs) graph['6'].inputs.text = params.prompt
+  if (graph['7']?.inputs) graph['7'].inputs.text = LTX_NEGATIVE
+  if (graph['8']?.inputs && typeof graph['8'].inputs.length === 'number') graph['8'].inputs.length = params.length
+  if (graph['9']?.inputs) graph['9'].inputs.frame_rate = params.fps
+  if (graph['10']?.inputs) graph['10'].inputs.cfg = params.cfg
+  if (graph['11']?.inputs) graph['11'].inputs.sampler_name = params.samplerName || 'euler'
+  if (graph['12']?.inputs) graph['12'].inputs.steps = params.steps
+  if (graph['13']?.inputs) graph['13'].inputs.noise_seed = params.seed
+  if (graph['16']?.inputs) {
+    graph['16'].inputs.frame_rate = params.fps
+    graph['16'].inputs.filename_prefix = params.filenamePrefix || 'video/LTX23'
+  }
+  if (graph['21']?.inputs) graph['21'].inputs.strength_model = 1
+  if (textToVideo) {
+    delete graph['4']
+    delete graph['5']
+    graph['8'] = {
+      class_type: 'EmptyLTXVLatentVideo',
+      inputs: {
+        width,
+        height,
+        length: params.length,
+        batch_size: 1
+      },
+      _meta: { title: 'Empty LTX latent' }
+    }
+    if (graph['9']?.inputs) {
+      graph['9'].inputs.positive = ['6', 0]
+      graph['9'].inputs.negative = ['7', 0]
+    }
+    if (graph['12']?.inputs) graph['12'].inputs.latent = ['8', 0]
+    if (graph['14']?.inputs) graph['14'].inputs.latent_image = ['8', 0]
+  }
+  return graph
+}
+
 function applyV2IdentityPath(
   graph: WorkflowGraph,
   params: GenerateParams,
@@ -304,9 +381,10 @@ function labelsFrom(template: WorkflowGraph) {
 export const NODE_LABELS: Record<string, string> = {
   ...Object.fromEntries(labelsFrom(workflowV1 as WorkflowGraph)),
   ...Object.fromEntries(labelsFrom(workflowV2 as WorkflowGraph)),
+  ...Object.fromEntries(labelsFrom(workflowLtx as WorkflowGraph)),
   ...LABEL_OVERRIDES
 }
 
 export function isEncodingNode(node: string) {
-  return node === '92' || node === '105:91' || node === '172' || /encoding|saving mp4|create video/i.test(NODE_LABELS[node] || '')
+  return node === '92' || node === '16' || node === '105:91' || node === '172' || /encoding|saving mp4|create video|save mp4/i.test(NODE_LABELS[node] || '')
 }

utils/parseRecommend.ts

@@ -1,3 +1,5 @@
+import { parseShotScript } from './parseShots'
+
 export function parseEditPasses(text: string): string[] | null {
   const src = String(text || '').replace(/\r\n/g, '\n')
   const marker = /^[ \t]*---pass\s+(\d+)---[ \t]*$/gim
@@ -21,3 +23,40 @@ export function parseEditPasses(text: string): string[] | null {
 export function looksLikeShotScript(text: string) {
   return /^[ \t]*shot\s+[2-9]\d*\b/im.test(String(text || ''))
 }
+
+function stripFences(text: string) {
+  return String(text || '')
+    .replace(/^\s*```(?:[\w-]+)?\s*\n?/, '')
+    .replace(/\n?```\s*$/, '')
+    .trim()
+}
+
+function splitInlineShotMarkers(text: string) {
+  return String(text || '')
+    .replace(/\r\n/g, '\n')
+    .replace(/(^|[^A-Za-z])shot\s+(\d+)\s*[:.)\-]?\s*/gi, '$1\n\nshot $2\n')
+    .replace(/\n{3,}/g, '\n\n')
+    .trim()
+}
+
+function formatShotBody(body: string) {
+  let text = String(body || '').trim()
+  text = text.replace(/^\s*[:.\-–)]\s*/, '')
+  text = text.replace(/\s*\[SHOT CONFIGURATION\]\s*/gi, '\n\n[SHOT CONFIGURATION]\n')
+  text = text.replace(/\s*\[SUBJECT DIRECTION(?:\s*&\s*ACTION)?\]\s*/gi, '\n\n[SUBJECT DIRECTION & ACTION]\n')
+  text = text.replace(/\s*\[AUDIO CUES\]\s*/gi, '\n\n[AUDIO CUES]\n')
+  text = text.replace(/[ \t]+\n/g, '\n').replace(/\n{3,}/g, '\n\n').trim()
+  return text
+}
+
+export function normalizeShotScript(text: string) {
+  const src = splitInlineShotMarkers(stripFences(text))
+  const shots = parseShotScript(src)
+  if (!shots.length) return src
+  if (shots.length === 1 && !/^[ \t]*shot\s+\d+\b/im.test(src)) {
+    return formatShotBody(shots[0].prompt)
+  }
+  return shots
+    .map(shot => `shot ${shot.n}\n${formatShotBody(shot.prompt)}`.trim())
+    .join('\n\n')
+}

utils/videoModels.ts

@@ -0,0 +1,56 @@
+export type VideoEngineId = 'minimax' | 'ltx'
+export type VideoStartId = 'still' | 'text'
+export type MiniMaxGraphId = 'v1' | 'v2'
+export type VideoWorkflowId = 'v1' | 'v2' | 't2v' | 'ltx' | 'ltx-t2v'
+
+export function parseVideoWorkflow(raw: unknown): VideoWorkflowId {
+  const value = String(raw || '').trim()
+  if (value === 'v2') return 'v2'
+  if (value === 't2v') return 't2v'
+  if (value === 'ltx' || value === 'ltx-i2v') return 'ltx'
+  if (value === 'ltx-t2v') return 'ltx-t2v'
+  return 'v1'
+}
+
+export function videoEngineOf(id: VideoWorkflowId): VideoEngineId {
+  return id.startsWith('ltx') ? 'ltx' : 'minimax'
+}
+
+export function videoStartOf(id: VideoWorkflowId): VideoStartId {
+  return id === 't2v' || id === 'ltx-t2v' ? 'text' : 'still'
+}
+
+export function isLtxWorkflow(id: VideoWorkflowId) {
+  return videoEngineOf(id) === 'ltx'
+}
+
+export function isTextToVideo(id: VideoWorkflowId) {
+  return videoStartOf(id) === 'text'
+}
+
+export function minimaxGraphOf(id: VideoWorkflowId): MiniMaxGraphId {
+  return id === 'v2' ? 'v2' : 'v1'
+}
+
+export function workflowForExtension(id: VideoWorkflowId): VideoWorkflowId {
+  if (id === 't2v') return 'v1'
+  if (id === 'ltx-t2v') return 'ltx'
+  return id
+}
+
+export function composeVideoWorkflow(engine: VideoEngineId, start: VideoStartId, graph: MiniMaxGraphId = 'v1'): VideoWorkflowId {
+  if (engine === 'ltx') return start === 'text' ? 'ltx-t2v' : 'ltx'
+  if (start === 'text') return 't2v'
+  return graph === 'v2' ? 'v2' : 'v1'
+}
+
+export function defaultVideoSteps(turbo: boolean, workflow: string = 'v1') {
+  const id = parseVideoWorkflow(workflow)
+  if (!turbo) return isLtxWorkflow(id) ? 12 : 20
+  if (isLtxWorkflow(id)) return 8
+  return id === 'v2' ? 6 : 8
+}
+
+export const LTX_UNET = 'PinkCherry_FineTune_int8_v1_8_LTX23.safetensors'
+export const LTX_DISTILLED_LORA = 'ltx-2.3-22b-distilled-lora-384-1.1.safetensors'
+export const LTX_NEGATIVE = 'blurry, low quality, jitter, deformed, static frame, freeze, watermark, bad anatomy, text, subtitles'

utils/videoSteps.ts

@@ -1,4 +0,0 @@
-export function defaultVideoSteps(turbo: boolean, workflow = 'v1') {
-  if (!turbo) return 20
-  return workflow === 'v2' ? 6 : 8
-}