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>

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) {
-  const fallback = turbo ? 1.5 : 4
+function parseCfg(raw: string | undefined, turbo: boolean, ltx = false) {
+  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) {
-  return SAMPLERS.has(raw || '') ? raw! : 'res_multistep'
+function parseSampler(raw: string | undefined, ltx = false) {
+  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' && fields.useIdentityRefs === 'true'
-
-  const { width, height } = resolveOutputSize(fields.aspect, image.data, Number(fields.width), Number(fields.height))
+  const useIdentityRefs = workflow === 'v2' && !textToVideo && !isLtxWorkflow(workflow) && fields.useIdentityRefs === 'true'
+  const { width, height } = resolveOutputSize(
+    fields.aspect,
+    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 samplerName = parseSampler(fields.sampler_name)
+  const cfg = parseCfg(fields.cfg, turbo, isLtxWorkflow(workflow))
+  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)
@@ -124,15 +130,17 @@ 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({
-    ownerKey,
-    folderId,
-    filename: image.filename,
-    data: image.data,
-    width,
-    height,
-    hideInput
-  })
+  const still = image
+    ? await rememberInputStill({
+      ownerKey,
+      folderId,
+      filename: image.filename,
+      data: image.data,
+      width,
+      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,