From 1314e8e89fde0d804ebd5af407be93246d6478db Mon Sep 17 00:00:00 2001
From: Towsty <ianjmcarr@gmail.com>
Date: Wed, 2 Sep 2026 08:19:21 -0500
Subject: [PATCH] Let failed studio jobs retry instead of sitting in the queue
 with only Remove.

Co-authored-by: Cursor <cursoragent@cursor.com>
---
 pages/index.vue                            | 18 +++++++++
 pages/queue.vue                            | 22 +++++++++++
 server/api/studio-queue/[id]/retry.post.ts |  7 ++++
 server/utils/studioQueue.ts                | 45 ++++++++++++++++++++--
 4 files changed, 88 insertions(+), 4 deletions(-)
 create mode 100644 server/api/studio-queue/[id]/retry.post.ts

diff --git a/pages/index.vue b/pages/index.vue
index 1537b5d..eb7c173 100644
--- a/pages/index.vue
+++ b/pages/index.vue
@@ -1167,6 +1167,14 @@
               >
                 {{ job.cutIn ? 'Next' : (canCutInStudioJob ? 'Run next' : 'Start now') }}
               </button>
+              <button
+                v-if="job.status === 'error' || (job.status === 'held' && !job.holdForCutIn)"
+                type="button"
+                class="shrink-0 rounded-full border border-amber-300/40 px-2.5 py-1 text-xs text-amber-100 hover:border-amber-200 hover:text-white"
+                @click="retryStudioJob(job.id)"
+              >
+                Retry
+              </button>
               <button
                 v-if="job.status === 'waiting' || job.status === 'held' || job.status === 'running' || job.status === 'error'"
                 type="button"
@@ -5085,6 +5093,16 @@ async function cutInStudioJob(id: string) {
   }
 }
 
+async function retryStudioJob(id: string) {
+  try {
+    await $fetch(`/api/studio-queue/${id}/retry`, { method: 'POST' })
+    toast('Queued to run as soon as the GPU is free.')
+    await refreshStudioQueue()
+  } catch (error: any) {
+    toast(error?.data?.statusMessage || error?.statusMessage || 'Could not retry that job')
+  }
+}
+
 async function dismissStudioJob(id: string) {
   try {
     await $fetch(`/api/studio-queue/${id}`, { method: 'DELETE' })
diff --git a/pages/queue.vue b/pages/queue.vue
index 3d6f57d..90ef684 100644
--- a/pages/queue.vue
+++ b/pages/queue.vue
@@ -84,6 +84,15 @@
             >
               {{ job.cutIn ? 'Next after current shot' : (canCutIn ? 'Run next' : 'Start now') }}
             </button>
+            <button
+              v-if="job.status === 'error'"
+              type="button"
+              class="rounded-full border border-amber-300/40 px-3 py-1.5 text-xs text-amber-100 hover:border-amber-200 disabled:cursor-not-allowed disabled:opacity-50"
+              :disabled="Boolean(busyId)"
+              @click="retryJob(job)"
+            >
+              Retry
+            </button>
             <template v-if="job.shots">
               <template v-if="canProcessShots(job)">
                 <button
@@ -756,6 +765,19 @@ async function cutIn(job: StudioJobRow) {
   }
 }
 
+async function retryJob(job: StudioJobRow) {
+  busyId.value = job.id
+  error.value = ''
+  try {
+    await $fetch(`/api/studio-queue/${job.id}/retry`, { method: 'POST' })
+    await loadQueues()
+  } catch (err: any) {
+    error.value = err?.data?.statusMessage || err?.statusMessage || err?.message || 'Could not retry that job'
+  } finally {
+    busyId.value = ''
+  }
+}
+
 function listenJob(job: StudioJobRow) {
   if (!job.liveJobId || sources.has(job.id)) return
   const source = new EventSource(`/api/generate/${job.liveJobId}/stream`)
diff --git a/server/api/studio-queue/[id]/retry.post.ts b/server/api/studio-queue/[id]/retry.post.ts
new file mode 100644
index 0000000..e7350c4
--- /dev/null
+++ b/server/api/studio-queue/[id]/retry.post.ts
@@ -0,0 +1,7 @@
+import { retryStudioJob } from '~/server/utils/studioQueue'
+
+export default defineEventHandler(async (event) => {
+  const { owner } = assertLibraryOwner(event)
+  const id = String(getRouterParam(event, 'id') || '')
+  return await retryStudioJob(owner, id)
+})
diff --git a/server/utils/studioQueue.ts b/server/utils/studioQueue.ts
index c057fff..e7d98de 100644
--- a/server/utils/studioQueue.ts
+++ b/server/utils/studioQueue.ts
@@ -492,6 +492,36 @@ export async function requestCutIn(owner: string, id: string) {
   return updated
 }
 
+export async function retryStudioJob(owner: string, id: string) {
+  const jobsNow = readJobs(owner)
+  const target = jobsNow.find(job => job.id === id)
+  if (!target) throw createError({ statusCode: 404, statusMessage: 'Queued job not found' })
+  if (target.status !== 'error' && target.status !== 'held' && target.status !== 'waiting') {
+    throw createError({ statusCode: 409, statusMessage: 'Only a failed or paused job can be retried' })
+  }
+  const updated = await mutate(owner, (jobs) => {
+    const job = jobs.find(item => item.id === id)
+    if (!job) throw createError({ statusCode: 404, statusMessage: 'Queued job not found' })
+    if (job.status !== 'error' && job.status !== 'held' && job.status !== 'waiting') {
+      throw createError({ statusCode: 409, statusMessage: 'Only a failed or paused job can be retried' })
+    }
+    job.status = 'waiting'
+    job.cutIn = true
+    job.lastError = undefined
+    job.liveJobId = undefined
+    job.pausedByUser = false
+    job.pauseAfterCurrent = false
+    job.holdForCutIn = false
+    job.updatedAt = Date.now()
+    for (const item of jobs) {
+      if (item.id !== id && item.status === 'waiting') item.cutIn = false
+    }
+    return structuredClone(job)
+  })
+  await kickStudioQueue()
+  return updated
+}
+
 function pruneDone(jobs: StudioJob[]) {
   const cutoff = Date.now() - 1000 * 60 * 60 * 24
   return jobs.filter(job => {
@@ -585,6 +615,10 @@ function pendingAlive(job: StudioJob) {
   ))
 }
 
+function isTransientComfyError(error?: string) {
+  return /still busy|COMFY_BUSY|offline|never answered|host agent|unreachable|asleep|Starting Comfy|ComfyUI \(1\)|instance picker/i.test(String(error || ''))
+}
+
 function repairStaleJobs(jobs: StudioJob[]) {
   for (const job of jobs) {
     if (job.status === 'error' && /allowIdentityRefs is not defined/i.test(String(job.lastError || ''))) {
@@ -594,6 +628,13 @@ function repairStaleJobs(jobs: StudioJob[]) {
       job.updatedAt = Date.now()
       continue
     }
+    if (job.status === 'error' && isTransientComfyError(job.lastError)) {
+      job.status = 'waiting'
+      job.liveJobId = undefined
+      job.lastError = undefined
+      job.updatedAt = Date.now()
+      continue
+    }
     if (job.status !== 'running') continue
     const live = job.liveJobId ? getJob(job.liveJobId) : undefined
     const liveBusy = live && (live.status === 'queued' || live.status === 'uploading' || live.status === 'running')
@@ -1123,10 +1164,6 @@ function remainingStudioShots(job: Job) {
   return Math.max(0, (library.chainTotal || 1) - (library.chainStep || 1))
 }
 
-function isTransientComfyError(error?: string) {
-  return /still busy|COMFY_BUSY|offline|never answered|host agent|unreachable|asleep|Starting Comfy|ComfyUI \(1\)|instance picker/i.test(String(error || ''))
-}
-
 export async function onLiveVideoSettled(job: Job) {
   const owner = job.library?.ownerKey
   if (!owner) {