Bump ComfyUI to 0.34.0; fix Docker Hub description sync for org tokens

.changeset/bump-comfyui-0-34.md

@@ -0,0 +1,9 @@
+---
+"worker-comfyui": minor
+---
+
+Bump the pinned ComfyUI version from 0.29.0 to 0.34.0 (current upstream latest). Workflows exported from recent ComfyUI versions use core nodes that don't exist in 0.29-era images and fail at runtime with "missing node" errors — notably every image consumed as a base by comfyui-wizard. ComfyUI's dependency floors are unchanged between 0.29 and 0.34, so the existing torch 2.11.0+cu128 and transformers<5 pins still apply, and the build-time smoke test gates startup.
+
+Also widens `allowedCudaVersions` to include 13.1 and 13.2. The image bundles its own CUDA 12.8 runtime, so any host with a driver at or beyond the 12.8 floor (driver >= 570) runs it identically — newer host CUDA versions are backward compatible. Without this, endpoints were locked out of the growing share of hosts advertising 13.1/13.2 (including most RTX 5090 and RTX PRO 6000 capacity).
+
+Also fixes the release workflow's Docker Hub description sync: the previous action authenticated via Docker Hub's legacy user-login API, which rejects organization access tokens unconditionally — it 401'd on every release since the org token rotation and aborted the job before the GitHub release was created (5.9.0's images shipped to Docker Hub, but its GitHub release was never created, so the Hub never indexed it). The step is now a direct PATCH to the modern namespaces endpoint using the same token, and runs after the GitHub release step so a description failure can never block a release again.

.github/workflows/release.yml

@@ -132,14 +132,26 @@ jobs:
       - name: Checkout
         uses: actions/checkout@v4
 
-      - name: Update description on Docker Hub
-        uses: peter-evans/dockerhub-description@v3
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_TOKEN }}
-          repository: ${{ vars.DOCKERHUB_REPO }}/${{ vars.DOCKERHUB_IMG }}
-
       - name: Create GitHub release
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         run: gh release create "${{ needs.release.outputs.version }}" --generate-notes
+
+      # peter-evans/dockerhub-description uses the legacy Hub API
+      # (/v2/users/login + /v2/repositories/...), which rejects organization
+      # access tokens with 401/403 regardless of scopes. The namespaces
+      # endpoint accepts the org token as a Bearer token directly.
+      # Runs after the GitHub release so a description failure can't block it.
+      - name: Update description on Docker Hub
+        env:
+          DOCKERHUB_TOKEN: ${{ secrets.DOCKERHUB_TOKEN }}
+          DOCKERHUB_REPO: ${{ vars.DOCKERHUB_REPO }}
+          DOCKERHUB_IMG: ${{ vars.DOCKERHUB_IMG }}
+        run: |
+          jq -n --rawfile readme README.md '{full_description: $readme}' \
+            | curl -sSf -X PATCH \
+                -H "Authorization: Bearer ${DOCKERHUB_TOKEN}" \
+                -H "Content-Type: application/json" \
+                -d @- \
+                -o /dev/null \
+                "https://hub.docker.com/v2/namespaces/${DOCKERHUB_REPO}/repositories/${DOCKERHUB_IMG}"

.runpod/hub.json

@@ -9,7 +9,7 @@
     "containerDiskInGb": 20,
     "gpuIds": "ADA_24",
     "gpuCount": 1,
-    "allowedCudaVersions": ["12.8", "12.9", "13.0"],
+    "allowedCudaVersions": ["12.8", "12.9", "13.0", "13.1", "13.2"],
     "env": [
       {
         "key": "COMFY_ORG_API_KEY",

Dockerfile

@@ -5,7 +5,7 @@ ARG BASE_IMAGE=nvidia/cuda:12.8.1-cudnn-runtime-ubuntu24.04
 FROM ${BASE_IMAGE} AS base
 
 # Build arguments for this stage with sensible defaults for standalone builds
-ARG COMFYUI_VERSION=0.29.0
+ARG COMFYUI_VERSION=0.34.0
 ARG CUDA_VERSION_FOR_COMFY=12.8
 ARG ENABLE_PYTORCH_UPGRADE=false
 ARG PYTORCH_INDEX_URL

docker-bake.hcl

@@ -11,7 +11,7 @@ variable "RELEASE_VERSION" {
 }
 
 variable "COMFYUI_VERSION" {
-  default = "0.29.0"
+  default = "0.34.0"
 }
 
 # Global defaults for standard CUDA 12.8.1 images