NC33: text2image_stablediffusion2 ExApp is enabled and polling tasks, but Assistant does not register any image generation provider

Which version of assistant are you using?

Assistant version: 3.3.0

Which version of Nextcloud are you using?

Nextcloud version: 33.0.0.16

Which browser are you using? In case you are using the phone App, specify the Android or iOS version and device please.

chrome

Describe the Bug

I set up Nextcloud Assistant image generation with AppAPI, Docker Socket Proxy and HaRP on an Unraid host. The ExApp text2image_stablediffusion2 is deployed, enabled, and actively polling /ocs/v2.php/taskprocessing/tasks_provider/next, but Assistant does not expose it as an image generation provider.

In the Assistant admin settings:

• with integration_openai enabled, only ollama appears for image generation
• with integration_openai disabled, Assistant shows: “None of your currently installed apps provide image generation functionality”

So the ExApp appears to be running, but it is not registered/usable as an Assistant image provider.

Environment

• Nextcloud version: 33.0.0.16
• Assistant version: 3.3.0
• AppAPI version: 33.0.0
• integration_openai version: 4.3.0
• ExApp: text2image_stablediffusion2 1.3.0
• Deployment: Docker / HaRP
• Host OS: Unraid
• Docker network for ExApps: custom bridge network
• Docker engine version: 27.5.1

What works

• Docker Socket Proxy works
• HaRP daemon registration works
• HaRP connection test succeeds
• ExApp is deployed and enabled
• app_api:daemon:list shows the daemon correctly
• app_api:app:list shows:
  text2image_stablediffusion2 (Local Image Generation: Stable Diffusion): 1.3.0 [enabled]
• assistant.text_to_image_picker_enabled is set to 1
• taskprocessing:task-type:set-enabled core:text2image 1 succeeds

What does not work

• Assistant does not show the Stable Diffusion ExApp as a selectable image generation backend
• When I ask Assistant to generate an image, only core:text2text:chat tasks are created, not image tasks
• Assistant either routes the request to Ollama/chat, or shows no image provider at all

Expected Behavior

text2image_stablediffusion2 should appear as a valid image generation provider in Assistant and image requests should create image-related taskprocessing jobs.

Actual behavior
The ExApp is enabled and polling taskprocessing endpoints, but Assistant does not register any usable image provider from it.

Relevant OCC output

• php occ app_api:app:list
  shows:
  text2image_stablediffusion2 (Local Image Generation: Stable Diffusion): 1.3.0 [enabled]

• php occ app_api:daemon:list
  shows both my DSP daemon and HaRP daemon, with HaRP active

• php occ config:list assistant
  shows:
  text_to_image_picker_enabled: "1"

• php occ config:app:get app_api default_daemon_config
  returns:
  harp_proxy_docker

• php occ app:list | grep -iE "assistant|app_api|openai|ollama|text2image"
  returns:

  • app_api: 33.0.0
  • assistant: 3.3.0
  • integration_openai: 4.3.0

• php occ taskprocessing:task:stats
  shows existing tasks

• php occ taskprocessing:task:list
  only shows tasks like:

  • type: core:text2text:chat
  • appId: assistant:chatty-llm

Relevant logs
The ExApp repeatedly polls:
GET /ocs/v2.php/taskprocessing/tasks_provider/next?format=json

The most suspicious log entry is:
The loading of lazy AppConfig values have been triggered by app "core"

Stack trace points into:

• OC\TaskProcessing\Manager->getPreferredProvider()
• OC\Core\Controller\TaskProcessingApiController->intersectTaskTypesAndProviders()

I also see repeated deprecated alias warnings such as:

• OCP\IServerContainer
• Request

But those look secondary and not like the root cause.

Additional notes
Disabling integration_openai removes the visible ollama entry from Assistant, but Assistant still does not recognize text2image_stablediffusion2 as an image provider. Instead it reports that no installed app provides image generation.

This suggests a registration/provider-selection problem between Assistant, AppAPI and the ExApp on NC33.

To Reproduce

1. Install and configure Nextcloud 33 with the following apps enabled:

   • Assistant
   • AppAPI
   • text2image_stablediffusion2
   • integration_openai (optional, used here with Ollama)

2. Configure AppAPI with a working Docker Socket Proxy daemon and a working HaRP daemon.

3. Verify that:

   • php occ app_api:app:list shows text2image_stablediffusion2 as enabled
   • php occ app_api:daemon:list shows the configured daemon(s)
   • php occ taskprocessing:task-type:set-enabled core:text2image 1 succeeds
   • assistant.text_to_image_picker_enabled is set to 1

4. Open Assistant administration settings in Nextcloud.

5. Go to the image generation section / unified task processing settings.

6. Observe the provider selection:

   • with integration_openai enabled, only ollama is shown
   • with integration_openai disabled, Nextcloud reports that no installed app provides image generation

7. Try to generate an image in Assistant with a prompt such as:
   Create an image of a farm at sunset

8. Check the created tasks with:

   • php occ taskprocessing:task:list

9. Observe that only core:text2text:chat tasks are created, and no image generation provider is used.

10. Check the Nextcloud log and observe repeated requests to /ocs/v2.php/taskprocessing/tasks_provider/next and exceptions related to getPreferredProvider().

Start with OC\TaskProcessing\Manager->getPreferredProvider() and OC\Core\Controller\TaskProcessingApiController->intersectTaskTypesAndProviders(), then inspect requests to /ocs/v2.php/taskprocessing/tasks_provider/next. Reproduce with the listed OCC commands and enabled apps, comparing provider registration with integration_openai enabled and disabled. Done means the Stable Diffusion provider appears in Assistant and image prompts create image-related taskprocessing jobs.