Skip to content

Unify AI connections and fix GPT image generation #1436

Open

Description

@jamesgraves52

I get the "No Image models Found" Error when trying to enable model generation. I know that GPT-5.6-sol has image generation capablility can we Simple Chat configured to accept GPT-5.6-Sol as an image generation model???

Activity

  1. added
    bugSomething isn't working
    P2Priority 2: important, scheduled after P0/P1
    on Sep 7, 2026
  2. changed the title [-]Image Generation Does Not Work Or Pick up Azure Open AI GPT-5.6-SOL LLM Model[/-] [+]Unify AI connections and fix GPT image generation[/+] on Sep 8, 2026
  3. paullizer commented on Sep 8, 2026

    @paullizer
    Collaborator

    Updated implementation scope

    The agreed direction is one shared AI Connections framework, reusing the existing model-endpoint registry rather than creating a second image-only endpoint manager.

    • Configure an endpoint and its credentials once; chat and image-generation defaults reference the connection/model by stable IDs rather than copying configuration.
    • Wire chat and image generation now. Keep the existing Image action and one image-model selection; do not require a second image-backend picker.
    • Add capability-aware discovery, selection, and runtime validation. Preserve existing chat/agent/workflow/vision behavior when image-only models enter the shared registry.
    • Automatically import existing direct/APIM image settings, preserving the active model, credentials, routing, and existing connections. Do not force the irreversible chat multi-endpoint switch or publish imported image connections as new chat models.
    • Correct the GPT image-generation API path, unify generation callers, and address the coupled image-persistence and misleading-error problems. Actual Azure image-tool availability/default routing still requires confirmation against the configured resource.
    • Establish extension points for future embeddings, transcription, speech, computer use, and other capabilities; their runtime integrations and new controls are explicitly deferred.

    Priority remains P2, assigned to Paul Lizer (@paullizer). Size is L for the expanded framework, compatibility, migration, and UI scope. The implementation plan is being revised; application code has not been changed yet.

  4. moved this from Pending Evaluation to In progress in Simple Chat Roadmapon Sep 8, 2026
  5. paullizer commented on Sep 8, 2026

    @paullizer
    Collaborator

    Implementation ready for review

    The shared AI Connections framework is implemented in the working branch paullizer-gpt-image-generation for 0.261.102. This is not a deployment or a claim of live Azure availability.

    • One connection/credential registry now supports independent chat and image model defaults, capability-aware discovery/selection, and reusable operation adapters.
    • Existing image settings import automatically with active-model/auth/API/APIM preservation, scoped staged credentials, conditional settings writes, and recovery without erasing retained settings.
    • GPT image requests use Azure v1 Responses; dedicated image requests retain the appropriate Images API. Image mode does not require a usable text model. Shared image edits preserve configuration/refusal/rate-limit statuses.
    • Classic and V2 use one grouped image-model picker. Image-only models are excluded from text consumers, unverified model variants cannot inherit image-tool support, and duplicated connections retain image operation profiles.
    • Framework documentation, administrator guidance, generated documentation inventory, release notes, and the application version are updated. Embeddings, transcription, speech, computer use, and other runtime integrations remain deferred.

    Validation

    • All changed functional tests plus route policies: 324 passed, with 457 subtests passed.
    • Final Classic/V2 browser workflows: 28 passed.
    • Classic/V2 model-choice scripts, TypeScript typecheck, and production V2 build passed.
    • Documentation inventory/coverage: 7/7; site quality: 6/6.
    • Real OpenAI 1.109.1 request construction was exercised with mocked HTTP, including complete registered /api/chat image requests. Reported code-review findings were corrected and regression-covered.

    Deployment boundary

    No live Azure inference or resource provisioning was performed. Actual image-tool entitlement, regional availability, and any service-required backing deployment remain properties of the configured Azure resource and must be confirmed there. The application does not invent deployment bindings or silently switch models.

  6. moved this from In progress to In review in Simple Chat Roadmapon Sep 8, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Labels

P2Priority 2: important, scheduled after P0/P1bugSomething isn't workingenhancementNew feature or request

Type

No type

Projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions