[PSA-Team-source]

Summary

Adds UberLLM as an external image provider, so one API key reaches image models from several makers in Invoke: Z-Image Turbo, FLUX.2 Klein 4B, FLUX.2 Pro, Qwen Image 3, Seedream 4.5, Krea 2 Turbo, Ideogram V4 and Nano Banana 2.

• UberLLMProvider calls POST /v1/images/generations with response_format: "b64_json". UberLLM answers image requests synchronously, so there is no submit/poll/download step. Its errors ({"error": {"message"}}) are surfaced as-is; 429 becomes a rate-limit error with Retry-After.
• external_uberllm_api_key / external_uberllm_base_url settings, provider registration, uberllm_image_generation node, and the node mapping in both frontends plus a key card.
• Eight txt2img starter models. Every UberLLM image model takes the same five size values (1024x1024, 1024x768, 768x1024, 1536x1024, 1024x1536), so each starter declares them as aspect_ratio_sizes and ExternalGenerationService snaps any request to one of them (e.g. a 1368x768 preset goes out as 1536x1024). This is also why the existing OpenAI provider with a base URL override is not a good fit: it sends the raw width/height and only ships OpenAI starter models.
• Regenerated openapi.json / schema.ts and docs/src/generated/settings.json; added UberLLM to the API Keys provider list in the models guide.

Mirrors #9476 (Atlas Cloud), including the follow-ups there: node types registered in both frontends, and a test that goes through ExternalGenerationService.

Related Issues / Discussions

None.

QA Instructions

• uv run --no-sync pytest tests/app/services/external_generation/test_uberllm_provider.py — 16 passed.
• uv run --no-sync pytest -n 4 tests/app/services/external_generation tests/app/invocations/test_external_image_generation.py tests/app/invocations/test_frontend_external_canvas_graphs.py tests/app/routers/test_app_info.py tests/app/routers/test_starter_model_is_installed.py tests/app/routers/test_model_manager_authorization.py tests/backend/model_manager/test_external_api_config.py tests/backend/model_manager/test_starter_models.py tests/backend/model_manager/test_starter_models_package.py tests/test_config.py tests/app/util/test_custom_openapi.py tests/app/test_api_dependencies.py — 754 passed.
• uv tool run ruff@0.11.2 check and format --check on the changed Python files — clean.
• Regenerated with the commands behind make frontend-openapi, make frontend-typegen and generate-docs-data (scripts/generate_docs_json.py); a second run gives no diff; pnpm lint and pnpm test in invokeai/frontend/api pass.
• node --test scripts/mock-backend-fixtures.test.mjs in webv2 — 51 passed (provider list and count updated).
• Not run: webv1/webv2 tsc/eslint and a browser check of the new key card (no full frontend install on this machine). The frontend changes are one map entry each, plus the key card entry.
• Live: one generation through ExternalGenerationService + UberLLMProvider with a 1368x768 request (snapped to 1536x1024), and one with an invalid key, which fails with status 401: Invalid, revoked, or expired API key.

Tested with venice/z-image-turbo: one 1536x1024 PNG returned in 7.5 s, usage.cost_usd 0.0079.

Review

No material findings. Limits: txt2img only (UberLLM's edit models are separate model ids, so img2img is left for a follow-up); no seed or negative prompt, since the models' request schemas do not take them.

Compatibility / Rollout

Additive: two new optional settings and one new invocation. Existing configs and graphs are unchanged.

Checklist

• The PR has a short but descriptive title, suitable for a changelog
• Meaningful regression coverage added / updated where needed; obsolete tests/code removed
• Persisted-state and API changes include required migrations / compatibility validation
• Relevant performance/efficiency opportunities considered; material claims have evidence
• Material review findings resolved and relevant checks rerun
• Documentation added / updated (if applicable)
• Updated What's New copy (if doing a release after this PR)

Model prices (below list price): https://uberllm.dev/models

I work on UberLLM.

---

^{Open in [code]smith · Need help on this PR? Tag @codesmith-bot with what you need.}