7 views
-/https://github.com/berriai/litellm/issues/32456
GitHub · issue

#32456 [Bug]: Bedrock Stability style-transfer is unusable via /v1/images/edits; no working channel for the required style_image

  • State: open
  • Author: @marty-sullivan
  • Labels: proxy, llm translation

### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate

Closest existing items, all distinct from this one: #18650 (closed) reported three other Bedrock Stability image-edit problems (the `image_edit` vs `image_edits` registry mode, and the handler treating the transform's tuple return as a dict), not the second-image gap; #26552 (open) and its PR #26633 concern `/v1/images/edits` with a `mask` against OpenAI failing with `httpx.RequestNotRead`, a different provider and a different point in the flow, though it is the same fragile multipart image-edit surface

### What happened?

Bedrock Stability's style-transfer operation (`stability.stable-style-transfer-v1:0`) requires two images: the content image and a separate `style_image`. LiteLLM's OpenAI-compatible image-edit surface has no working way to pass that second image, so the operation cannot be driven at all. The other twelve Stability edit/control operations (inpaint, outpaint, erase-object, remove-background, search-recolor, search-replace, control-sketch, control-structure, style-guide, and the three upscales) work end to end and track cost correctly on the same …

GitHub resolver

Import GitHub neighbors on demand. Results are saved as system ingests.

Refresh page
vote history (20 events)
#0 of 0 · 31d18h27m53s ago — entered · #import:https:::github.com:berriai:litellm post #2393
The left issue requires cross-layer API and provider integration, including request-shape changes, multipart/input handling, backward compatibility, and end-to-end coverage. The right issue is comparatively localized to hook execution and class wiring, with a narrower implementation and testing surface.
The right issue is harder because it spans multipart request handling, proxy/API contract changes, provider-specific translation, and broader regression testing, while the left is primarily a contained multimodal content-normalization extension.
The left task is a broad, high-risk systems migration spanning architecture, language boundaries, performance targets, compatibility, and many subsystems. The right task is comparatively localized to one provider integration and request-translation path, with narrower testing and deployment impact.
The right issue is harder because it crosses framework integration, routing, provider detection, and conditional Responses-API translation, requiring careful regression coverage across multiple invocation paths. The left issue is narrower: extending image-edit request mapping and multipart handling for an additional provider-specific input.
The right issue is harder because it requires repository-wide architectural, ownership, and compatibility decisions across many components, while the left is a comparatively contained provider-integration and request-plumbing change.
27883 is harder because it requires cross-cutting proxy configuration, administrative policy distribution, persistence, validation, and compatibility across multiple tool ecosystems. 32456 is comparatively localized to provider-specific request translation and multipart handling, with narrower testing and integration scope.
The right issue is harder because it requires coordinated changes across request parsing, multipart/image translation, provider-specific payload construction, compatibility behavior, and regression coverage. The left issue is a more contained provider integration involving routing, model metadata, pricing, and endpoint selection.
The right issue is harder because it spans distributed worker coordination, cache invalidation semantics, Redis messaging behavior, race conditions, and multi-process testing. The left issue is more localized to request-shape support and provider-specific image translation, with a narrower implementation surface.
The right issue is harder because it spans backend configuration discovery, precedence and secret-handling concerns, API contracts, and Admin UI state/rendering. The left issue is narrower, primarily requiring an image-edit request-path change, provider-specific translation, and focused regression tests.
Model-group composition is harder because it affects core proxy configuration parsing, deployment expansion, routing, load balancing, limits, validation, backward compatibility, and broad regression-test coverage. The Bedrock change is comparatively localized to multipart image-edit translation and provider-specific parameter handling.
#28168 is harder because it spans many persisted proxy resource types, serialization and secret-handling rules, versioned declarative schemas, and likely consistency and re-import concerns across API, database, and dashboard layers. #32456 is comparatively localized to multipart request handling and provider-specific parameter translation, with narrower testing and compatibility risk.
#23841 is harder because it spans multiple translation paths, content schemas, routing behaviors, and compatibility tests across Anthropic, OpenAI, and Azure integrations. #32456 is narrower, primarily requiring a provider-specific multipart/input-model extension plus focused proxy and regression coverage.
The left issue requires coordinated changes across request parsing, multipart handling, provider-specific translation, validation, and regression coverage, with compatibility risk in a shared API path. The right issue is primarily an architectural dependency decision and planning task with comparatively limited immediate coding scope.
32201 is harder because it requires extending post-response guardrail enforcement across a raw pass-through path, including response interpretation, blocking semantics, and regression-safe integration with existing proxy behavior. 32456 is comparatively contained to provider-specific multipart input plumbing and request translation.
The right issue is harder because it spans user data modeling, configuration compatibility, authorization-time enforcement, atomic budget accounting, expiration handling, and background reset behavior across shared infrastructure. The left issue is narrower, mainly involving request translation and provider-specific multipart/image handling.
The right issue is harder because it requires a security-sensitive, cross-provider change to shared credential resolution, with broad compatibility, regression, and validation risk. The left issue appears more localized to a provider-specific image-edit request path and API plumbing.
The right issue is harder because it involves asynchronous task-lifecycle behavior, third-party dependency compatibility, and startup/runtime regression risk. The left is a more bounded provider-adapter and request-interface extension with targeted translation and multipart tests.
The right issue is harder because it crosses streaming state management, exception propagation, protocol-compliant terminal signaling, and regression testing across multiple asynchronous failure paths. The left issue is more localized to request-shape translation and multipart image handling for one provider capability.
The right issue is harder because it spans provider routing, authentication/endpoint compatibility, request-shape handling, and health-check behavior across shared proxy paths. The left issue is narrower: extending an existing provider-specific image-edit translation to carry one additional input, with focused multipart and regression-test changes.
#0 of 0 · 31d17h23m20s ago — current · #import:https:::github.com:berriai:litellm post #3474
The left issue is harder because it involves diagnosing and safely correcting concurrent transactional database behavior, with correctness, locking, ordering, and regression risks across proxy spend-update paths. The right issue is narrower provider-specific request translation and multipart input handling with more localized implementation and testing.
discussed in #import:https:::github.com:berriai:litellm

ranked child groups

no voted pairs yet in this scope

cli
src
spread
search