Omni-Rewriter supports Qwen-Image prompt expansion (and an optional generate client)

Hi — Omni-Rewriter is an open prompt-expansion harness (schema → deterministic validation → bounded repair → dialect render). We already ship a Qwen-Image PE profile for T2I / edit-style packing.

Scope

• PE: metadata.image_pe_profile=qwen_image_edit (and Seedream-style packing for T2I). Expand ≠ generate.
• Optional generate client: OpenAIImagesClient mapped to a public /v1/images/generations contract (for example SGLang-Diffusion’s Qwen-Image-2512 registry). We do not treat every SGLang/vLLM build as interchangeable.

Links

• Repo: https://github.com/WayneJin0918/Omni-Rewriter
• Image PE: https://github.com/WayneJin0918/Omni-Rewriter/blob/main/docs/dialects/image-pe.md
• Compatibility notes: https://github.com/WayneJin0918/Omni-Rewriter/blob/main/docs/dialects/generation-adapters.md
• Replay / PE gallery: https://huggingface.co/spaces/Wayne-King/omni-rewriter-replay

FYI only — not a request to change Qwen-Image. Happy to correct any contract misread.