Initial SQL-only 1C adapter baseline

registry/model-cards/qwen-image-edit.yaml

@@ -0,0 +1,32 @@
+id: qwen-image-edit
+name: Qwen Image Edit
+type: image-diffusion-model
+task:
+  - image-editing
+  - inpainting
+  - image-generation
+language:
+  - ru
+  - en
+  - multilingual
+source: huggingface
+upstream_id: Qwen/Qwen-Image-Edit
+upstream_url: https://huggingface.co/Qwen/Qwen-Image-Edit
+license: apache-2.0
+status: candidate
+created_at: 2026-06-19
+storage_path: /models/image/qwen-image-edit
+format: diffusers
+quantization: bf16
+context_length: null
+vram_required_gb: 24
+parameters: 20B
+base_model: Qwen/Qwen-Image
+adapter_for: null
+datasets: []
+eval_suites: []
+deployment:
+  target: docker-gpu.cin.su
+  runtime: transformers
+  served_model_name: qwen-image-edit
+notes: "Heavy image editing candidate with stronger text/rendering/editing capabilities than SDXL. On 2026-06-20 it loaded as QwenImageEditPipeline in about 31s on RTX 4090 with CPU offload, but a 512x512 1-step edit did not finish within 1800s. Keep it as an experimental/manual-switch model until a quantized runtime or larger GPU is available."