images: stop video's flash-attn backend leaking to image models (Z-Image attn_mask crash)

diffusers' Model.set_attention_backend() doesn't just set per-processor backends
— it ALSO flips a process-wide active backend (attention_dispatch's
_active_backend). The video path sets that to flash-attn for the Wan transformer;
image and video share the nvidia-engine process, so the global stayed flash and
leaked to the next image model. Z-Image's transformer sets no backend of its own
(passes backend=None → uses the global) and its attention is masked, so it
crashed with "`attn_mask` is not supported for flash-attn 2" → image/environment
generation 400. reset_attention_backend() clears per-processor backends but NOT
the global, so it didn't help.

Fix: restore the diffusers global backend to the env default (native/SDPA)
(a) before every image generation — bulletproof against a leaked flash backend —
and (b) in the video pipeline teardown (_free_pipeline_vram), so it can't persist
after a video pipe is freed. Masked image attention (SDPA) now always works; the
video transformer keeps its own per-processor backend.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RdMufYvtTbtGDWsiZVoXce

codai/api/images.py

@@ -967,6 +967,23 @@ async def _generate_with_diffusers(pipeline, request, global_args, http_request=
         except Exception as _ip_err:
             print(f"Warning: IP-Adapter injection failed ({_ip_err}), continuing without character refs")
 
+    # Reset the diffusers GLOBAL attention backend to the environment default
+    # (native/SDPA) before generating. diffusers' Model.set_attention_backend()
+    # ALSO flips a process-wide active backend, and the video path sets it to
+    # flash-attn — which then leaks to image transformers that don't set their own
+    # (e.g. Z-Image passes backend=None → uses the global) and crashes with
+    # "`attn_mask` is not supported for flash-attn 2". Image + video share the
+    # engine process, so restore the default here so masked image attention (SDPA)
+    # always works. Cheap + idempotent; no-op if diffusers lacks the dispatcher.
+    try:
+        from diffusers.models.attention_dispatch import (
+            _AttentionBackendRegistry, AttentionBackendName)
+        from diffusers.utils.constants import DIFFUSERS_ATTN_BACKEND
+        _AttentionBackendRegistry.set_active_backend(
+            AttentionBackendName(DIFFUSERS_ATTN_BACKEND))
+    except Exception:
+        pass
+
     try:
         result = await asyncio.to_thread(pipeline, **call_kwargs)
     except TaskCancelled:

codai/api/video.py

@@ -754,6 +754,18 @@ def _free_pipeline_vram(pipe) -> None:
                         _c.reset_attention_backend()
                     except Exception:
                         pass
+            # reset_attention_backend() clears per-processor backends but NOT the
+            # process-wide active backend that set_attention_backend() flipped to
+            # flash — restore the env default so it can't leak to a later image
+            # model (Z-Image's masked attention rejects flash-attn 2).
+            try:
+                from diffusers.models.attention_dispatch import (
+                    _AttentionBackendRegistry, AttentionBackendName)
+                from diffusers.utils.constants import DIFFUSERS_ATTN_BACKEND
+                _AttentionBackendRegistry.set_active_backend(
+                    AttentionBackendName(DIFFUSERS_ATTN_BACKEND))
+            except Exception:
+                pass
             try:
                 _unload_video_loras(pipe)
             except Exception: