Merge 0f6ab9c1de into 83082a51c4

comfy_execution/jobs.py

@@ -170,6 +170,19 @@ def is_previewable(media_type: str, item: dict) -> bool:
     return False
 
 
+def is_text_preview(media_type: str, item: dict) -> bool:
+    """
+    Check if a previewable output item is textual rather than visual media.
+
+    Saved text files (SaveText's .txt/.md/.json) are real outputs but must not
+    outrank visual media when picking the job preview.
+    """
+    if media_type == 'text':
+        return True
+    filename = item.get('filename', '').lower()
+    return any(filename.endswith(ext) for ext in TEXT_EXTENSIONS)
+
+
 def normalize_queue_item(item: tuple, status: str) -> dict:
     """Convert queue item tuple to unified job dict.
 
@@ -259,8 +272,13 @@ def get_outputs_summary(outputs: dict) -> tuple[int, Optional[dict]]:
     Returns (outputs_count, preview_output).
 
     Preview priority (matching frontend):
-    1. type="output" with previewable media
-    2. Any previewable media
+    1. type="output" visual media (saved images/video/audio/3d)
+    2. any other previewable visual media (e.g. temp/preview images)
+    3. saved text file (e.g. SaveText's .txt/.md/.json)
+    4. raw text (only when the job produced nothing else previewable)
+
+    Text is kept in its own slots so node/execution order can't let a text
+    output mask a visual one (e.g. a text node that runs before an image).
 
     Text content entries (strings under 'text') are preview-only metadata,
     matching the frontend's METADATA_KEYS: they can serve as the fallback
@@ -269,6 +287,8 @@ def get_outputs_summary(outputs: dict) -> tuple[int, Optional[dict]]:
     count = 0
     preview_output = None
     fallback_preview = None
+    text_file_fallback = None
+    text_fallback = None
 
     for node_id, node_outputs in outputs.items():
         if not isinstance(node_outputs, dict):
@@ -296,8 +316,8 @@ def get_outputs_summary(outputs: dict) -> tuple[int, Optional[dict]]:
                                     'nodeId': node_id,
                                     'mediaType': media_type
                                 }
-                                if fallback_preview is None:
-                                    fallback_preview = enriched
+                                if text_fallback is None:
+                                    text_fallback = enriched
                         continue
                     # normalize_output_item returned a dict (e.g. 3D file)
                     item = normalized
@@ -314,12 +334,15 @@ def get_outputs_summary(outputs: dict) -> tuple[int, Optional[dict]]:
                     }
                     if 'mediaType' not in item:
                         enriched['mediaType'] = media_type
-                    if item.get('type') == 'output':
+                    if is_text_preview(media_type, item):
+                        if text_file_fallback is None:
+                            text_file_fallback = enriched
+                    elif item.get('type') == 'output':
                         preview_output = enriched
                     elif fallback_preview is None:
                         fallback_preview = enriched
 
-    return count, preview_output or fallback_preview
+    return count, preview_output or fallback_preview or text_file_fallback or text_fallback
 
 
 def apply_sorting(jobs: list[dict], sort_by: str, sort_order: str) -> list[dict]:

tests/execution/test_jobs.py

@@ -280,6 +280,86 @@ class TestGetOutputsSummary:
         assert preview['filename'] == 'model.glb'
         assert preview['mediaType'] == '3d'
 
+    def test_media_preview_preferred_over_text(self):
+        """A visual output wins the preview even when a text node is iterated
+        first (regression: text could mask a later temp/preview image)."""
+        outputs = {
+            'text_node': {'text': ['a caption']},
+            'image_node': {'images': [{'filename': 'preview.png', 'type': 'temp'}]},
+        }
+        count, preview = get_outputs_summary(outputs)
+        # Text is preview-only metadata and not counted; only the image counts.
+        assert count == 1
+        assert preview['filename'] == 'preview.png'
+        assert preview['mediaType'] == 'images'
+
+    def test_text_used_as_preview_when_no_media(self):
+        """Text is the preview only when the job produced no media output."""
+        outputs = {
+            'text_node': {'text': ['hello world']},
+        }
+        count, preview = get_outputs_summary(outputs)
+        assert count == 0  # text entries are not counted as outputs
+        assert preview['mediaType'] == 'text'
+        assert preview['content'] == 'hello world'
+
+    def test_media_preview_preferred_over_saved_text_file(self):
+        """A visual output wins the preview over a saved text file (SaveText),
+        even a temp/preview image iterated after the text node."""
+        outputs = {
+            'save_text': {
+                'text': ['the text'],
+                'files': [{'filename': 'ComfyUI_00001.txt', 'subfolder': '', 'type': 'output'}],
+            },
+            'preview_image': {'images': [{'filename': 'preview.png', 'type': 'temp'}]},
+        }
+        count, preview = get_outputs_summary(outputs)
+        assert count == 2  # the .txt file and the image; raw text is metadata
+        assert preview['filename'] == 'preview.png'
+        assert preview['mediaType'] == 'images'
+
+    def test_saved_media_preferred_over_saved_text_file(self):
+        outputs = {
+            'save_text': {
+                'text': ['the text'],
+                'files': [{'filename': 'ComfyUI_00001.txt', 'subfolder': '', 'type': 'output'}],
+            },
+            'save_image': {'images': [{'filename': 'result.png', 'type': 'output'}]},
+        }
+        count, preview = get_outputs_summary(outputs)
+        assert count == 2
+        assert preview['filename'] == 'result.png'
+
+    def test_mime_format_file_preferred_over_saved_text_file(self):
+        """Custom-node outputs previewable via MIME format (e.g. VHS videos
+        under arbitrary keys) rank as visual media, above saved text files."""
+        outputs = {
+            'save_text': {
+                'files': [{'filename': 'notes.md', 'subfolder': '', 'type': 'output'}],
+            },
+            'video_node': {
+                'files': [{'filename': 'clip.webm', 'format': 'video/webm', 'type': 'output'}],
+            },
+        }
+        count, preview = get_outputs_summary(outputs)
+        assert count == 2
+        assert preview['filename'] == 'clip.webm'
+
+
+    def test_saved_text_file_preferred_over_raw_text(self):
+        """With no media in the job, the saved text file (a real, counted
+        output) is the preview rather than the raw text metadata."""
+        outputs = {
+            'save_text': {
+                'text': ['the text'],
+                'files': [{'filename': 'ComfyUI_00001.txt', 'subfolder': '', 'type': 'output'}],
+            },
+        }
+        count, preview = get_outputs_summary(outputs)
+        assert count == 1
+        assert preview['filename'] == 'ComfyUI_00001.txt'
+        assert preview['mediaType'] == 'files'
+
 
 class TestHas3DExtension:
     """Unit tests for has_3d_extension()"""