Add video transcription, fix model names
This commit is contained in:
@@ -255,6 +255,17 @@ def sendImageCaption(captionText, nodeID):
|
||||
},
|
||||
)
|
||||
|
||||
def sendVideoTranscription(transcriptionText, nodeID):
|
||||
PromptServer.instance.send_sync(
|
||||
"runwareVideoTranscription",
|
||||
{
|
||||
"success": True,
|
||||
"transcriptionText": transcriptionText,
|
||||
"nodeID": nodeID,
|
||||
"widgetName": "prompt", # Target the prompt widget (matching imageCaptioning pattern)
|
||||
},
|
||||
)
|
||||
|
||||
def inferenecRequest(genConfig):
|
||||
global RUNWARE_API_KEY, RUNWARE_API_BASE_URL, SESSION_TIMEOUT
|
||||
RUNWARE_API_KEY = os.getenv("RUNWARE_API_KEY")
|
||||
|
||||
Reference in New Issue
Block a user