nodes/Runware-ComfyUI-Runware
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add steps parameter to audio inference, fix empty video/audio object handling, add Creatify models, and improve VideoObject save_to method

Browse Source
This commit is contained in:
Sirshendu Ganguly
2026-01-12 12:48:19 -05:00
parent 745acacd2b
commit b2efe563b3
7 changed files with 136 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files
clientlibs/utils.js
+11 -1
View File
@@ -966,6 +966,8 @@ function audioInferenceToggleHandler(audioInferenceNode) {
const sampleRateWidget = audioInferenceNode.widgets.find(w => w.name === "sampleRate");
const useBitrateWidget = audioInferenceNode.widgets.find(w => w.name === "useBitrate");
const bitrateWidget = audioInferenceNode.widgets.find(w => w.name === "bitrate");
const useStepsWidget = audioInferenceNode.widgets.find(w => w.name === "useSteps");
const stepsWidget = audioInferenceNode.widgets.find(w => w.name === "steps");
// Helper function to toggle widget enabled state (exact same pattern)
function toggleWidgetState(useWidget, paramWidget, paramName) {
@@ -1024,6 +1026,10 @@ function audioInferenceToggleHandler(audioInferenceNode) {
if (useBitrateWidget && bitrateWidget) {
toggleWidgetState(useBitrateWidget, bitrateWidget, "bitrate");
}
if (useStepsWidget && stepsWidget) {
toggleWidgetState(useStepsWidget, stepsWidget, "steps");
}
}
function acceleratorOptionsToggleHandler(acceleratorNode) {
@@ -1748,6 +1754,10 @@ function videoModelSearchFilterHandler(videoModelSearchNode) {
"Bria": [
"bria:60@1 (Bria Video Eraser)",
],
"Creatify": [
"creatify:aurora@fast (Creatify Aurora Avatar Model API (720p))",
"creatify:aurora@0 (Creatify Aurora Avatar Model API (720p))",
],
};
const MODEL_DIMENSIONS = {
@@ -1879,7 +1889,7 @@ function videoModelSearchFilterHandler(videoModelSearchNode) {
"sync:react-1@1": "720p",
"bria:60@1": "720p",
"creatify:aurora@fast": "720p",
"creatify:aurora@0": "480p",
"creatify:aurora@0": "720p",
};
const DEFAULT_DIMENSIONS = {"width": 1024, "height": 576};
modules/audioInference.py
+115 -23
View File
@@ -5,6 +5,7 @@ import numpy as np
import librosa
import tempfile
import os
import comfy.model_management
from .utils import runwareUtils as rwUtils
@@ -72,7 +73,7 @@ class RunwareAudioInference:
"default": "URL",
"tooltip": "Output type for the generated audio"
}),
"outputFormat": (["MP3"], {
"outputFormat": (["MP3", "MP4"], {
"default": "MP3",
"tooltip": "Format of the output audio"
}),
@@ -82,6 +83,16 @@ class RunwareAudioInference:
"max": 3,
"tooltip": "Number of audio files to generate"
}),
"useSteps": ("BOOLEAN", {
"tooltip": "Enable to include steps parameter in API request. Disable if your model doesn't support steps.",
"default": False,
}),
"steps": ("INT", {
"tooltip": "Number of inference steps for audio generation. More steps generally result in higher quality but longer generation time.",
"default": 20,
"min": 1,
"max": 100,
}),
},
"optional": {
"inputs": ("RUNWAREAUDIOINFERENCEINPUTS", {
@@ -102,30 +113,102 @@ class RunwareAudioInference:
"""Main function to generate audio using Runware API"""
params = self._extractParameters(kwargs)
genConfig = self._buildGenConfig(params)
genResult = rwUtils.inferenecRequest(genConfig)
self._logResponse(genResult)
self._validateResponse(genResult)
try:
# Debug: Print the request being sent
print(f"[DEBUG] Sending Audio Inference Request:")
print(f"[DEBUG] Request Payload: {rwUtils.safe_json_dumps(genConfig, indent=2)}")
genResult = rwUtils.inferenecRequest(genConfig)
# Debug: Print the response received
print(f"[DEBUG] Received Audio Inference Response:")
print(f"[DEBUG] Response: {rwUtils.safe_json_dumps(genResult, indent=2)}")
print(f"[Debugging] Generation config: {rwUtils.safe_json_dumps(genConfig, indent=2)}")
except Exception as e:
# Re-raise the original error without modification
raise e
audioData = genResult["data"][0]
audioUrl = self._extractAudioUrl(genResult)
# Extract task UUID for polling
taskUUID = genConfig[0]["taskUUID"]
hasAudio = audioUrl is not None
hasVideo = bool(audioData.get("videoURL") or audioData.get("videoBase64Data", False))
if hasAudio:
audioObj = self._downloadAndProcessAudio(audioUrl, params["sampleRate"])
print(f"[DEBUG] Audio URL found, returning audio. Audio URL: {audioUrl}")
return (audioObj, None)
if hasVideo:
videos = rwUtils.convertVideoB64List(genResult, width=None, height=None)
print(f"[DEBUG] No audio URL in response, but video is present. Returning video.")
return (None, videos)
raise Exception("No audio or video data received from API")
# Poll for audio completion
while True:
# Check for interrupt before each poll
comfy.model_management.throw_exception_if_processing_interrupted()
# Poll for audio result
pollResult = rwUtils.pollVideoResult(taskUUID)
print(f"[Debugging] Poll result: {rwUtils.safe_json_dumps(pollResult, indent=2) if isinstance(pollResult, (dict, list)) else pollResult}")
# Check for errors first
if pollResult and "errors" in pollResult and len(pollResult["errors"]) > 0:
error_info = pollResult["errors"][0]
error_message = error_info.get("message", "Unknown error")
# Extract more detailed error info if available
if "responseContent" in error_info:
response_content = error_info["responseContent"]
# Handle both string and dict response content
if isinstance(response_content, str):
detailed_message = response_content
elif isinstance(response_content, dict):
detailed_message = response_content.get("message", str(response_content))
else:
detailed_message = str(response_content)
if detailed_message:
error_message = f"{error_message}\nProvider Error: {detailed_message}"
# Include taskUUID for debugging
task_uuid = error_info.get("taskUUID", "unknown")
raise Exception(f"Audio generation failed (Task: {task_uuid}): {error_message}")
if pollResult and "data" in pollResult and len(pollResult["data"]) > 0:
audioData = pollResult["data"][0]
# Check status directly
if "status" in audioData:
status = audioData["status"]
if status == "success":
audioUrl = self._extractAudioUrl(pollResult)
hasAudio = audioUrl is not None
hasVideo = bool(audioData.get("videoURL") or audioData.get("videoBase64Data", False))
if hasAudio:
audioObj = self._downloadAndProcessAudio(audioUrl, params["sampleRate"])
print(f"[DEBUG] Audio URL found, returning audio. Audio URL: {audioUrl}")
# Return empty video object when only audio is present (prevents errors in downstream nodes)
emptyVideoObj = rwUtils.VideoObject("", width=0, height=0)
return (audioObj, emptyVideoObj)
if hasVideo:
videos = rwUtils.convertVideoB64List(pollResult, width=None, height=None)
print(f"[DEBUG] No audio URL in response, but video is present. Returning video.")
# Extract first video object from tuple (ComfyUI expects single VideoObject, not tuple)
if len(videos) > 0:
# Return empty audio object when only video is present (prevents errors in downstream nodes)
emptyAudioObj = {
"waveform": torch.zeros((1, 1, 1)), # [batch, channels, samples]
"sample_rate": params["sampleRate"]
}
return (emptyAudioObj, videos[0])
else:
raise Exception("No video object found in response")
raise Exception("No audio or video data received from API")
# If status is "processing", continue polling
# Check for interrupt before waiting
comfy.model_management.throw_exception_if_processing_interrupted()
# Wait before next poll (split into smaller chunks to allow more frequent interrupt checks)
for _ in range(10): # 10 x 0.1 second = 1 second total
comfy.model_management.throw_exception_if_processing_interrupted()
rwUtils.time.sleep(0.1)
def _extractParameters(self, kwargs):
"""Extract and validate parameters from kwargs"""
@@ -143,6 +226,8 @@ class RunwareAudioInference:
"outputFormat": kwargs.get("outputFormat", "MP3"),
"negativePrompt": kwargs.get("negativePrompt", ""),
"numberResults": kwargs.get("numberResults", 1),
"steps": kwargs.get("steps", 20),
"useSteps": kwargs.get("useSteps", False),
"inputs": kwargs.get("inputs", None),
"providerSettings": kwargs.get("providerSettings", None),
}
@@ -170,7 +255,7 @@ class RunwareAudioInference:
"model": params["model"],
"outputType": params["outputType"],
"outputFormat": params["outputFormat"],
"deliveryMethod": "sync",
"deliveryMethod": "async",
"includeCost": True,
"numberResults": params["numberResults"],
}]
@@ -205,6 +290,13 @@ class RunwareAudioInference:
else:
print(f"[DEBUG] Skipped duration")
# Add steps parameter only if enabled
if params["useSteps"]:
genConfig[0]["steps"] = params["steps"]
print(f"[DEBUG] Added steps: {params['steps']}")
else:
print(f"[DEBUG] Skipped steps")
# Add optional parameters
if params["negativePrompt"]:
genConfig[0]["negativePrompt"] = params["negativePrompt"]
modules/utils/runwareUtils.py
+5
View File
@@ -596,6 +596,11 @@ class VideoObject:
def save_to(self, filename, **kwargs):
"""Save video to file by downloading from URL with retry logic"""
# If video URL is empty, this is a placeholder object (e.g., when only audio is returned)
if not self.video_url or self.video_url.strip() == "":
print(f"[Video Download] Skipping download - empty video URL (placeholder object)")
return False
max_retries = 10
retry_delays = [2, 5, 10, 15, 20] # Longer delays for retries with larger videos
modules/videoModelSearch.py
+2 -2
View File
@@ -92,7 +92,7 @@ class videoModelSearch:
],
"Creatify": [
"creatify:aurora@fast (Creatify Aurora Avatar Model API (720p))",
"creatify:aurora@0 (Creatify Aurora Avatar Model API (480p))",
"creatify:aurora@0 (Creatify Aurora Avatar Model API (720p))",
],
}
@@ -281,7 +281,7 @@ class videoModelSearch:
# Creatify Models
"creatify:aurora@fast": "720p",
"creatify:aurora@0": "480p",
"creatify:aurora@0": "720p",
}
MODEL_ARCHITECTURES = [
workflows/Runware_Audio_Inference_Mirelo_Provider_Settings_generate_audio.json
+1
View File
File diff suppressed because one or more lines are too long
workflows/Runware_Audio_Inference_Mirelo_Provider_Settings_generate_video.json
+1
View File
File diff suppressed because one or more lines are too long
workflows/Runware_Video_Inference_Creatify_Aurora_fast.json
+1
View File
File diff suppressed because one or more lines are too long