Nodes/IF_VideoPrompts/IF Video Prompts 🎥🧠
ComfyUI Node

IF Video Prompts 🎥🧠

A ComfyUI node in ImpactFrames💥🎞️/LLM with 20 inputs and 4 outputs.

By if-ai·Created about a year ago·Updated about a year ago· 57
IF Video Prompts 🎥🧠
  • images
  • sequence_description
  • scene_breakdown
  • preview_image
  • negative_prompt
input_modeFrames
model_nameQwen2.5-VL-3B-Instruct-AWQ
profileHyVideoAnalyzer - Simple one line prompt
max_new_tokens512
frame_sample_count16
temperature0.70
analysis_typeFull sequence
languageEnglish
video_file
fps8.0
max_pixels262144
fallback_frame_count4
custom_system_prompt
prefix
suffix
seed-1
negative_promptNone
model_offloadYes
precisionfloat16
CategoryImpactFrames💥🎞️/LLM

Inputs (20)

NameTypeDefaultDescription
input_modeCOMBOFramesSelect input mode: use pre-loaded frames or direct video file
model_nameCOMBOQwen2.5-VL-3B-Instruct-AWQSelect the Qwen2.5-VL model to use
profileCOMBOHyVideoAnalyzer - Simple one line promptSelect a profile with predefined system prompt and rules
max_new_tokensINT5121–2048Maximum number of new tokens to generate
frame_sample_countINT161–32Number of frames to sample from entire sequence
temperatureFLOAT0.700–2Higher values increase creativity but reduce coherence
analysis_typeCOMBOFull sequenceType of analysis to perform on the video
languageCOMBOEnglishLanguage for the output
imagesoptIMAGEInput frames from a LoadVideo node
video_fileoptCOMBOSelect a video file from the input directory
fpsoptFLOAT8.00.1–60Frames per second for video processing. Higher values sample more frames.
max_pixelsoptINT2621440–921600Max pixels for video processing (0 = default)
fallback_frame_countoptINT41–16Number of frames to use in fallback mode if initial processing fails. Lower values use less VRAM.
custom_system_promptoptSTRINGCustom system prompt to override the profile
prefixoptSTRINGText to add before the generated prompt
suffixoptSTRINGText to add after the generated prompt
seedoptINT-1Random seed for generation (use -1 for random)
negative_promptoptCOMBONonePredefined negative prompt to use
model_offloadoptCOMBOYesOffload model from GPU when not in use to save VRAM
precisionoptCOMBOfloat16Model precision - lower precision uses less VRAM but may reduce quality

Outputs (4)

NameTypeDescription
sequence_descriptionSTRING
scene_breakdownSTRING
preview_imageIMAGE
negative_promptSTRING