Vantage-DreamOmni2

TextEncodeDreamOmni2

Produces: - enhanced prompt (STRING) - list of image tensors (IMAGE_LIST) in ComfyUI tensor format [B,H,W,C] - fused multimodal embedding (TENSOR) from Qwen2.5-VL (final hidden states) This ensures we carry both the text and the VLM fused embedding downstream.

DreamOmni2 · 4 entradas · 2 parámetros · 1 salida

VantageAdaptiveImageGrid

Takes a list of differently sized images and creates an adaptive grid. Use CombineImagesList before this node to avoid forced resizing. match_image_size = False → uniform grid layout (equal cell sizes) match_image_size = True → adaptive flow (aligns by average row height) canvas_width / canvas_height → defines final grid size (keeps aspect ratio per image, no cropping)

Vantage/Tools · 4 entradas · 3 parámetros · 1 salida