Qwen2_5_OmniBatchImage

This is the pack's quiet utility node, and it solves a real annoyance. Qwen2_5_OmniBatchImage takes a handful of images and bundles them into a single BatchImage output that QWEN2_5_OMNI_Run accepts on its BatchImage input. Which matters more than it sounds: asking the Omni model to compare or summarize across several images in one prompt - "which of these two character designs is more consistent?" - is exactly the job these models are good at, and this node is the ergonomic way to hand it the evidence.

How it works

The clever bit is that it has no fixed inputs at all. Check the node's schema and required is empty - the inputs are added dynamically by the pack's JavaScript as you connect images. Wire in a first image and an input appears; wire a second and another one appears. Under the hood each input is saved to ComfyUI's temp directory as a PNG and turned into a file:// URI, and the node returns the list. The run node then treats each one as a separate image in the conversation.

Output is a single BatchImage that goes straight into the run node's BatchImage socket. That's the entire loop: batch the images, ask your question, read the answer.

When to reach for it (and when not to)

The run node's plain image input already handles an image tensor with multiple frames - if you're loading one folder's worth of frames through a batch loader, you don't need this node at all; just feed the tensor in and the run node will split it into multiple images itself. Where Qwen2_5_OmniBatchImage earns its keep is when your images come from different sources - a couple of LoadImage nodes, some outputs from different branches of the graph - and you want them all in one prompt as distinct items. That's a genuinely nicer workflow than hunting for a node that concatenates tensors.

Practical notes

• It saves PNGs to ComfyUI's temp directory each run. Lots of batch runs mean lots of temp files; ComfyUI cleans those up on restart, so it's not something to lose sleep over.
• Like every node in this pack, it doesn't do anything alone - it exists to feed QWEN2_5_OMNI_Run, which also needs the model loaded by DownloadAndLoadQWEN2_5_OMNIModel.
• Install is the same as the rest of the pack: ComfyUI Manager (search "ComfyUI-Qwen2_5-Omni") or git clone into custom_nodes plus pip install -r requirements.txt, then restart.

It's a tiny node and it won't make headlines, but if you've ever wanted to line up several images in front of a local model and say "tell me what's different," this is the piece that makes that a five-minute workflow instead of a file-hoisting project.