Eric Qwen Batch to Video
Turn a flat latent batch back into frames
- latent
- latent
EricQwenBatchToVideo is the smallest node in this pack and it knows it: one input, one output, one reshape. Its entire job is to take a flat latent batch and give it back a frame dimension, the exact reverse of what the pack's Latent Cut to Batch node does.
What it does
Feed it a 4D latent [B, C, H, W] and it returns a 5D latent [1, B, C, H, W] - it inserts the frame dimension, turning your stack of layers into a single video-style latent with each layer as a frame. If the input is already 5D, it passes through.
That's the whole thing. No parameters to set, no skip_first toggle, no options. It's a shape utility, pure and simple.
When you'd bother
Qwen-Image-Layered thinks in frames. The multi-layer latent is 5D, the KSampler samples 5D, and then most of the pack's tooling wants a flat batch so it can treat each layer independently. But sometimes you go the other way - you've decoded layers, edited them, re-encoded them as a flat batch, and now you want to run them back through something frame-aware, like a video node or another multi-frame sampling pass. BatchToVideo restores that frame structure so the downstream node sees what it expects.
In short: use Latent Cut to Batch when layers are about to be decoded one-by-one; use Batch to Video when already-separate layers need to be viewed as one multi-frame latent again. They're a matched pair, and having both means you never have to hand-reshape a latent yourself.
The honest note
This is the sort of node you will forget exists until a specific workflow needs it - and then it's genuinely handy, because hand-writing unsqueeze(0) calls in a node-less way is exactly the fiddly thing ComfyUI graphs hate. It's in the pack's experimental "Native" category, so don't expect much hand-holding, but for a one-line reshape there's not much to go wrong.
Install the pack via ComfyUI Manager (search "Eric Qwen") or git clone https://github.com/EricRollei/Qwen_Layers_Diffuser_Pipeline_Comfyui into custom_nodes, then restart. No model, no VRAM, no dependencies beyond the pack's own.
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| latent | LATENT | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| latent | LATENT | — |