Frames Concat Latent

Exactly what it says: Frames Concat Latent joins two latents into one by concatenating their frames. Feed it two latent clips (or a latent and a set of guide frames) and it hands back a single latent that runs one after the other. It's connective tissue for the pack's video and temporal workflows - the node you use when you need to assemble a frame sequence in latent space rather than after decoding.

Nothing glamorous, but if you're building video-to-video or long-sequence Wan graphs, you end up needing to splice latents constantly, and doing it in latent space (instead of decoding, concatenating images, and re-encoding) keeps everything cleaner and avoids extra VAE round-trips.

How it works

It takes two LATENT inputs and stitches them along the frame dimension: frames_0 first, then frames_1, producing one longer latent. Think of it as "append clip B to the end of clip A" while both are still latents. The result is a single LATENT that downstream nodes treat as one continuous sequence.

The inputs and outputs that matter

Dead simple - two in, one out:

• frames_0 (LATENT) - the first segment. Comes out at the front.
• frames_1 (LATENT) - the second segment, appended after the first.
• latent (LATENT, output) - the combined sequence.

Order matters: frames_0 leads. If your stitched result plays in the wrong order, swap the two inputs.

How to install it

ComfyUI Manager: search RES4LYF, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/ClownsharkBatwing/RES4LYF
cd RES4LYF
pip install -r requirements.txt

Portable ComfyUI: use the embedded pip. Restart, hard-refresh F5. No downloads.

Common issues & troubleshooting

Shape mismatch error. The two latents need to be compatible to concatenate - same spatial dimensions and channel layout. If one clip is 512×512 and the other 768×768, or they came from different model families, concatenation will fail. Match resolution and source before joining.

Order is reversed. frames_0 is the front of the result. If B is playing before A, you wired them backwards - swap the inputs.

Seam between clips. Concatenation is a hard join; it doesn't blend or interpolate across the boundary. If you see a jump where the two segments meet, that's inherent to a straight concat. Smoothing the transition is a separate step (overlap, interpolation, or an unsampling pass over the boundary), not something this node does.

It's for latents, on purpose. If you already have decoded images, you can concatenate those with a normal image-batch node. This one exists so you can splice before decoding and stay in latent space - which matters in the pack's video and guide workflows.