Pipe In Context Video v2

Pipe In Context Video v2 is a video-generation context bus with a twist that tells you exactly what kind of workflow it was built for: it carries two models. The inputs model_1 and model_2 - plus clip and vae - mean this node is designed for pipelines that run two video models, typically a "first pass, cheap" model and a "second pass, better" model, or two model variants you want to A/B inside one run. If you only need one model, this node is arguably more than you need - the single-model sibling is the one to grab.

Like every node in the context family, it's an rgthree-style dict: optional base_ctx in, every connected field overrides, unconnected fields inherit, and everything comes out both on the context wire and as its own output. The video vocabulary is the familiar Wan/VideoHelperSuite set - frame_rate, width, height, frame_load_cap, skip_first_frames, select_every_nth - plus the image slots that make the two-model pattern work: images_input, images_ref_start, images_ref_end (reference frames to condition the start and end of the clip) and images_output for the finished frames. audio_input / audio_output cover audio-conditioned video, mask_1/mask_2 handle masking, and there are five images_pp_1…images_pp_5 post-processing image slots plus five any_1…any_5 catch-alls so the context can carry odds and ends. steps, cfg, seed and path round out the settings.

Note what's not here: no positive/negative conditioning and no latent. This is a "moving frames and models" bus, not a "prompt and sample" bus - the v3 and v4 versions of this node add the conditioning side if you need it.

Install. ComfyUI Manager → search "RvTools" → install ComfyUI-RvTools_v2 and restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/r-vage/ComfyUI-RvTools_v2
# restart ComfyUI afterwards

No models or keys. Manual clones: pip install -r requirements.txt (opencv-python, pynvml, piexif, Pillow, pilgram).

The catches. Two-model is a feature, but it's also a commitment: if your workflow only ever loads one checkpoint, model_2 just sits empty and this node's extra slot count is pure canvas clutter - prefer the single-model Pipe In Context Video v3 (or v4 if you also want conditioning). And remember the family caveats: the context output is pipe-typed and unpacks with a matching RvTools/rgthree-style Out node, dict keys are pack-specific, and the pack itself is unmaintained with ComfyUI_Eclipse as the README-designated successor. This one works fine today; just know you're on legacy plumbing when you inherit a workflow built around it.