WAN Text_Embeds Multi-Switch
Pick a prompt you already encoded
- input1
- input2
- input3
- input4
- input5
- text_embeds
WanVideoWrapper works with text in two stages: first your prompt gets encoded into WANVIDEOTEXTEMBEDS conditioning, then the sampler consumes those embeddings. That split is why this switch exists. If you're A/B-testing prompts on a video workflow, the expensive part - the text encoder pass - happens before sampling, and re-running it every time you tweak a word both costs time and changes the conditioning object. Encode each prompt once, park the results in this node, and pick between them by bypassing. The switch's entire job is routing.
The WAN Text_Embeds Multi-Switch takes up to five WANVIDEOTEXTEMBEDS inputs and returns the first one that isn't None. First-non-None is the whole algorithm, and it's also the whole UX: no widget, you choose by bypassing inputs. The README's placement rule still rules - bypassable options in the early slots, the always-on default last - because an always-enabled encode in slot 1 is never None and will shadow slots 2 through 5 forever.
A concrete shape: slot 1 is your "cinematic" prompt encode (bypassable), slot 2 is your "close-up" encode (bypassable), slot 3 is a neutral default that's never bypassed. Bypass one or two of the first slots and the survivor wins; bypass none and the default takes over. That gives you a real prompt-picker for video without touching a single cable, and without paying the encode cost more than once.
It carries the Purge_VRAM toggle (required widget, default off). On a text-embeds switch it's mostly theater - these are tensors, not models - but it's there for symmetry with the model switches in the pack, and if you ever want the whole VRAM slate wiped before the branch downstream runs, it'll do it.
Output is text_embeds, which wires into the WanVideoWrapper sampler's conditioning input.
Installing it
Ships in ComfyUI-RvTools_v2 - Manager → search ComfyUI-RvTools_v2, or clone https://github.com/r-vage/ComfyUI-RvTools_v2 into custom_nodes and restart. The WANVIDEOTEXTEMBEDS type comes from ComfyUI-WanVideoWrapper, which you need installed for any of these WAN nodes to have sockets at all - it's the implicit dependency, not in RvTools' own requirements. No model downloads for the switch itself.
The pack is unmaintained (README → ComfyUI_Eclipse); the [RvTools] suffix is from v2's rename after v1's repo vanished.
Common issues
- No text-embeds sockets - WanVideoWrapper missing. Install it, restart, then the node has something to route.
- Slot 1 always wins - an always-enabled encode in slot 1 shadows everything after it. Default last, options first.
- "Missing Input" error - all five slots bypassed or empty. Keep one wired.
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| Purge_VRAM | BOOLEAN | false | — |
| input1opt | WANVIDEOTEXTEMBEDS | — | |
| input2opt | WANVIDEOTEXTEMBEDS | — | |
| input3opt | WANVIDEOTEXTEMBEDS | — | |
| input4opt | WANVIDEOTEXTEMBEDS | — | |
| input5opt | WANVIDEOTEXTEMBEDS | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| text_embeds | WANVIDEOTEXTEMBEDS | — |