Prompts to Pipe

This is the "prompts" half of ArtVenture's pipe system - the same idea as AV_CheckpointModelsToParametersPipe, which bundles checkpoint/VAE/LoRA names into a PIPE, except this one bundles the creative side: your positive and negative text, plus an optional reference image and mask. It exists for the same reason: instead of running separate wires for prompt text, image, and mask across a big workflow, you collect them here and carry the lot as a single object.

Where it earns its keep is templated or automation-driven graphs. If you're building something an external caller feeds different prompts into every run - an API endpoint, a batch job, a reusable subgraph - having "everything about this generation's content" collapse into one wire makes the graph much easier to reason about, and easier to swap wholesale.

How it works

You type (or wire in) a positive and a negative prompt, optionally attach a reference image and mask, and the node packs all of it into a PIPE. It also accepts an existing pipe as an optional input - so you can build the full manifest across two nodes: start with AV_CheckpointModelsToParametersPipe for the model selections, feed its PIPE output into this node's pipe input, and this node extends it with the prompt and image data. One combined bundle, assembled in two steps, still one wire by the time it reaches the rest of your graph.

The inputs and outputs that matter

Two required inputs:

• positive - multiline text, default "Positive".
• negative - multiline text, default "Negative".

Three optional inputs:

• pipe - an existing PIPE to extend rather than start fresh. This is how you chain it after a model-parameters pipe node.
• image - a reference image to carry alongside the prompts (useful for img2img or inpainting-flavored pipelines built around the pipe system).
• mask - a mask to carry the same way.

The single output is a PIPE, which the pack's unpacking node - AV_ParametersPipeToPrompts - reads back into plain positive/negative strings plus image/mask, ready for a CLIP Text Encode or a sampler.

How to install it

Comes with the Art Venture pack. ComfyUI Manager: search comfyui-art-venture, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/sipherxyz/comfyui-art-venture

Restart ComfyUI. No model downloads - it's pure data bundling.

Common issues & troubleshooting

Nothing downstream sees my prompts. This node only builds the bundle; something has to unpack it. If there's no AV_ParametersPipeToPrompts (or another pipe-aware node) reading the PIPE output, the prompts just sit inside an object nobody opens. This node makes sense as one half of a pair, not standalone.

Lost the image/mask I attached. Confirm they're wired into this node's optional inputs, not left unconnected - and that whatever reads the pipe downstream actually pulls image/mask back out. AV_ParametersPipeToPrompts does; a node that only reads positive/negative won't surface them.

Chaining pipes and one branch's data disappeared. If you're extending an existing pipe from AV_CheckpointModelsToParametersPipe, make sure that node's PIPE output is actually wired into this node's pipe input - not left as a separate, second wire heading elsewhere. Two disconnected pipes don't merge; you need the chain.

More setup than a normal workflow needs. For a one-off generation, two CLIP Text Encode nodes and a plain image loader are simpler and clearer. The pipe pattern pays off in large templated or API-driven workflows where consolidating "everything about this run's content" into one wire is worth the extra node - if you're not building that, you probably don't need this.