DP Crazy Prompt Mixer

When you're fishing for ideas and a fixed prompt keeps giving you the same five images, this node scrambles it. You paste in some text, and it chops and reshuffles it into a new arrangement every run - same words, different order, sometimes surprising combinations. It's a variety machine, not a quality machine. Think of it as a lazy wildcard: instead of building a whole __wildcard__ file system, you drop a blob of comma-separated ideas in and let it remix them.

The output is a plain STRING, so it wires straight into a CLIP Text Encode. Nothing fancy under the hood - it's rearranging your own text, not calling an LLM or inventing new concepts. That's the honest framing: it won't make a bad prompt good, but it'll pull you out of a rut when every generation is starting to look identical.

How it works

You feed it text and pick how to slice it. The three controls:

• mixing_mode - the important one. split_by_commas treats each comma-separated phrase as a unit and shuffles those units (keeps your phrases intact, just reorders them). chunks breaks the text into groups of chunk_size words and shuffles the chunks. full_random is the chaos setting - it scrambles at the word level, so grammar goes out the window and you get genuinely weird combinations.
• chunk_size (1–10, default 3) - only matters in chunks mode; it's how many words ride together in each shuffled block.
• generation_mode - randomize gives you a fresh shuffle each run; fixed locks it so you get the same rearrangement repeatedly (useful when you found one you like and want to keep it while you change other things).

The input_prompt is an optional multiline field - that's where your source text goes.

The realistic setup

Write your prompt as comma-separated fragments - "neon city, rain, lone figure, cinematic lighting, wide shot" - set mixing_mode to split_by_commas, leave generation_mode on randomize, and queue a batch. Each image gets the same ingredients emphasized in a different order, which nudges the model's attention around without you rewriting anything. Reach for full_random when you actively want nonsense to jog something loose; skip it when you need coherent output.

How to install it

Via ComfyUI Manager: search ComfyUI-Desert-Pixel-Nodes, install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/DesertPixelAi/ComfyUI-Desert-Pixel-Nodes

then restart ComfyUI. This is a pure text node - no models, no heavy dependencies. It lives under the "DP" text category; type "DP Crazy" in the node search.

Common issues & troubleshooting

Output is grammatical garbage. That's full_random doing its job - it shuffles words, not phrases. If you want the model to still parse the prompt, switch to split_by_commas and make sure your source is actually comma-delimited.

Every run looks the same. You're on generation_mode: fixed. Flip it to randomize. (And remember your KSampler seed also has to move - if the sampler seed is fixed and the prompt shuffle is fixed, nothing changes.)

It's not adding new ideas. By design. It only ever rearranges the words you gave it. If you want the node to generate fresh concepts, that's a different tool - this pack has dedicated random-prompt generators for superheroes, vehicles, art styles and the like. The Mixer is for remixing what you already wrote.