Prompt with Flux Guidance MXD

Standard ComfyUI text-to-image plumbing takes three nodes: encode your positive prompt, encode your negative, and then separately attach a guidance value to the conditioning. It's mechanical, and for Flux it's especially silly because Flux guidance is a per-generation number (typically 3.5) that you set in the conditioning itself, not on the sampler. Prompt with Flux Guidance MXD collapses encode + guidance into one node: type your text, plug in the CLIP, set guidance, and a ready-to-sample CONDITIONING comes out the other side.

What it actually does

Three steps under one hood:

1. Tokenize + encode your text with the connected CLIP model (the same clip.encode_from_tokens_scheduled call the core CLIP Text Encode node makes).
2. Stamp a guidance value onto the resulting conditioning via conditioning_set_values - the exact mechanism the core "Flux Guidance" node uses.
3. Hand you the conditioned CONDITIONING output, ready for a sampler.

So instead of CLIP Text Encode → Flux Guidance wiring, you get one node with a text box, a clip socket, and a guidance number. The guidance default of 3.5 is right in the community's sweet spot for Flux.1 Dev - that's where most people land for standard generations, and it's a good starting point before you start tuning toward 1–2 for more literal adherence or higher for more prompt adherence.

Inputs and outputs

• text - your prompt, multiline.
• clip - the CLIP model. The tooltip is honest: "The CLIP model used for encoding the text." For Flux.1 that means the T5 + CLIP-L text encoders that ship with the checkpoint.
• guidance - 0 to 100, default 3.5. This is Flux's own guidance, not CFG - it travels inside the conditioning and the sampler applies it accordingly.
• Output: CONDITIONING, wired to a sampler's positive input.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Maxed-Out-99/ComfyUI-MaxedOut.git

Restart, or ComfyUI Manager → search "Maxed Out" → install. No extra dependencies - it uses core ComfyUI node helpers.

Troubleshooting

• "CLIP model is None. Your checkpoint may not contain a text encoder." - you connected a checkpoint that has no usable text encoder (some distilled/single-file checkpoints drop it). Connect a proper CLIP loader output instead.
• Guidance seems to do nothing - if your sampler is CFG-based rather than guidance-aware, the value is ignored. This is built for the guidance-style models (Flux family), not the CFG world.
• "One node vs two" - if you want to keep CLIP Text Encode separate for reuse across multiple nodes, this isn't the one; the whole point is collapsing the pair. Nothing wrong with either approach, just know what you're buying.

It's a quality-of-life node through and through - the kind of "why is this two nodes anyway" compaction that makes a template workflow cleaner to read and faster to build.