CLIP debug

CLIPDebug is the most boring node in this pack, and that's the whole point. It takes a conditioning, prints what's inside it to the terminal, and hands it back completely untouched. It's a wiretap, not a processor. You reach for it when a workflow is doing something mysterious with your conditioning and you want to see the actual numbers instead of guessing.

That's a real job here, because the pack it ships in - ComfyUI-FluxRegionAttention - does its magic by editing your conditioning. RegionAttention clones the text embeddings, concatenates each region's prompt embeddings onto the base, and re-emits the whole thing as a new conditioning. When the output of that chain looks wrong, you want to know what shape that tensor took. CLIPDebug is how you look.

The inputs are exactly what you'd expect: clip (the CLIP model used for encoding the text) and condition (the CONDITIONING you're curious about). Feed it a wire, run the queue, and it prints to the console where you launched ComfyUI:

• how many items are in the conditioning,
• the shape of condition[0][0] - for Flux that's the concatenated clip_l + t5_xxl embedding tensor, which is the thing this pack extends,
• the keys inside the conditioning's extra dict (condition[0][1]), and
• the shape of pooled_output.

Then it returns the exact same conditioning on the output, so you can drop it inline in a workflow and it can't corrupt anything. It's a pure inspection node.

One honest gotcha: it accepts the clip input but never actually uses it - the code only inspects condition. Wire it if you want to be tidy, but don't expect the prints to involve the CLIP at all. And the prints go to the terminal, not the canvas. Beginners wire this up, run it, see nothing on the graph, and assume it's broken. Look at the console window. If nothing prints there either, your queue probably didn't re-run because the graph didn't change - bump a seed or hit "Run" rather than just loading.

Installing is the standard two-step:

cd ComfyUI/custom_nodes
git clone https://github.com/attashe/ComfyUI-FluxRegionAttention

then restart ComfyUI - or just search "ComfyUI-FluxRegionAttention" in ComfyUI Manager and click install. There are no model files to download and no heavy dependencies; it's a small pure-Python pack, so this node comes along for free.

That's it. It won't fix anything - it shows you what's there. But when regional conditioning is misbehaving, "what's in my conditioning" is often the first question you actually need answered.