AnyAnglePromptModeReadT8

The whole node is a dropdown with two values: studio-plus-input and input-full. Those are exactly the two options on the AnyAngle Studio encoder's prompt-mode widget, and that's the point. This node is how the browser-side Studio finds out which of the two you picked.

Why it exists

AnyAngle Studio - the interactive camera and reference workbench in this pack - runs in your browser and needs values that live in the node graph. Normally that means images and text, which the pack reads with its own reader nodes. But ComfyUI also lets you convert a widget on a node into an input link, and if you've done that to the encoder's prompt_mode, the chosen value is now an input rather than a widget the frontend could just look at.

So the Studio probes for it the same way it probes for everything else: the frontend assembles a throwaway API prompt out of the upstream chain, adds one terminal node with a UUID id, POSTs it to /prompt, and reads the answer out of the ui payload - outputs[<uuid>].anyangle_text[0]. AnyAnglePromptModeReadT8 is that terminal, and the value it hands back is the mode string. It has no outputs because the ui payload is the only channel that survives from a custom node through execution history to the browser; there's no socket into a JavaScript panel. It's flagged as an output node for the same reason.

Note the shared plumbing: this node returns its value under the same UI key as AnyAngleTextReadT8. Same transport, different payload - one carries a string, one carries a combo choice.

What the two modes actually do

This is the part worth understanding, because it changes your output text substantially.

• studio-plus-input - the Studio's own generated prompt (the camera-change sentence with the image numbers) is kept, and the text you wired into the encoder is appended to it. It's the default in the shipped workflows, and it's the forgiving choice: the model gets a coherent instruction in the pack's own phrasing, plus whatever detail you wanted to add.
• input-full - your wired text is used verbatim, empty string included, and nothing is appended. The Studio's prompt is ignored. Reach for this when you're writing the whole instruction yourself, when you're driving the pack from an API workflow, or when you specifically need an empty prompt and no implicit material-usage text sneaking in.

The distinction has teeth precisely because of the empty-string case: in input-full, an empty prompt means an empty prompt. In studio-plus-input, an empty input still leaves you with the Studio's text.

Using it

You don't wire it. The frontend injects it when it reads a mode, exactly as it does for the reference and text readers - meaning the node shows up in ComfyUI's queue and history for a moment and never appears in your saved workflow. If you add one by hand you get a two-option dropdown, no outputs, and no way to use it; that's not a malfunction.

The moment it earns its place is a batch camera run. When you prepare a batch, the Studio freezes the mode along with your prompt and negative prompt, then submits every angle with that frozen configuration. Twelve angles, one consistent instruction.

Install

The node comes with the pack, so installing the pack is installing it:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-Qwen-Image-2.1-MultiAngle-T8.git ComfyUI-AnyAngle-Studio-T8
cd ComfyUI-AnyAngle-Studio-T8
python -m pip install -r requirements.txt
python install_assets.py --download-lora

Restart ComfyUI, then hard-refresh the browser (Ctrl+F5). Use the dot-free folder name shown above - a 2.1 in the directory name makes the current frontend truncate the source badge down to Comfyui-Qwen-Image-2, which reads like a failed install. Skip --download-lora if you're running base Qwen Image 2.1 with structure guides, and skip install_reconstruction.py (3.78 GB) unless you need image-to-3D. The node itself needs no models, but the Studio it serves wants a ComfyUI build with the Qwen Image 2.1 native nodes.

Where people trip

Changing the mode mid-batch does nothing. The mode is read at batch prepare time and frozen. Flip the widget after preparing and every queued view still uses the old one. Re-prepare the batch if you want the new setting.

Wrong mode, wrong prompt. When output looks like the Studio's text was thrown away, or like extra sentences appeared that you didn't write, check the mode before chasing model behaviour. input-full is a switch that disables a feature people assume is always on.

"人物参考清单与提示词不同步" is the encoder refusing to run because the Studio's stored prompt no longer matches its material list - re-apply the Studio and run again.

Reading requires execution. The read is a genuine queued job, so an upstream node that errors, is muted, or isn't executable produces a failure in the panel rather than a silent default. The last thing you want there is a silent default anyway - the mode decides who owns your prompt.