Projectorz Parameter

Here's the situation: Stable Projectorz is a Windows app for AI-texturing 3D models, and it wants to talk to an Automatic1111-style backend. You want it to run on ComfyUI instead, because ComfyUI is the only practical way to get Flux or Qwen into the loop. So the bridge makes ComfyUI impersonate that A1111 API - and ProjectorzParameter is how the settings you twiddle in Projectorz's UI actually flow into your graph.

It's a getter, exactly like ProjectorzControlnetParameter but for the top-level generation settings instead of one ControlNet unit. You get one setting per node, as a string. The value input is the star: it defaults to !!!Autofill when executed!!! and is replaced with the real setting every time Projectorz fires a request. You don't type here, you read.

The 19 settings it can read

The dropdown covers the A1111 basics plus the highres-fix family:

• Text and model: prompt, negative_prompt, sampler_name
• Geometry: width, height, batch_size, n_iter
• Sampling: steps, cfg_scale, seed, tiling
• Refiner: refiner_checkpoint, refiner_switch_at
• Highres fix: enable_hr, hr_upscaler, hr_sampler_name, hr_scale, denoising_strength, hr_second_pass_steps

The output is a single STRING. That's the bridge's whole philosophy - everything arrives as text, and you convert it where you need a number.

How you actually wire it

The shipped sample workflow is the best tutorial you'll get. Two ProjectorzParameter nodes feed prompt and negative_prompt straight into the two CLIPTextEncode nodes. Then width and height each go through a ProjectorzStringToInt and into an EmptyLatentImage - note the KSampler in that sample is set to 15 steps at cfg 8, hardcoded, while Projectorz's own step/cfg settings stay unused. That's the key mental model: the bridge only passes through what you explicitly wire. Want steps, cfg_scale, and seed to follow Projectorz? Add nodes for them and convert them into the KSampler's widgets. Otherwise the sample's defaults rule.

For the highres-fix settings to do anything, you'd have to build actual two-pass sampling in your graph and feed denoising_strength and hr_scale into it. The bridge makes the values available; it won't build the pipeline for you. That's a real gap in the out-of-box experience, but it's also the point - you're in ComfyUI now, so the pipeline is yours.

Install and gotchas

ComfyUI Manager → search "Stable Projectorz Bridge", or:

cd ComfyUI/custom_nodes
git clone https://github.com/tianlang0704/ComfyUI-StableProjectorzBridge

Restart. No models, no extra pip dependencies.

Three things to keep in mind. One: don't edit value, it's overwritten per run. Two: remember everything's a string - seed is a string too, and it'll happily pass "2147483647" to a widget that wants an int, so route numbers through the converters. Three: the sample workflow is hardwired to one checkpoint and a specific sampler, so a workflow you build by cloning it needs your own CheckpointLoaderSimple and the ControlNet model of your choice. None of this is a bug; it's just a bridge, and bridges only carry what you connect.