Hunyuan Batch Image to 3D

What it is

Hunyuan Batch Image to 3D converts a whole folder of images to 3D models in one queue press, instead of running Hunyuan Image to 3D over and over by hand. The pack's own README calls this the "easiest" way to batch - point it at a folder, hit queue once, walk away, come back to a folder full of GLBs and a summary of what happened. If you've got a set of product shots, character turnarounds, or asset references you want converted en masse, this is the node built for exactly that instead of the single-image one.

Same underlying mechanism as the rest of the pack: every image in the batch gets sent to Tencent Cloud's Hunyuan 3D Global API v3 and billed per generation. Batch just automates the looping - it doesn't make the individual conversions any cheaper or faster, so budget accordingly before you point it at a folder of 50 images.

How it works

The node scans input_folder for files matching file_pattern, then works through them one at a time - uploading each to Tencent's API, waiting up to max_wait_time for that image's job to finish, saving the result into output_folder, and moving to the next. It's sequential, not parallel, which matters for how long a big batch actually takes: the README's own guidance is to process 5–10 images at a time and expect roughly 2–5 minutes per image, so a batch of ten can realistically run 20–50 minutes end to end.

Inputs and outputs

• input_folder - where the source images live, relative to your ComfyUI directory (default input).
• output_folder - where the resulting GLBs get written (default batch_output).
• file_pattern - a glob pattern for which files to pick up (default *.png) - change it if your source images are JPGs or you want to scope it to a subset.
• max_images - a hard cap on how many files get processed in one run, from 1 to 100 (default 10). This is your safety rail against accidentally kicking off a huge, expensive batch from a folder you forgot was full.
• enable_pbr, face_count, generate_type, polygon_type - the same generation settings as the single-image node, applied uniformly across the whole batch. There's no per-image override here; every image in the run gets the same face count, PBR toggle, and mesh style.
• max_wait_time - per-image polling ceiling (60–3600s, default 600), same as elsewhere in the pack.
• config - the HUNYUAN_CONFIG from a Hunyuan Config node. Required, same credentials used for every image in the batch.

Outputs are batch_summary (a string report of what ran, what succeeded, what didn't) and model_paths (a string listing the GLB paths produced). Feed model_paths into a Hunyuan Batch Model Preview node to step through the results in ComfyUI.

Install

Via ComfyUI Manager: search "Hunyuan 3D", install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/exedesign/Hunyuan-3D-v3.git
cd Hunyuan-3D-v3
pip install -r requirements.txt

The README also documents two alternative batch approaches - a standalone Python script (batch_hunyuan3d.py) and just manually re-running the single-image node per file - but this node is the one-click version of the same idea, wired directly into the graph.

Troubleshooting

A ResourceInsufficient error mid-batch means your Tencent Cloud credits ran out partway through, not that the node broke - check batch_summary to see how far it got before it started failing, top up the account, and re-run against the remaining images. Keep an eye on your API credits generally during a batch: at $0.10–0.60 per image, a batch of even 20 images can add up fast, and there's no built-in cost estimate before you queue it - max_images is your only real guardrail. And because this can genuinely run for tens of minutes across a full batch, if you're on a platform with its own job timeout - comfy.icu's default is 5 minutes per run - you'll need to raise that ceiling well beyond the default before a real batch has any chance of finishing inside it.