📷PhotoMaker Adapter Loader from hub🤗

A bare SDXL checkpoint doesn't know anything about identity injection. This is the node that teaches it. PhotoMakerAdapter_Loader_fromhub downloads TencentARC's photomaker-v1.bin adapter from Hugging Face and merges it into the pipe that your base model loader produced - which is what turns a plain SDXL model into a PhotoMaker model that can paste a reference face into a new generation.

In the pack's chain, this sits immediately after the base loader: BaseModel_Loader_fromhub → PhotoMakerAdapter_Loader_fromhub → (optional LoRALoader) → NEW_PhotoMaker_Generation. Skip it and the generation node still runs - it just ignores your reference images entirely, which is a confusing failure mode to debug. If your output looks like plain SDXL output, check this node first.

Mechanically, from the source, it's straightforward: hf_hub_download grabs the adapter file (the default TencentARC/PhotoMaker repo, photomaker-v1.bin - a bit over a gigabyte), then the pipeline's load_photomaker_adapter() registers it with trigger_word="img", and the scheduler is swapped to EulerDiscreteScheduler (which is what PhotoMaker's official demo uses - you don't set this yourself, the node does). The trigger word is important: it tells the tokenizer where in your prompt to inject the identity embeddings. You must actually write img into your prompt after the class word - photo of a man img wearing sunglasses - or the identity never activates.

The inputs you'll touch:

• repo_id - which Hugging Face repo to pull from. The default TencentARC/PhotoMaker is the official source and the right answer 99% of the time.
• filename - defaults to photomaker-v1.bin, the official V1 weights. PhotoMaker V2 was never released as a public model, so "v1" isn't a downgrade.
• pipe - wire in the MODEL out of your base loader.

The output is a MODEL (it's the same pipe, now adapter-loaded), which continues into the LoRA loader or the generation node.

Install the pack once via ComfyUI Manager (search "PhotoMaker") or:

cd ComfyUI/custom_nodes
git clone https://github.com/ZHO-ZHO-ZHO/ComfyUI-PhotoMaker-ZHO
cd ComfyUI-PhotoMaker-ZHO
pip install -r requirements.txt

Restart, then let this node do its thing - the first run downloads the adapter, and that's it, it caches. Budget around 10GB VRAM for the whole pipeline; this is a diffusers wrapper, and it's been unmaintained since April 2024, so treat it as a stable-but-old tool rather than something to build a forward-looking stack on. If you'd rather not download from the hub, the PhotoMakerAdapter_Loader_local twin loads the same file from a path you give it.