Get IPAdapter Data (Texturaizer)

IP-Adapter reads a reference image and pushes the model toward matching its style or content without any training - you hand it a picture instead of a LoRA. This node retrieves whatever IP-Adapter setup Blender configured for the render: up to two reference images, their individual weights, which preset to use, and a plain on/off toggle, all bundled from Blender's export instead of you rebuilding it by hand in ComfyUI every time.

Worth knowing before you wire this up

IP-Adapter is genuinely useful, but it's also squarely an SD 1.5/SDXL-era tool at this point - Tencent shipped its last variant in January 2024, the community's ComfyUI implementation (cubiq's ComfyUI_IPAdapter_plus) is in maintenance mode, and nothing in the IP-Adapter line loads on Flux or newer architectures. If Blender's checkpoint selection (via Texturaizer_GetModelName) is pointed at something newer, this data won't have anywhere useful to go - check what base model you're actually running before assuming an IP-Adapter setup that "isn't working" is a ComfyUI-side bug.

How it works

Same read as the rest of the pack - directory_optional, the addon's global directory if left blank, or data_optional for an already-loaded dictionary. What it pulls out specifically is the IP-Adapter block of Blender's export: two independent reference image slots (Texturaizer supports up to two at once, each weighted separately), a preset choice, and an enable flag.

The outputs that matter

• ip image 1 / ip1 weight and ip image 2 / ip2 weight - up to two reference images, each with its own strength.
• preset - one of LIGHT - SD1.5 only (low strength), STANDARD (medium strength), VIT-G (medium strength), PLUS (high strength), PLUS FACE (portraits), or FULL FACE - SD1.5 only (portraits stronger). These are the same named presets from cubiq's IPAdapter Unified Loader - PLUS FACE and FULL FACE are the face-focused variants people reach for on portraits (community shorthand: "ip adapter plus full face... gets you 80% there"), while LIGHT and FULL FACE are flagged SD1.5-only right in their own names, so don't expect them to behave sensibly on an SDXL checkpoint.
• use IP - a boolean; the on/off switch for the whole feature.
• IP data - a dictionary bundling the above, presumably meant for a node further downstream (or a custom IPAdapter apply node) that wants the whole configuration in one object rather than as separate wires.
• data_hash - the standard debug-only fingerprint.

Wire the image and weight pairs, plus preset, into whichever IP-Adapter loader/apply nodes you're using (from ComfyUI_IPAdapter_plus or equivalent) - this node reads the configuration, it doesn't run IP-Adapter itself.

Installing it

ComfyUI Manager → search Texturaizer → Install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/LatentSpaceDirective/ComfyUI-Texturaizer

then restart ComfyUI. Note this node only reports the configuration - you still need an actual IP-Adapter implementation pack (like ComfyUI_IPAdapter_plus) installed separately to do anything with the values it returns, plus the IP-Adapter and CLIP vision model files that pack requires.

Where people get burned

Two things: check use IP before debugging anything else - if it's off, both image slots are irrelevant no matter how they're configured. And match preset to your actual base model - picking LIGHT or FULL FACE against an SDXL checkpoint, or any preset at all against a Flux-class model, is reaching for a tool that architecturally doesn't apply there.