Encode Diffusers Outpaint Prompt

This is CLIP Text Encode for the diffusers pipeline running underneath this pack - the README says it straight: "works as clip text encode (prompt), and specifies what to add to the image." If you've ever wired up a normal ComfyUI text-to-image workflow you already know how to use it; the only reason it's a separate node from ComfyUI's built-in CLIP Text Encode is the same reason every node in this pack is separate - it's feeding a diffusers pipeline internally, not ComfyUI's native conditioning path.

What you set: text, a multiline string - your prompt, describing what you want to appear in the outpainted area. clip takes a CLIP model. You'll use two of these nodes in a working graph: one for what you want (feeding positive on the Diffusers Image Outpaint node) and one for what you don't (feeding negative) - same pattern as any standard positive/negative conditioning pair. device and dtype are here too, defaulting to auto like the rest of the pack's nodes; there's rarely a reason to touch them.

Where the clip input comes from is worth getting right: the pack's older versions needed their own tokenizer and text encoder downloads, but a 2024 update removed that requirement - ComfyUI's own CLIP now works here directly. The easiest route is a Checkpoint Loader Simple node, which hands you a CLIP output alongside the VAE you'll need later, no separate clip selection needed. If you'd rather use a dedicated (Dual) Clip Loader instead - say you're already running one for another part of your graph - and you want to pair it with an SDXL-type model like RealVisXL_V5.0_Lightning, you'll need clip_l and clip_g files, which the README points to Comfy-Org's stable-diffusion-3.5-fp8 text encoders repo for.

Output: diffusers_conditioning, a CONDITIONING type - the tooltip describes it plainly as "a conditioning containing the embedded text used to guide the diffusion model." Wire it into either positive or negative on the main outpaint node.

Installing it: it's part of the same pack as everything else here - git clone ComfyUI-DiffusersImageOutpaint into custom_nodes (or install through the Manager), restart, and run pip install -r requirements.txt from the extension folder if dependencies didn't already get pulled in. Nothing extra to download specifically for this node - it just needs a CLIP model, which you're almost certainly loading for the rest of the workflow anyway.

What actually trips people up here isn't this node itself - it's almost always upstream. A mismatched CLIP (wrong architecture for the unet you loaded in Load Diffuser Model) will produce garbage conditioning or an outright error, so if your outpaint results look completely unrelated to your prompt, check that your CLIP source matches your model choice before you start second-guessing the prompt wording. And since this pack is SDXL-only end to end, don't reach for a Flux or SD 1.5 CLIP setup here - it's built around SDXL's dual text-encoder expectations.