Difforum · Model Catalog

Difforum Model Catalog is the pack's built-in answer to the eternal question: "which model do I even download for this?" You pick one of twelve curated recipes and it hands you the full story - a DIFFORUM_RECIPE bundle describing the model, quant, resolution, and steps for that path, plus a guide STRING with download and training instructions. It's an output node, so the guide renders as a text panel right in the UI.

The menu covers every rendering path the pack supports:

• SD1.5 animation - sd15_fast_anim, sd15_toon_anim, sd15_real_anim, plus sd15_illusion for the QR/pattern-lock trick.
• Wan 2.2 - wan22_12gb and wan22_24gb, the modern default for serious video.
• LTX - ltxv_preview, the fast preview workhorse.
• SDXL - sdxl_quality and sdxl_fast (the fast one expects a Turbo/LCM LoRA), the classic Deforum-style feedback base.
• Flux - flux_quality and flux_fast.
• SD3.5 - sd35_quality.

Why this node exists

The README's templates all carry yellow notes listing exactly which models to download and where they go - this node is that knowledge in node form, for when you're building your own graph instead of loading a template. It pairs naturally with Model Profile: Profile resolves the settings for your GPU, Catalog tells you what files to fetch. Profile says "Wan 2.2 5B GGUF Q5_K_M"; Catalog walks you to QuantStack/Wan2.2-TI2V-5B-GGUF and the unet/text_encoders/vae folders.

The DIFFORUM_RECIPE output is a real data bundle you can wire into downstream nodes (or read from), not just a text blurb - the text is the human-readable version of the same structured recipe.

Install and gotchas

One-pack install, as with every node here:

cd ComfyUI/custom_nodes
git clone https://github.com/chillithebillis/Difforum.git difforum

Restart ComfyUI ([Difforum] loaded N nodes; ComfyUI Manager: search "Difforum"). Deps: numpy.

The honest caveat: a recipe tells you what to download, it doesn't download it for you, and the models still land in your ComfyUI models/ folders by hand (checkpoints, unet, loras, etc. - the guide names the folder). Also, recipes assume you'll match the family: an SD1.5 ControlNet won't help an SDXL recipe, and the README hammers the rule that ControlNets must match their base family. If a recipe points at a GGUF, you'll also need the ComfyUI-GGUF loader - the guide mentions the loader, but you have to install that pack yourself.