🐸 Load: Model + CLIP + VAE

Frog Loader is the ComfyUI equivalent of deleting three nodes before you even start. A stock text2img graph opens with a Checkpoint Loader, a separate CLIP Loader, and a VAE Loader - three dropdowns, three sets of outputs, three wires to route. This node collapses all of it into one: pick a diffusion model, a text encoder, and a VAE from three dropdowns on a single node, and out come MODEL, CLIP, and VAE from one connection point. If you've ever stared at a workflow and asked why the model needs to be "loaded" three times, this is the answer: it doesn't. One loader, three outputs, done.

It's the natural entry point for the whole Frog Node Pack, which is built around the idea of fewer, denser nodes instead of a dozen thin ones. The pack leans hard toward Anima - a 2B flow-matching anime model on Cosmos-Predict2 with a Qwen3 text encoder - and this loader's dropdowns are where that stack gets wired. There's nothing Anima-specific about the node itself; it loads any checkpoint, CLIP, and VAE ComfyUI can see. It's just that the pack's example workflows assume you're using it.

The inputs that matter

Four of the five inputs are dropdowns:

• diffusion_model - the checkpoint. This is your base model, the thing that does the actual denoising.
• clip_encoder - the text encoder that turns your prompt into conditioning.
• vae_model - the VAE that decodes latents into pixels. On Anima, that's the Qwen-Image VAE.
• Turbo_Lora - an optional LoRA slot, defaulting to "Empty". This is where the Anima Turbo LoRA goes if you're running the packed-in FrogFlow example.
• lora_strength - the multiplier for that LoRA (1.0 = as trained). Ignored while Turbo_Lora is Empty, so don't worry about it until you actually pick a LoRA.

What comes out

Three outputs, and they're exactly what the names say: MODEL, CLIP, and VAE. MODEL goes to your sampler, CLIP to your text encoder node, and VAE wherever you decode. The pack's own 🐸 CLIP Text Encode and 🐸 KSampler are wired to expect these, but any standard ComfyUI node accepts them too - this is not a custom pipe type, just three standard wires leaving one node.

Installing it

The whole pack installs in one step. In ComfyUI Manager, search for "Frog Node Pack" (or RabbitThatIsPink/FrogNodePack) and install. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/RabbitThatIsPink/FrogNodePack

Restart ComfyUI, hard-refresh the browser with Ctrl+Shift+R, and search the node list for "🐸". One genuine relief: requirements.txt declares no extra dependencies - the only library used is Pillow, which ships with ComfyUI. No pip, no model downloads, no missing-module roulette. The models themselves are your job, and if you want the example workflows to run, you'll need the Anima Turbo LoRA from Civitai, which the README flags as a required download.

Where people trip

The most common mistake is expecting this node to apply your LoRA the way a separate LoRA Loader would. It won't - only the single Turbo_Lora slot gets applied, and only at lora_strength. If you stack a dozen LoRAs on your Anima base, that's what the 🐸 Library's LoRA stack system is for, not this loader. Also note there's a sibling, 🐸 Load: Model + CLIP + VAE + Name, that adds a model_name string output for auto-loading quality prompts by checkpoint; grab that one instead if you're building a per-model prompt library. FrogLoader is the plain one, and that's exactly what most workflows need.