Load Flux.1 Models

FLUX.1 isn't one file, it's a stack. You need the diffusion model, two text encoders (the T5-XXL and the smaller CLIP-L that Flux.1 still uses), and the VAE. In a stock ComfyUI graph that's three separate loader nodes and three sets of dropdowns - the kind of boilerplate every Flux workflow starts with. Load Flux.1 Models from the Diztraido pack collapses that into a single node with one model, one clip, and one vae output. It's the loader half of the pack's "Flux.1 pipeline in four nodes" story, and the reason the pack exists.

The tradeoff, upfront: it's a convenience wrapper, not a new loading engine. Everything underneath is ComfyUI's own UNETLoader, DualCLIPLoader, and VAELoader - so it's as reliable as the native nodes, but you don't get any extra loading magic.

How it works

The node builds its widget list by asking ComfyUI for the required inputs of those three native loaders and merging them into one panel. The one meaningful change it makes is to set the DualCLIPLoader's type default to flux for you, because that's the only value that makes sense for Flux.1 anyway. When you run, it instantiates the three native loaders, hands each the fields it owns, and passes back their three outputs.

The inputs and outputs that matter

• unet_name - your diffusion model, from models/diffusion_models/. This is the big one; a full fp16 Flux.1 dev is ~24GB, so most people run fp8 or GGUF here.
• weight_dtype - default, fp8_e4m3fn, fp8_e4m3fn_fast, or fp8_e5m2. fp8_e4m3fn is the usual VRAM-saver and quality is close enough that most people can't tell; default is full precision. If you're on a 16GB card, this dropdown is your friend.
• clip_name1 / clip_name2 - the two text encoders. For Flux.1 that's t5xxl_fp8 (or fp16) and clip_l, from models/text_encoders/.
• type - the DualCLIPLoader type list. Defaults to flux; leave it there. (The other entries exist because the native loader serves SDXL, SD3, etc.)
• vae_name - default pixel_space, the VAE Flux uses (ae.safetensors).

Outputs: model (MODEL), clip (CLIP), vae (VAE) - exactly the three things everything downstream wants.

How to install

Part of Diztraido Nodes. Install via ComfyUI Manager (search "Diztraido Nodes", registry package diztraido-nodes) and restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/jadervasque/ComyUI-Diztraido.git

Restart and it's under the Diztraido flux category. No extra Python dependencies - it literally wraps ComfyUI's own loaders.

Common issues

The node downloads nothing. If a dropdown is empty, the model files aren't where ComfyUI expects them - Flux.1 dev/schnell diffusion models go in models/diffusion_models/, text encoders in models/text_encoders/, VAE in models/vae/. Two things that trip people up: the two-encoder requirement means a Flux.1 workflow won't run if you only drop the diffusion model in - you need both t5xxl and clip_l. And the license reality is worth knowing: Flux.1 dev is non-commercial under BFL's license, while the distilled schnell tier is Apache 2.0. The pack won't stop you either way; just know which one you're shipping.