Nodes/Diztraido Nodes/Load Flux.2 Models + LoRAs
ComfyUI Node

Load Flux.2 Models + LoRAs

Flux.2's loader plus a 16-slot LoRA chain — mostly for the Klein crowd

By jadervasque·Created 2 months ago·Updated 2 months ago· 0
Load Flux.2 Models + LoRAs
    • model
    • clip
    • vae
    unet_name
    weight_dtype
    clip_name
    type
    vae_name
    lora_count0
    lora_1
    strength_model_11.00
    strength_clip_10.00
    lora_2
    strength_model_21.00
    strength_clip_20.00
    lora_3
    strength_model_31.00
    strength_clip_30.00
    lora_4
    strength_model_41.00
    strength_clip_40.00
    lora_5
    strength_model_51.00
    strength_clip_50.00
    lora_6
    strength_model_61.00
    strength_clip_60.00
    lora_7
    strength_model_71.00
    strength_clip_70.00
    lora_8
    strength_model_81.00
    strength_clip_80.00
    lora_9
    strength_model_91.00
    strength_clip_90.00
    lora_10
    strength_model_101.00
    strength_clip_100.00
    lora_11
    strength_model_111.00
    strength_clip_110.00
    lora_12
    strength_model_121.00
    strength_clip_120.00
    lora_13
    strength_model_131.00
    strength_clip_130.00
    lora_14
    strength_model_141.00
    strength_clip_140.00
    lora_15
    strength_model_151.00
    strength_clip_150.00
    lora_16
    strength_model_161.00
    strength_clip_160.00

    If you're running Flux.2 and using LoRAs, you are almost certainly on Klein. The community's verdict on the dev tier - gorgeous, but 18–24GB even quantized and slower than molasses - pushed the actual day-to-day usage onto the 4B/9B Klein variants, and that's exactly the crowd that wants a LoRA stack without building a LoRA spaghetti graph. Load Flux.2 Models + LoRAs bundles the Flux.2 loader (UNet + single CLIP + VAE) and a 16-slot sequential LoRA stack into one node. Set lora_count, pick files, dial strengths, get model/clip/vae out the far side.

    It's the Flux.2 counterpart of the pack's Load Flux.1 Models + LoRAs, and the reasoning is the same: LoRA chains in raw ComfyUI mean one LoraLoader node per LoRA, each with its own strength widgets, all wired in series. This collapses that into a panel.

    How it works

    The loader half wraps UNETLoader, CLIPLoader (type defaulted to flux2), and VAELoader, same as DiztraidoLoadFlux2Models. The LoRA half adds lora_count plus 16 slots of lora_name / strength_model / strength_clip. At execution it loads the base model, then applies the active LoRAs in slot order through ComfyUI's LoraLoader, feeding each result into the next - so slot order is application order. Empty slots and double-zero strengths are skipped. One difference from the Flux.1 variant worth noting: here the per-slot strength_clip defaults are set to 0.0 explicitly (the Flux.1 version's schema also ends up there, this one is just deliberate about it), which is the right starting point for Flux LoRAs.

    The inputs that matter

    • unet_name, weight_dtype (fp8_e4m3fn is the sensible default), clip_name (single text encoder), type (leave on flux2), vae_name (pixel_space).
    • lora_count - 0 to 16 active slots.
    • lora_1lora_16 - file pickers from models/loras/.
    • strength_model_N / strength_clip_N - defaults 1.0 / 0.0; strength_clip stays 0 for Flux.

    Outputs: model, clip, vae.

    How to install

    Comes with Diztraido Nodes. ComfyUI Manager → search "Diztraido Nodes" → install diztraido-nodes → 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 dependencies; models and LoRAs come from your existing ComfyUI/models/ tree.

    Common issues

    The same trio of gotchas as every stacker: order matters (top slot applies first), a stack of 1.0-strength LoRAs is usually too hot - start around 0.6–0.8 each - and an empty lora_name dropdown means the file isn't in models/loras/. One Flux.2-specific reality check: Klein's distilled checkpoints are guidance-distilled and behave differently from dev (the community found the stock Flux2Scheduler sigmas, built for dev, need adjusting for Klein). That's a sampler concern, not this loader's, but it means if your Klein output looks off, the fix is in the sampler schedule, not your LoRA stack.

    CategoryDiztraido/flux

    Inputs (54)

    NameTypeDefaultDescription
    unet_nameCOMBO0 options:
    weight_dtypeCOMBO4 options: default, fp8_e4m3fn, fp8_e4m3fn_fast, fp8_e5m2
    clip_nameCOMBO0 options:
    typeCOMBO28 options: stable_diffusion, stable_cascade, sd3, stable_audio, mochi, ltxv, +22
    vae_nameCOMBO1 options: pixel_space
    lora_countINT00–16
    lora_1COMBOThe name of the LoRA.
    strength_model_1FLOAT1.00-100–100
    strength_clip_1FLOAT0.00-100–100
    lora_2COMBOThe name of the LoRA.
    strength_model_2FLOAT1.00-100–100
    strength_clip_2FLOAT0.00-100–100
    lora_3COMBOThe name of the LoRA.
    strength_model_3FLOAT1.00-100–100
    strength_clip_3FLOAT0.00-100–100
    lora_4COMBOThe name of the LoRA.
    strength_model_4FLOAT1.00-100–100
    strength_clip_4FLOAT0.00-100–100
    lora_5COMBOThe name of the LoRA.
    strength_model_5FLOAT1.00-100–100
    strength_clip_5FLOAT0.00-100–100
    lora_6COMBOThe name of the LoRA.
    strength_model_6FLOAT1.00-100–100
    strength_clip_6FLOAT0.00-100–100
    lora_7COMBOThe name of the LoRA.
    strength_model_7FLOAT1.00-100–100
    strength_clip_7FLOAT0.00-100–100
    lora_8COMBOThe name of the LoRA.
    strength_model_8FLOAT1.00-100–100
    strength_clip_8FLOAT0.00-100–100
    lora_9COMBOThe name of the LoRA.
    strength_model_9FLOAT1.00-100–100
    strength_clip_9FLOAT0.00-100–100
    lora_10COMBOThe name of the LoRA.
    strength_model_10FLOAT1.00-100–100
    strength_clip_10FLOAT0.00-100–100
    lora_11COMBOThe name of the LoRA.
    strength_model_11FLOAT1.00-100–100
    strength_clip_11FLOAT0.00-100–100
    lora_12COMBOThe name of the LoRA.
    strength_model_12FLOAT1.00-100–100
    strength_clip_12FLOAT0.00-100–100
    lora_13COMBOThe name of the LoRA.
    strength_model_13FLOAT1.00-100–100
    strength_clip_13FLOAT0.00-100–100
    lora_14COMBOThe name of the LoRA.
    strength_model_14FLOAT1.00-100–100
    strength_clip_14FLOAT0.00-100–100
    lora_15COMBOThe name of the LoRA.
    strength_model_15FLOAT1.00-100–100
    strength_clip_15FLOAT0.00-100–100
    lora_16COMBOThe name of the LoRA.
    strength_model_16FLOAT1.00-100–100
    strength_clip_16FLOAT0.00-100–100

    Outputs (3)

    NameTypeDescription
    modelMODEL
    clipCLIP
    vaeVAE