Model
Trends
.ai
Model
Trends
.ai
die besten Open-Source-KI-Modelle

Load LoRA-Node in ComfyUI

Load LoRA (LoraLoader) applies a LoRA file to the MODEL and CLIP with separate strengths, so a style, character or concept can be added to any checkpoint without merging.
Node-Paket: ComfyUI core
Kategorie: Loader
Diese Seite ist noch nicht übersetzt und wird deshalb auf Englisch angezeigt.

Was Load LoRA macht

A LoRA is a small patch trained on top of a base model. Load LoRA takes the MODEL and CLIP coming from a checkpoint loader, applies the patch, and outputs new MODEL and CLIP objects. Everything downstream (prompts, sampler) then uses the patched versions. Several Load LoRA nodes can be chained to stack LoRAs.
strength_model scales how much the LoRA changes the diffusion network, strength_clip how much it changes the text encoder. Most LoRAs are made for 0.6 to 1.0 on both. Negative strengths push the model away from the concept. LoraLoaderModelOnly is the same node without the CLIP side, which is all most Flux and video LoRAs need.

Eingänge

Name
Typ
Bedeutung
model
MODEL
The model to patch, from Load Checkpoint, Load Diffusion Model or a previous LoRA node.
clip
CLIP
The text encoder to patch.
lora_name
COMBO
The LoRA file from models/loras.
strength_model
FLOAT
Weight applied to the model patch. 1.0 is the trained strength; 0 disables it.
strength_clip
FLOAT
Weight applied to the text encoder patch.

Ausgänge

Name
Typ
Bedeutung
MODEL
MODEL
The patched model for the sampler or the next LoRA.
CLIP
CLIP
The patched text encoder for CLIP Text Encode.
AD
Spar dir die Einrichtung: ComfyUI-Workflows vorinstalliert in der Cloud
BitVector führt fertige ComfyUI-Workflows auf eigenen GPUs aus. Keine Installation, keine fehlenden Nodes, keine roten Kästen. Auf dem Handy oder Laptop öffnen und in einer Minute generieren.

So verwendest du Load LoRA

  1. Put the LoRA file in ComfyUI/models/loras.
  2. Insert Load LoRA between the checkpoint loader and the sampler: model and clip in, MODEL and CLIP out.
  3. Connect the CLIP output to the CLIP Text Encode nodes, not the original checkpoint CLIP.
  4. Pick lora_name and set both strengths, starting at 0.8.
  5. Add the trigger words from the LoRA page to the positive prompt.

Einstellungen und Tipps

  • The LoRA must match the base family: an SDXL LoRA does nothing useful on SD 1.5 and a Flux LoRA fails on SDXL.
  • When stacking LoRAs, lower each strength so the total stays near 1.0 to 1.5; stacks at full strength burn the image.
  • A LoRA that only changes the model (no text encoder layers) prints "lora key not loaded" for CLIP keys; that is harmless.
  • For a quick A/B, set strength_model to 0 instead of unplugging the node; the graph stays wired.
  • Use a LoRA stacker node from a custom pack (rgthree Power Lora Loader, Efficiency LoRA Stacker) when you regularly use more than three.

Fehlerbehebung bei Load LoRA

lora key not loaded: lora_unet_... (dozens of lines)

Warum es passiert
The LoRA was trained for a different architecture than the model it is applied to, so none of its layer names match.

So behebst du es
Check the LoRA page for its base model (SD 1.5, SDXL, Pony, Illustrious, Flux) and pair it with a checkpoint of that family. A handful of unmatched keys is normal; hundreds mean the wrong family.

The LoRA has no visible effect

Warum es passiert
The CLIP Text Encode nodes are still wired to the checkpoint CLIP instead of the LoRA output, the trigger word is missing, or strength is 0.

So behebst du es
Rewire both prompt nodes to the CLIP that leaves Load LoRA, add the trigger words, and raise strength_model to 1.0 to confirm it works before tuning down.

Images are oversaturated, deep-fried or full of artifacts

Warum es passiert
Combined LoRA strength is too high, often because several LoRAs run at 1.0 together or an overtrained LoRA needs less than 1.

So behebst du es
Lower strength_model to 0.5-0.7 per LoRA and keep the stack total around 1.2 or less. Raise cfg slightly if detail disappears.

Error: Expected size for first two dimensions of batch2 tensor to be... (shape mismatch)

Warum es passiert
A LoRA of the wrong rank or family is being applied to CLIP, typically an SDXL LoRA with strength_clip above 0 on a Flux CLIP.

So behebst du es
Use LoraLoaderModelOnly for Flux and video LoRAs, or set strength_clip to 0.

AD
Diesen Workflow vom Handy aus starten
Jeder Workflow auf dieser Seite ist auf BitVector vorinstalliert, mit Modellen und Custom Nodes an Ort und Stelle. Einen auswählen, Prompt eingeben, fertig. Null Einrichtung, nichts herunterladen.

Fragen zu Load LoRA

Can I load several LoRAs?

Yes. Chain Load LoRA nodes: the MODEL and CLIP outputs of one go into the next. Order rarely matters; strengths do.

Should strength_model and strength_clip be the same?

Usually. Lower strength_clip when the LoRA hijacks the prompt too strongly, lower strength_model when the style is too heavy.

Do LyCORIS / LoCon / DoRA files work here?

Most do, in the same node. Very old formats may need the LyCORIS custom node pack.

Verwandte Nodes

Load Checkpoint
ComfyUI core
Load Checkpoint (CheckpointLoaderSimple) opens a .safetensors or .ckpt model file and hands out the three parts every workflow needs: the diffusion MODEL, the CLIP text encoder and the VAE.
CLIP Text Encode (Prompt)
ComfyUI core
CLIP Text Encode turns a text prompt into CONDITIONING using the model text encoder. One node holds the positive prompt, a second one the negative prompt.
KSampler
ComfyUI core
KSampler runs the denoising loop: it takes the model, prompts and a latent and produces the finished latent image, controlled by seed, steps, cfg, sampler, scheduler and denoise.
Load Diffusion Model
ComfyUI core
Load Diffusion Model (UNETLoader) loads a bare denoising network such as Flux, SD3.5, Wan or HunyuanVideo from models/diffusion_models, with a weight_dtype option for fp8 to save VRAM.
AD
Keine Lust mehr, Nodes zu reparieren? Lass es die Cloud machen
BitVector hält hunderte ComfyUI-Workflows installiert, aktuell und getestet auf schnellen Cloud-GPUs. Kein Python, keine CUDA-Fehler, keine VRAM-Grenzen. Läuft in jedem Browser.

Weitere ComfyUI-Nodes

ComfyUI Manager
ComfyUI-Manager
ComfyUI Manager is the extension that installs, updates and fixes custom node packs and models from inside the interface, resolves missing nodes in imported workflows and snapshots your setup.
Save Image
ComfyUI core
Save Image writes the IMAGE tensor to ComfyUI/output as a PNG, with the whole workflow embedded in the file metadata so the picture can be dragged back into ComfyUI to restore the graph.
VAE Decode
ComfyUI core
VAE Decode converts the sampled LATENT into a pixel IMAGE with the VAE; VAE Decode (Tiled) does the same in tiles for very large images.
Empty Latent Image
ComfyUI core
Empty Latent Image creates the blank latent canvas (width, height, batch_size) that text-to-image sampling starts from; dimensions must be multiples of 8 and match the model family.
Load Image
ComfyUI core
Load Image reads a picture from the ComfyUI input folder (or an upload) and outputs an IMAGE tensor plus a MASK taken from the alpha channel, the starting point for img2img, inpainting, ControlNet and IP-Adapter graphs.
Apply ControlNet
ComfyUI core
Apply ControlNet attaches a guide image and a loaded ControlNet to the positive and negative CONDITIONING with strength and start/end percentages, so the sampler follows edges, depth or pose from the guide.