ComfyUI Node
Flux Panorama Loader
A ComfyUI node in DiT360Plus/pipeline with 8 inputs and 1 output.
Flux Panorama Loader
- pipeline
◄model▾►
◄base_pipelineblack-forest-labs/FLUX.1-dev►
◄lora_id►
◄dtypefloat16►
◄enable_vae_tilingtrue►
◄enable_vae_slicingtrue►
◄cpu_offloadmodel►
◄balanced_offload_gb8.0►
CategoryDiT360Plus/pipeline
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| model | COMBO | FLUX model from your ComfyUI diffusion_models folder. | |
| base_pipeline | COMBO | black-forest-labs/FLUX.1-dev | HuggingFace repo for pipeline components (text encoders, VAE, scheduler). |
| lora_id | STRING | Optional HuggingFace repo ID or local path for LoRA weights. Leave empty for no LoRA. | |
| dtype | COMBO | float16 | Model precision. float16 uses less VRAM, bfloat16 may be more stable. |
| enable_vae_tiling | BOOLEAN | true | Enable VAE tiling to reduce VRAM usage during encode/decode. |
| enable_vae_slicing | BOOLEAN | true | Enable VAE slicing for memory efficiency. |
| cpu_offload | COMBO | model | CPU offload strategy. 'model' moves whole models to CPU when idle (fast, best for 1024). 'balanced' splits transformer across GPU/CPU (good for 1536-2048). 'sequential' moves one layer at a time (slowest, lowest VRAM). 'off' keeps everything on GPU. |
| balanced_offload_gb | FLOAT | 8.01–48 | GPU memory budget (GB) for transformer layers in 'balanced' mode. Only used in balanced mode. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| pipeline | DIT360_PIPELINE | — |