FlowState Chef
The powerhouse of the Chef pipeline. This node executes a two-stage, instruction-based image editing and refinement workflow.
- An all-in-one image editing pipeline that seamlessly combines:
- A Qwen model for executing direct, instruction-based edits on input images.
- A FLUX model for high-quality refinement of the initial edit.
- Intelligent stage caching to avoid re-running stages where parameters haven't changed, saving time.
- Full control over both stages, including models, LoRAs, VAEs, samplers, and more.
Inputs
| Name | Type | Description |
|---|---|---|
chef_ingredients |
ANY |
The output from the FlowState Chef Ingredients node. |
qwen_model |
STRING |
Qwen Model: The Qwen UNET model used for the initial editing stage. |
flux_model_filetype |
STRING |
FLUX File Type: Specify whether the FLUX model is a full checkpoint or a UNET-only file. |
flux_model |
STRING |
FLUX Model: The FLUX model used for the final refinement stage. |
weight_dtype |
STRING |
Weight Datatype (DType): The data type for model weights (e.g., float16). |
qwen_optimization_lora |
STRING |
Qwen Opt. LoRA: An optimization LoRA applied only to the Qwen model. |
qwen_style_lora |
STRING |
Qwen Style LoRA: A style LoRA applied only to the Qwen model. |
qwen_style_lora_strength |
FLOAT |
Qwen LoRA Strength: The strength of the Qwen style LoRA. |
flux_style_lora |
STRING |
FLUX Style LoRA: A style LoRA applied only to the FLUX model. |
flux_style_lora_strength |
FLOAT |
FLUX LoRA Strength: The strength of the FLUX style LoRA. |
sage_attention |
STRING |
Sage Attention Mode: The type of Sage Attention patching to apply. |
qwen_clip |
STRING |
Qwen CLIP: The CLIP model compatible with the selected Qwen model. |
qwen_vae |
STRING |
Qwen VAE: The VAE used for the Qwen stage. |
flux_clip_1 |
STRING |
FLUX CLIP 1: The first CLIP model for the FLUX stage. |
flux_clip_2 |
STRING |
FLUX CLIP 2: The second CLIP model for the FLUX stage. |
flux_vae |
STRING |
FLUX VAE: The VAE used for the FLUX stage. |
tiled_decode |
BOOLEAN |
Tiled Decode: Enables tiled VAE decoding for higher resolutions to conserve VRAM. |
resolution |
STRING |
Resolution Selector: Select a preset resolution or 'Custom'. |
orientation |
STRING |
Orientation Selector: Swaps the aspect ratio of the selected preset. |
latent_type |
STRING |
Latent Type: The source for the initial latent (e.g., Empty, from Image). |
custom_width |
INT |
Custom Width: Defines the width of the output when resolution is 'Custom'. |
custom_height |
INT |
Custom Height: Defines the height of the output when resolution is 'Custom'. |
custom_batch_size |
INT |
Batch Size: The number of images to generate. |
seed |
INT |
Seed: The seed used to generate the initial random noise. |
sampling_algorithm |
STRING |
Sampling Algorithm: The sampler to use for the diffusion process. |
scheduling_algorithm |
STRING |
Scheduling Algorithm: The scheduler to use for the diffusion process. |
qwen_steps |
INT |
Qwen Steps: The number of steps for the Qwen sampling stage. |
qwen_denoise |
FLOAT |
Qwen Denoise: The denoise amount for the Qwen sampling stage. |
flux_guidance |
FLOAT |
FLUX Guidance: The guidance scale (CFG) for the FLUX stage. |
flux_steps |
INT |
FLUX Steps: The number of steps for the FLUX refinement stage. |
flux_denoise |
FLOAT |
FLUX Denoise: The denoise amount for the FLUX refinement stage. |
Outputs
| Name | Type | Description |
|---|---|---|
image |
IMAGE |
The final generated image batch after FLUX refinement. |
latent |
LATENT |
The final latent batch from the FLUX refinement process. |
Video tutorials can be found on YouTube at 🌊 FlowState Creator Suite Playlist (Coming Soon)