-
-
Notifications
You must be signed in to change notification settings - Fork 4
🎉 Releases
This page documents the major milestones and releases for the FlowState Creator Nodes project.
Date: 2025-10-01
-
Added Tiled Decoding to WAN Studio to help the decoding bottleneck with larger videos.
-
Updated the Chef stage execution pipeline to make the loading a little more efficient.
Date: 2025-09-30 Introducing the 🌊👩🏻🍳 FlowState Chef and 🌊🥗 FlowState Chef Ingredients nodes! This is a revolutionary two-stage pipeline designed for powerful, instruction-based image editing and refinement. It combines the advanced image editing capabilities of Qwen with the high-quality refinement of FLUX into a single, cohesive, and incredibly efficient workflow.
-
Instruction-Based Image Editing: Use the new dual-prompt system to give direct commands to the Qwen model (e.g., "place the woman on the left, change the man's shirt to blue") and then use a separate, descriptive prompt to guide the final FLUX refinement.
-
Multi-Image Composition: The Chef Ingredients node allows you to input up to four separate images, which are automatically stitched together into a single composition, making it easy to create scenes with multiple subjects.
-
Intelligent Stage Caching: The Chef node is built for efficiency. It intelligently detects which parameters have changed between runs. If you only adjust a setting in the FLUX refinement stage, the node will skip re-running the initial Qwen edit, saving significant generation time.
-
Total Creative Control: As with all FlowState nodes, you have granular control over every aspect of the pipeline, including models, LoRAs, samplers, schedulers, and advanced settings for both the Qwen and FLUX stages.
Date: 2025-09-22
Introducing 🌊🎬 FlowState WAN Studio, the new flagship node for the FlowState Creator Suite! This is our most advanced node yet, designed to be an all-in-one, comprehensive pipeline for generating high-quality video. It encapsulates a complex, multi-stage workflow into a single, convenient, and powerful interface, saving you time and graph complexity.
-
Dual-Model Sampling Pipeline: Utilize two separate UNET models in a sequential workflow. The node intelligently splits the sampling steps, allowing you to use one model for the initial high-noise steps and a second, specialized model for refining low-noise details.
-
Advanced LoRA & Model Patching: Get granular control over your model augmentations. Apply separate optimization LoRAs to the high-noise and low-noise stages, patch in an overall style LoRA, and enable Sage Attention, all from within the node.
-
Flexible Video Sizing: Easily define your output resolution. Choose from a list of pre-selected aspect ratios, enter custom dimensions, or have the node automatically inherit the resolution from an optional starting frame.
-
Complete, Self-Contained Workflow: WAN Studio handles the entire process from start to finish. It manages loading models, encoding prompts, creating the initial latent, patching augmentations, running the two-stage sampling, and decoding the final video.
-
Seamless Image-to-Video: By providing an image to the optional starting_frame input, you can guide the video generation process, turning a static image into a dynamic clip with ease.
Date: 2025-09-17
Updates to the System Initialization & model loading/patching routines, as well as to the import pipelines.
- Makes the System Initialization ~0.02s faster.
- More sophisticated model patching strategy results in better model persistence, which leads to fewer model loads.
Date: 2025-09-16
Introduced a more flexible strategy for checking the availability of KJNodes, due to varying naming schemas, depending on how the use installed KJNodes.
Date: 2025-09-16
Now added Flux LoRA styling for the 🌊🚒 FlowState Flux Engine. Simply select model & strength, or select none.
Date: 2025-09-16
The official wiki documentation for the project was created, providing users with a central place to find information about the nodes and their usage. The Nodes.md file was also updated with detailed input/output information.
Date: 2025-09-16
This update brought checkpoint integration to the 🌊🚒 FlowState Flux Engine, allowing for use of bundled checkpoints (e.g., Flux Dev.1 fp8). It also included new media for the Comfy Registry.
Date: 2025-09-15
A significant upgrade to the 🌊🚒 FlowState Flux Engine, integrating Sage Attention for more precise and context-aware image manipulation.
Date: 2025-09-15
The initial release of the 🌊🚒 FlowState Flux Engine, a powerful tool for transforming and manipulating latent images. This release laid the groundwork for future integrations and features.
Date: 2025-09-15
The first major node release since the clean slate initiative. The 🌊🌱 FlowState Latent Source node provides a stable and reliable way to generate latent images for your workflows.
Date: 2025-09-14
This commit marked a fresh start for the repository, renewing its focus and introducing a new direction for the project. The initiative cleared out old work to make way for a new set of powerful and efficient nodes.