v0.19.2
Latest update patches a critical OOM bug in quantized models and introduces new JSON and text generation controls.
The open-source project ComfyUI, maintained by Comfy-Org, has rolled out its latest incremental update, version 0.19.2. This release is primarily focused on stability and quality-of-life improvements for its massive user base, evidenced by its 109k GitHub stars. The headline fix addresses a significant Out-Of-Memory (OOM) regression that occurred in the `_apply()` function when performing inference with quantized models—a common technique to run large models on consumer hardware. This patch, contributed by a new developer, ensures more reliable and efficient generation sessions.
Beyond critical bug fixes, the update expands the toolkit available to workflow creators. A new 'JsonExtractString' node, added by the project's lead developer 'comfyanonymous,' provides a structured way to parse and extract data from JSON within complex node graphs. Another notable addition is the ability to disable the default template in text generation nodes, offering users more raw control over prompt engineering. The release also includes minor partner-node updates, like adding a 1080p resolution option for the SeeDance 2.0 model, showcasing the ecosystem's continued growth.
- Patches a critical OOM (Out-Of-Memory) regression bug that affected inference with quantized models, preventing crashes.
- Introduces a new 'JsonExtractString' node for parsing JSON data directly within visual workflows.
- Adds a toggle to disable default templates in text gen nodes, giving advanced users more control over prompt formatting.
Why It Matters
For professionals using ComfyUI for stable diffusion workflows, this update ensures greater stability and provides new tools for building complex, automated image generation pipelines.