ComfyUI - disappearing workflows
A persistent bug in the popular AI tool is causing users to lose complex Stable Diffusion workflows.
A significant and frustrating bug has surfaced in ComfyUI, the popular open-source, node-based graphical interface for running Stable Diffusion workflows. Users are reporting that upon launching the application, only a single project tab is restored, despite having multiple complex workflows open when the program was closed. More critically, these tabs appear to be overwriting each other in an uncontrolled manner. As detailed in a viral Reddit post, one user lost a meticulously crafted SDXL workflow when an older project, opened briefly the previous day, inexplicably overwrote the newer file. This represents a major disruption for artists and developers who rely on ComfyUI's visual programming environment to build intricate, multi-step image generation and processing pipelines.
The core of the issue seems to be related to how ComfyUI handles session state and tab management upon exit and restart. Unlike applications that reliably save and restore all open documents, ComfyUI's current behavior is causing data loss and workflow corruption. The community is actively seeking solutions, with discussions pointing towards manual backup strategies—like frequently exporting workflows as JSON files—as a temporary workaround. The bug highlights a critical stability problem for a tool that is central to many professional AI art and prototyping workflows, where recreating a lost node-based setup from memory can be nearly impossible. The development team has yet to issue an official fix, leaving users vulnerable to unexpected data loss.
- A critical bug causes ComfyUI to restore only one tab on launch, losing multiple open workflows.
- Workflows are being unpredictably overwritten, with older files replacing newer ones, leading to irreversible data loss.
- The issue forces users to adopt manual JSON backups, disrupting professional Stable Diffusion pipeline development.
Why It Matters
This bug threatens hours of work for AI artists and developers, undermining trust in a key tool for complex generative AI projects.