AUTO PASS 6 · STOP / UNLOAD

Every STOP node in this pack saves a checkpoint, frees VRAM, and hands the queue to the next pass. SayaImagePhase6Stop is the one with nobody left to hand it to. It's the terminal of the six-pass pipeline - after it fires, the automated sequence doesn't queue a pass 7. It returns the system to IDLE, and whatever final upscale, color work and saving happened in pass 6 is the finished product.

Mechanically it's identical to its five siblings, so I won't re-litigate the details: image, seed, positive_prompt, negative_prompt, and the models_json/vaes_json/samplers_json metadata notes go in (record-keeping only - nothing steers the sampler), with optional source_path and an ignored save_receipt. It writes phase_6_final.candidate.png/.json, atomically promotes them to phase_6_final.png/.json under ComfyUI/output/image/checkpoints/, and emits the completion event with next_phase: 0. Outputs: image, validated_path, manifest_json.

What "STOP" really means here

The completion event phase 6 emits tells the frontend there is no next phase, so after the deferred VRAM unload the auto-controller resets to IDLE instead of queuing more work. That's the design telling you the run is over - a hard, unambiguous end-of-pipeline signal rather than silence. The phase_6_final checkpoint it leaves behind is also genuinely useful beyond the auto-run: it's the last validated artifact of the whole session, the thing you'd diff against if you re-run from an earlier phase to try a different final look, or the record you'd audit to see exactly what sampler and model produced your saved image.

Two honest notes

First, don't confuse this STOP with saving your final image. The pipeline treats phase 6's own SaveImage as part of the pass - this node persists the checkpoint, which is the pipeline's internal record, not necessarily your deliverable file. If you're grabbing the output for posting or print, take it from the phase's own save node; the checkpoint PNG in image/checkpoints is for the machine, not for sharing.

Second, and this applies to the whole family: this persistence layer is elaborate machinery that pays off only inside the author's automated pipeline. A single generate-upscale-save job doesn't need checkpoint ceremonies. But if you've committed to the six-pass run, this STOP is what lets the whole thing end cleanly - validated, documented, VRAM freed, controller back at IDLE, ready for the next generation.

Install via ComfyUI Manager (search saya-comfy-couple-plus) or git clone https://github.com/alphaziod/saya-comfy-couple-plus into custom_nodes, restart, and keep backups - the pack is explicit WIP and the README warns ports and behavior can change between updates.