MiniMaxRef Save Image
A save node that respects loops, batches, and changing filenames
- images
- images
Let's be blunt: you already have a Save Image node. ComfyUI ships one, and it's fine. So the question anyone sane asks about MiniMaxRefSaveImage is why this pack bothered, and the answer is one small behavior that saves real pain: it re-executes every loop iteration even when nothing else changed, so a filename that varies per pass actually gets saved under a different name each time.
The stock save nodes cache their execution based on inputs, which is usually the right trade - but it's the wrong one inside a loop where the filename prefix comes from a computed string. MiniMaxRefSaveImage forces its fingerprint to NaN, meaning "always run me," so the filename_prefix you feed it - say, from MiniMaxRefJoinString building scene_{value1} from a loop index - is honored on every pass instead of the first one. That one property is the entire reason this node exists in the pack, and it's a genuinely better fit for the Director loop than the default save node.
The inputs are as minimal as a save node gets. filename_prefix is a string (default Tenz - a nod to the pack's author, who publishes as Tenz-8), and files land in ComfyUI/output, standard ComfyUI behavior. images is the IMAGE tensor you want written. If images is None the node just returns None rather than erroring, which makes it safe to drop into a workflow where a segment might legitimately produce nothing. The images output passes the saved images straight through, so you can chain a preview after it without adding nodes.
It's an output node (is_output_node), so it terminates that branch of the graph. If you're using it in the H3 workflow, the natural spot is saving the reference frames or keyframes your segments are built from - the first-frame reference image for a shot, say - rather than final video, which the Guide workflow routes through VHS VideoCombine instead.
Install it with the rest of the pack via ComfyUI Manager (search "MiniMaxRefDirector-ComfyUI") or:
cd ComfyUI/custom_nodes
git clone https://github.com/eaglering/MiniMaxRefDirector-ComfyUI
cd MiniMaxRefDirector-ComfyUI
pip install -r requirements.txt
then restart. The only new dependency is PyAV, used by the audio/video nodes, not this one.
Where you might get burned: the default prefix is Tenz, and if you forget to change it, everything lands in ComfyUI/output/Tenz* - harmless, but easy to mistake for a bug the first time you search your output folder. Also worth knowing: this is a brand-new pack with zero search footprint, so there's no community lore to fall back on - but the node is thin enough that the README and the example workflow in workflow/MinimaxH3 V3.1示例工作流.json are all the documentation you need.
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| filename_prefix | STRING | Tenz | The filename prefix for the saved files. Images are saved to the ComfyUI output directory. |
| images | IMAGE | The images to save. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| images | IMAGE | The saved images. |