πββ¬ Flux Prompt Saver
Save Flux images with metadata your Discord bots and organizers can actually read
- images
- params
Why a separate saver at all
ComfyUI's default Save Image embeds the full workflow JSON in the PNG - great for dragging the file back into ComfyUI and reconstructing your graph, useless for everything else. The other half of the ecosystem - Discord prompt-reader bots, prompt readers, image organization tools (the README's own list) - reads the classic A1111 "parameters" text chunk instead. This node writes that chunk.
So the workflow gets one sink: Flux Text Sampler β VAE decode β this, and every PNG carries positive, Negative prompt, Steps, Sampler, CFG scale, Seed, Size, Model in a format Stable Diffusion tools have read for years. If you've ever dragged a ComfyUI render into a bot and got back "no prompt found," you already know why this node exists.
What it takes in
images(IMAGE) - your decoded output.params(SAMPLER_PARAMS) - wire theparamsoutput of the bundled Flux Text Sampler here. The fields it reads directly aresteps,sampler,scheduler,width, andheight, withguidance,seed, andmodel_hashread leniently - so another SAMPLER_PARAMS source works as long as it carries those keys.positive(STRING) and optionalnegative(STRING) - wire from your prompt text nodes. An empty negative is recorded as "(not used)", which keeps the metadata clean for A1111 readers.model_name(STRING) - wire from the pack's Model Name node so the saved metadata records which model you actually used.filename_prefixandfilename- the two inputs that make this worth having.filename_prefixis the subfolder (default%date:yyyy-MM-dd%),filenameis the file name (defaultFLUX_%date:HHmmss%). Directories get created on the fly, and colliding filenames auto-increment to_1,_2, so nothing gets silently overwritten.
That split into separate path and filename inputs is the 2.0.0 breaking change - older versions took a single path string, so any pre-2.0 workflow needs its inputs rewired.
The date formatting
%date:FORMAT% where FORMAT uses readable tokens: yyyy, MM, dd, HH, mm, ss, each mapped to the equivalent strftime code. %date:yyyy-MM-dd% as the subfolder plus FLUX_%date:HHmmss% as the filename gives you output/2026-08-16/FLUX_143022.png and friends. Predictable and easy to read at a glance.
Two gotchas the README won't warn you about. First, the README's placeholder list includes %width%, %height%, and %seed% - but the shipped code only expands date tokens, so those stay literal in your filenames. Don't build filenames on the seed; you'll get files literally named img_%seed%.png. Second, this node writes the A1111 chunk and not the ComfyUI workflow JSON, so you can't drag these PNGs back into ComfyUI to restore the graph. They're for the metadata ecosystem, not for workflow sharing. Want both? Run a plain Save Image alongside it.
Install
cd ComfyUI/custom_nodes
git clone https://github.com/markuryy/ComfyUI-Flux-Prompt-Saver
then restart ComfyUI, or search "ComfyUI Flux Prompt Saver" in ComfyUI Manager. Pure Python, no pip dependencies, no models to download. The example workflow in the README needs Impact Pack, ComfyUI-Custom-Scripts, and Comfyroll installed before it loads - but the saver node itself runs on a stock install.
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| images | IMAGE | β | |
| params | SAMPLER_PARAMS | β | |
| positive | STRING | β | |
| model_name | STRING | β | |
| filename_prefix | STRING | %date:yyyy-MM-dd% | Subfolder to save the images in. Supports date formatting like %date:yyyy-MM-dd% |
| filename | STRING | FLUX_%date:HHmmss% | Filename for the image. Supports date formatting like %date:HHmmss% |
| negativeopt | STRING | β |
Outputs (0)
No outputs