Prompt Preset
DeepSoch's Prompt Preset
- prompt
- count
Prompt Preset is a checkbox list of style presets that emits a list of prompts - and because the output is a list, ComfyUI runs the entire downstream chain once per ticked preset. Tick "realistic_skin" and "comic_style", hit queue, get two images: one hyperreal, one comic, from the same source image. This is the node you reach for when you want to restyle something several ways in a single run instead of babysitting the prompt box between queues.
Mechanically it's straightforward and pleasantly open. At startup the node loads every preset from src/deepsoch_toolkit/data/presets/styles.json (nine ship built-in) and creates one BOOLEAN checkbox per preset. Each preset is a full instruction-style prompt - things like "Enhance the image into a hyper-realistic photograph style while preserving the exact subject, composition, layout, pose..." - written as directives to a modern LLM-encoded model, not as booru tags. Note what they all have in common: "do not add any new subjects." These are built for image-to-image restyling of an existing image, which is exactly what the bundled Transform-Image workflow does.
The inputs that matter:
select_all(BOOLEAN) - the master switch; when on, it overrides the individual checkboxes and emits every preset- the per-preset BOOLEAN checkboxes -
realistic_skin,realskin_game_engine,comic_style,vintage_documentary_bw,embroidered_tapestry,ink_wash_sumie,charcoal_newsprint,Ghibli (Clothes & Hair),Zombie (Game Style) custom_prompt(STRING, multiline) - your own text, appended to whatever's ticked
Outputs are prompt (STRING list) and count (INT, how many prompts). Because prompt is a list, each ticked preset re-runs CLIP encode → sampler → VAE → save independently - N presets, N images, one queue.
The nicest part is extensibility. User presets live in user/presets/custom_styles.json inside the toolkit folder - auto-created on first run, never overwritten by updates, and they override built-ins if a name collides. The node's IS_CHANGED checks the preset files' mtimes, so save a new preset and it appears as a checkbox on the next run without restarting ComfyUI. It's the rare custom node that gets the "add your own without editing my code" bit right.
Install: ComfyUI Manager → search "DeepSoch Toolkit", or:
cd ComfyUI/custom_nodes
git clone https://github.com/deepsoch/ComfyUI-deepsoch-toolkit
then restart. No extra Python dependencies.
Two gotchas. First, select_all really does override the checkboxes - if it's on, unticking individual presets does nothing, which trips people up more than it should. Second, and this one's the real trap: these are long prompts (550–870 characters) aimed at the T5/LLM-style encoders in Flux-era models. Feed them into an SDXL-lineage CLIP encoder and you're inside 77-token truncation land, where most of the instruction silently disappears. If no preset is ticked the node emits an empty string rather than erroring - you'll get an image, but it'll be the "I have no idea what you want" image. Use it with a modern model, tick what you want, and it's one of the laziest ways to batch-style something you'll find.
Inputs (11)
| Name | Type | Default | Description |
|---|---|---|---|
| select_all | BOOLEAN | false | — |
| Ghibli (Clothes & Hair) | BOOLEAN | false | — |
| Zombie (Game Style) | BOOLEAN | false | — |
| charcoal_newsprint | BOOLEAN | false | — |
| comic_style | BOOLEAN | false | — |
| embroidered_tapestry | BOOLEAN | false | — |
| ink_wash_sumie | BOOLEAN | false | — |
| realistic_skin | BOOLEAN | false | — |
| realskin_game_engine | BOOLEAN | false | — |
| vintage_documentary_bw | BOOLEAN | false | — |
| custom_promptopt | STRING | — |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| prompt | STRING | — |
| count | INT | — |