| name | style-explorer |
| user-invocable | true |
| argument-hint | [{"optional":"a seed style or theme"}] |
| description | Roam the TITLES (titles.xyz) artist-model library — generate a wide, varied spread of images across many different artists and many different prompts to discover styles and directions. A survey, not a deliverable. Use when the user wants to explore or be shown range: "explore styles", "what styles are on TITLES", "surprise me", "show me a range", "wander the catalog", "what can this do", "discover artists by generating", "what's new on TITLES", "show me new models", "style exploration". Samples broadly across the catalog, presents the board grouped by look with each artist credited, and points at where to go deeper. Runs on the TITLES MCP — if TITLES tools are missing, connect mcp.titles.xyz/mcp first (see titles-setup). NOT for: producing a finished asset set (promo-pack), or browsing already-published feed work (curation-digest).
|
style-explorer
Roam the TITLES artist library — a wide, varied spread of generations across many artists and many prompts, to discover what's possible. The point is breadth, not a decision.
Get connected
Check the tool list for TITLES tools (names contain titles_). If missing, hand off to the titles-setup skill and stop. Don't fall back to non-TITLES tools.
1. Ask what they want to explore — specific first, breadth second
Exploration is the ask, so you don't need a full concept. If they already gave a direction when they invoked the skill — a seed style/theme, "only new models", or "surprise me" — route on it, don't ask again. Otherwise ask whether they have something specific in mind before defaulting to breadth. Either way, route on:
- A particular style or direction — a theme, medium, mood, subject, or era ("something dreamlike", "3D creatures", "art deco"). Bias the roam toward it across several artists, but keep wandering within it.
- Only new / recent models — they want to see what's just landed. Roam the newest additions via
titles_search_models({ sort: "recent" }).
- Specific artists or mediums — center the roam on those and their neighbors.
- No preference / "surprise me" — then offer the full-breadth roam: sample across the catalog's whole range. Valid here; generating varied prompts and picking styles yourself is the feature (unlike promo-pack, where inventing a brief is off-limits — there the user has a fixed output in mind; here they've asked to be shown range).
Settle the breadth (how many images, or a budget) before spending.
2. Assemble a wide, diverse set of artists
Honor the step-1 focus: bias the searches to their style/direction, lead with sort: "recent" if they asked for new models, or sweep every axis for "surprise me". Either way the goal is range, not the top of one query: run several different searches across style axes (painterly, 3D, collage, glitch, photographic, illustrative, abstract), and vary the sort (popularity, recent, alpha) so you surface artists a single query would bury. Aim for a spread of architecture, medium, and mood. Note each artist's name and model_url.
3. Budget → price the ceiling first, then set breadth
Per-image cost varies a lot by architecture — a Nano Banana Pro or Flux sample can cost several times a Z-Image or Klein one. If you budget off the first (or cheapest) sample and extrapolate, you'll size the roam too generously and then have to skip the expensive models mid-roam to stay in budget — which cuts exactly the diversity this skill exists for. So price the ceiling before you talk breadth:
- Find the top per-sample cost in your candidate set, not the first. Reason from architecture (the pricier families cost multiples of the cheap ones), or probe the dearest-looking candidate and read
cost_usd off its submit response (older servers nest it as cost_quote.total_usd).
- Size the roam at that ceiling price — ask the user's exploration budget, then compute how many samples fit as if every one were the most expensive model, so no pricey model has to be cut for cost.
- Show the count and the worst-case per-sample cost, get a go or a cap, and keep a running total. Because you priced the ceiling, the real spend usually lands under the estimate — say so when it does.
4. Roam — vary the artist AND the prompt
Generate a spread where each sample is a different (artist, prompt) pairing. Unlike a controlled comparison, here you vary both: let a collage artist get a collage-native prompt, a creature model get a creature, a landscape model a landscape — play to each model's strengths so the board shows what each artist actually does best, not one subject forced through all of them. One at a time, shared session_id. Look at each result (titles_get_execution inlines previews).
5. Present the survey and point deeper
Show the board via the session_url, grouped by look/family, each image credited to its artist with the model_url. For each cluster, a line on what's distinctive — the discovery is the map, not just the pictures. Then route the user onward:
- Make something in a style that landed →
generate-image (one image) or promo-pack (a full asset set), passing that model_id.
- Go deep on an artist →
titles_set_following to keep their work surfacing, or start creating in their style (generate-image).
- Roam further → another spread, adjacent to whatever caught their eye, or off in a new direction.
Etiquette
Fire fast within the approved budget — this is meant to be a wide, quick survey, not a labored one. One canvas link, the board grouped and every artist credited, and an honest read on what's worth chasing. Range is the deliverable; a spread of near-identical looks is a failed roam.
