fix(skills): auteur follows the modern section order and routes assets through image_generate
SKILL.md is restructured to authoring standard 5 (When to Use, Prerequisites, How to Run, Quick Reference, Procedure, Pitfalls, Verification) — headings only, upstream body text kept. The asset references still carried the upstream per-CLI routing tables and command lines for other agent products; those are replaced with the native `image_generate` route (product names are allowed only in LICENSE and credit lines). `WebSearch` residue in recon docs/refscout becomes `web_search`. source.mjs wrote its ~2.6MB Google Fonts metadata cache to $TEMP||$TMPDIR||'.', which is the project cwd on most Linux shells; it now uses os.tmpdir() and Pitfalls documents the location. Network-access note now mentions that moodboard.mjs also downloads the image URLs the search hosts return. Docs page regenerated for this skill only.
This commit is contained in:
teknium1
@@ -22,7 +22,7 @@ metadata:
|
||||
|
||||
Auteur designs and builds web experiences the way a film director makes a film: script first, then assets, then the shoot, then the cut. It has three registers — **build** (an excellent conventional site), **direct** (a cinematic scroll-directed site) and **system** (a multi-screen product as one design system) — on one shared core of taste. Nothing ships until the page passes an executable anti-slop gate and the skill has looked at its own output.
|
||||
|
||||
### Use this when
|
||||
## When to Use
|
||||
|
||||
- A landing page, marketing site, hero section, portfolio or product page has to be **built or redesigned** — and looking generic is not acceptable.
|
||||
- The brief asks for **scroll animation, storytelling, or a site that feels like a film**.
|
||||
@@ -31,6 +31,15 @@ Auteur designs and builds web experiences the way a film director makes a film:
|
||||
|
||||
Not for polishing a UI someone else built, and not for backend-only work.
|
||||
|
||||
## Prerequisites
|
||||
|
||||
- **Node 18+** — every QA gate is a `.mjs` script run with `node` via the `terminal` tool.
|
||||
- **Playwright (for the QA gates)** — in the project directory: `npm install playwright` then `npx playwright install chromium`. Required by `scripts/shoot.mjs`, `motionqa.mjs`, `systemscan.mjs`, `refscout.mjs`, `chromadiff.mjs`, `moodboard.mjs` (the scripts import `playwright` at runtime; `slopscan.mjs` and `source.mjs` are dependency-light).
|
||||
- **ffmpeg** — optional; only for the video/score paths in `references/assets.md` and `references/scroll-flight.md`.
|
||||
- **Hermes tools** — use `image_generate` for image generation/editing, `terminal` for node/ffmpeg/npm, `write_file`/`read_file` for project files, `vision_analyze` to actually look at screenshots, and `browser_exec` for live-page inspection when a script isn't the right fit.
|
||||
|
||||
## How to Run
|
||||
|
||||
### What it actually does
|
||||
|
||||
1. Commits the art direction **in writing before any markup** — one hue, one type system, a motion budget, named anti-references.
|
||||
@@ -40,11 +49,29 @@ Not for polishing a UI someone else built, and not for backend-only work.
|
||||
|
||||
### Network access
|
||||
|
||||
The recon and sourcing scripts read live pages (awwwards, Bing/Pinterest/are.na image search, Poly Haven, Iconify, Google Fonts, Openverse, Coverr). Fetched content is **treated as reference data and licence metadata — never executed**, and no credentials, API keys or logins are involved. Skip phases 0–1 entirely if you don't want outbound requests; every other phase works offline.
|
||||
The recon and sourcing scripts read live pages (awwwards, Bing/Pinterest/are.na image search, Poly Haven, Iconify, Google Fonts, Openverse, Coverr). Fetched content is **treated as reference data and licence metadata — never executed**, and no credentials, API keys or logins are involved. Skip phases 0–1 entirely if you don't want outbound requests; every other phase works offline. `moodboard.mjs` also downloads the image URLs returned by those search hosts to build the contact sheet.
|
||||
|
||||
## Non-negotiables
|
||||
### Routing
|
||||
|
||||
These apply to every register, every phase, always — even if no reference file has been loaded. Match-and-refuse: if you are about to produce one of these, stop and restructure the element.
|
||||
Read the argument / brief and route:
|
||||
|
||||
1. **`direct`** or the brief smells cinematic — "wow", "cinematic", "immersive", "storytelling", "launch page", "premium brand", "make people stop scrolling" → load `references/direct.md` and follow its phases. This is the flagship register.
|
||||
2. **`build`** or the brief is ONE conventional surface — a marketing page, a landing, a single product page → load `references/build.md`.
|
||||
3. **`system`** or the brief has **more than one screen that must feel like one product** — app, dashboard, admin, settings, onboarding, a docs or content site with real navigation → load `references/system.md`. The unit of design becomes the component × state, the failure mode becomes drift rather than boredom, and there is deliberately **no peak**. If you are already in `build` and a second screen appears, stop and switch: half a system is worse than either.
|
||||
4. **`edit`** or the request modifies a page this skill built (the project contains `design/DESIGN.md`) — "add a section", "change the pricing", "swap the hero copy" → read `design/DESIGN.md` FIRST and follow its Editing protocol: reuse its tokens, section-opening patterns, and motion families; after the change run slopscan and re-shoot the affected viewports. An edit that ignores DESIGN.md is a regression even if it looks good in isolation.
|
||||
5. **`recon <brief>`** or the ask is only for reference material — "find references", "put together a moodboard", "what's the state of the art for X sites" → load `references/recon.md` and run just that phase: scout live sites, build the moodboard, hand back `design/refs/REFERENCES.md` (with the `steal:` lines filled) and `design/moodboard/contact-sheet.png` (with the read filled). No commit-sheet, no build.
|
||||
6. **`audit <path-or-url>`** → load `references/verify.md` and run the verification pipeline on an auteur-built page. If the target is an existing UI auteur didn't build and the user wants it *polished* rather than *rebuilt*, say that a dedicated UI-polish/critique pass (upstream paired auteur with a separate 'impeccable' skill, not vendored here) is the right tool and offer to continue only if they want a rebuild.
|
||||
7. **Ambiguous** (e.g. plain "make a landing page") → ask exactly one question: "A great conventional landing page, or cinema mode with scroll direction and generated assets?" (upstream phrased these example briefs in Russian; translated here.) Then route. (Multi-screen briefs are not ambiguous — they are `system`.) Don't ask anything else yet — each register runs its own intake.
|
||||
|
||||
All three registers share phase zero, and its centre of gravity is the commit-sheet. Order differs: **build** runs recon → commit-sheet → mockup; **direct** runs recon → storyboard → commit-sheet → mockup, because the film's scenes are what the six decisions get made *about*; **system** runs recon → system-sheet (route map + component inventory) → commit-sheet → mockup, because the six decisions get made about a product, not a page. Either way nothing is coded before the sheet is full.
|
||||
|
||||
### Working relationship with other skills
|
||||
|
||||
Auteur *builds*; it does not re-polish foreign UI. If the user has an existing interface that needs refinement, run a separate UI-critique pass (e.g. `vision_analyze` on screenshots plus the sibling design skills). Upstream paired auteur with an 'impeccable' critique skill (not vendored here); auteur's verify gate and an outside critique measure different things and coexist happily.
|
||||
|
||||
## Quick Reference
|
||||
|
||||
The non-negotiables and the phase table. They apply to every register, every phase, always — even if no reference file has been loaded. Match-and-refuse: if you are about to produce one of these, stop and restructure the element.
|
||||
|
||||
### Banned (rewrite, don't tweak)
|
||||
|
||||
@@ -84,35 +111,7 @@ A ban may be overridden only through a written `auteur-allow` (see Verification)
|
||||
- `prefers-reduced-motion` = an alternative art direction (gentler, not zero), never an afterthought.
|
||||
- Content must be readable with JS disabled: reveals enhance an already-visible default, never gate visibility.
|
||||
|
||||
## Routing
|
||||
|
||||
Read the argument / brief and route:
|
||||
|
||||
1. **`direct`** or the brief smells cinematic — "wow", "cinematic", "immersive", "storytelling", "launch page", "premium brand", "make people stop scrolling" → load `references/direct.md` and follow its phases. This is the flagship register.
|
||||
2. **`build`** or the brief is ONE conventional surface — a marketing page, a landing, a single product page → load `references/build.md`.
|
||||
3. **`system`** or the brief has **more than one screen that must feel like one product** — app, dashboard, admin, settings, onboarding, a docs or content site with real navigation → load `references/system.md`. The unit of design becomes the component × state, the failure mode becomes drift rather than boredom, and there is deliberately **no peak**. If you are already in `build` and a second screen appears, stop and switch: half a system is worse than either.
|
||||
4. **`edit`** or the request modifies a page this skill built (the project contains `design/DESIGN.md`) — "add a section", "change the pricing", "swap the hero copy" → read `design/DESIGN.md` FIRST and follow its Editing protocol: reuse its tokens, section-opening patterns, and motion families; after the change run slopscan and re-shoot the affected viewports. An edit that ignores DESIGN.md is a regression even if it looks good in isolation.
|
||||
5. **`recon <brief>`** or the ask is only for reference material — "find references", "put together a moodboard", "what's the state of the art for X sites" → load `references/recon.md` and run just that phase: scout live sites, build the moodboard, hand back `design/refs/REFERENCES.md` (with the `steal:` lines filled) and `design/moodboard/contact-sheet.png` (with the read filled). No commit-sheet, no build.
|
||||
6. **`audit <path-or-url>`** → load `references/verify.md` and run the verification pipeline on an auteur-built page. If the target is an existing UI auteur didn't build and the user wants it *polished* rather than *rebuilt*, say that a dedicated UI-polish/critique pass (upstream paired auteur with a separate 'impeccable' skill, not vendored here) is the right tool and offer to continue only if they want a rebuild.
|
||||
7. **Ambiguous** (e.g. plain "make a landing page") → ask exactly one question: "A great conventional landing page, or cinema mode with scroll direction and generated assets?" (upstream phrased these example briefs in Russian; translated here.) Then route. (Multi-screen briefs are not ambiguous — they are `system`.) Don't ask anything else yet — each register runs its own intake.
|
||||
|
||||
All three registers share phase zero, and its centre of gravity is the commit-sheet. Order differs: **build** runs recon → commit-sheet → mockup; **direct** runs recon → storyboard → commit-sheet → mockup, because the film's scenes are what the six decisions get made *about*; **system** runs recon → system-sheet (route map + component inventory) → commit-sheet → mockup, because the six decisions get made about a product, not a page. Either way nothing is coded before the sheet is full.
|
||||
|
||||
## The commit-sheet (before any code, both registers)
|
||||
|
||||
Slop is what happens when defaults make the decisions. The commit-sheet forces seven real decisions onto paper before the first line of code. Copy `templates/COMMIT-SHEET.md` into the project (e.g. `design/COMMIT-SHEET.md`) and fill all seven fields with non-defaults:
|
||||
|
||||
1. **Peak** — the ONE primary wow moment (direct) or signature element (build). One sentence. If you can't name it, you're not ready to build.
|
||||
2. **Color** — primary as OKLCH + commitment tier (restrained / committed / full-palette / drenched) + one line: *why this is not lavender, not cream, and not the category reflex* + **the background lightness as a number** (target mean L), because "dark feels premium" is where this skill drifts, and a number can be checked afterwards where a mood cannot.
|
||||
3. **Type** — display + text pairing on a contrast axis (serif+sans, geometric+humanist, mono+serif...) + one line: *why not Inter*.
|
||||
4. **Grid break** — the one concrete thing that breaks the symmetric-grid default: an overlap, an asymmetric split, a diagonal flow, a full-bleed interruption. Name it specifically.
|
||||
5. **Motion budget** — how many scroll-pattern families (≤3) and what they are.
|
||||
6. **Reflex check** — write down: (a) what a generic AI would do for this category (first-order reflex), (b) what a generic AI avoiding (a) would do (second-order reflex — e.g. fintech → "terminal dark mode" is *also* saturated now), (c) your chosen deviation from both. If recon ran, (a) is not a guess: whatever `design/refs/REFERENCES.md` showed five times *is* the reflex, dated and with receipts.
|
||||
7. **House tells broken** — name the **two (minimum)** items from `taste.md` §2.5 you are deliberately not doing this time, and what replaces each. Fields 6a/6b are the reflexes of the *category*; these are the reflexes of *this skill*, which recur across unrelated projects and are invisible from inside any one of them: near-black backgrounds, mono service labels, the logo/status/action header, the scroll-instruction footer, amber-or-acid accents, the wordmark-as-hero, glow standing in for lighting. Measured across nine showcase builds, eight were dark and three landed within 0.002 of the same lightness. A tell that genuinely belongs here can stay — say why, as with an `auteur-allow`.
|
||||
|
||||
Gate: every field filled with a specific, non-default answer. An empty or generic field ("modern, clean look") means stop and decide. This artifact is checked again at verification.
|
||||
|
||||
## Phases at a glance
|
||||
### Phases at a glance
|
||||
|
||||
| Phase | build register | direct register | system register | Reference to load |
|
||||
|---|---|---|---|---|
|
||||
@@ -126,7 +125,23 @@ The hero mockup gate (one static throwaway screen, screenshotted and approved be
|
||||
|
||||
Never skip a gate because the intermediate result "looks done". The gates exist because a page that merely looks done is exactly what every other AI ships.
|
||||
|
||||
## Reference files
|
||||
## Procedure
|
||||
|
||||
### The commit-sheet (before any code, both registers)
|
||||
|
||||
Slop is what happens when defaults make the decisions. The commit-sheet forces seven real decisions onto paper before the first line of code. Copy `templates/COMMIT-SHEET.md` into the project (e.g. `design/COMMIT-SHEET.md`) and fill all seven fields with non-defaults:
|
||||
|
||||
1. **Peak** — the ONE primary wow moment (direct) or signature element (build). One sentence. If you can't name it, you're not ready to build.
|
||||
2. **Color** — primary as OKLCH + commitment tier (restrained / committed / full-palette / drenched) + one line: *why this is not lavender, not cream, and not the category reflex* + **the background lightness as a number** (target mean L), because "dark feels premium" is where this skill drifts, and a number can be checked afterwards where a mood cannot.
|
||||
3. **Type** — display + text pairing on a contrast axis (serif+sans, geometric+humanist, mono+serif...) + one line: *why not Inter*.
|
||||
4. **Grid break** — the one concrete thing that breaks the symmetric-grid default: an overlap, an asymmetric split, a diagonal flow, a full-bleed interruption. Name it specifically.
|
||||
5. **Motion budget** — how many scroll-pattern families (≤3) and what they are.
|
||||
6. **Reflex check** — write down: (a) what a generic AI would do for this category (first-order reflex), (b) what a generic AI avoiding (a) would do (second-order reflex — e.g. fintech → "terminal dark mode" is *also* saturated now), (c) your chosen deviation from both. If recon ran, (a) is not a guess: whatever `design/refs/REFERENCES.md` showed five times *is* the reflex, dated and with receipts.
|
||||
7. **House tells broken** — name the **two (minimum)** items from `taste.md` §2.5 you are deliberately not doing this time, and what replaces each. Fields 6a/6b are the reflexes of the *category*; these are the reflexes of *this skill*, which recur across unrelated projects and are invisible from inside any one of them: near-black backgrounds, mono service labels, the logo/status/action header, the scroll-instruction footer, amber-or-acid accents, the wordmark-as-hero, glow standing in for lighting. Measured across nine showcase builds, eight were dark and three landed within 0.002 of the same lightness. A tell that genuinely belongs here can stay — say why, as with an `auteur-allow`.
|
||||
|
||||
Gate: every field filled with a specific, non-default answer. An empty or generic field ("modern, clean look") means stop and decide. This artifact is checked again at verification.
|
||||
|
||||
### Reference files
|
||||
|
||||
- `references/recon.md` — **phase 0 scouting**, two executable legs: `scripts/refscout.mjs` profiles live award-level sites (real stack, pinned scenes, scroll budget, fonts, painted palette, screenshots — mechanics, not skins) and `scripts/moodboard.mjs` builds a numbered contact sheet from Bing / Pinterest / are.na so the art direction is decided from live material instead of memory. Also: query craft, the steal rule, how recon feeds the commit-sheet, and the "reference images are not assets" line. Load at the top of phase 0.
|
||||
- `references/taste.md` — the full anti-slop system: extended bans with replacements, second-order category reflex table, color strategy tiers, typography pairing, copy rules. Load for any visual decision-making.
|
||||
@@ -134,41 +149,31 @@ Never skip a gate because the intermediate result "looks done". The gates exist
|
||||
- `references/build.md` — the standard register process. Load when routed to build.
|
||||
- `references/system.md` — the **multi-screen register**: route map, the component inventory as a gate, the state matrix (empty/loading/error are not edge cases), density rules, the no-peak rule, and `scripts/systemscan.mjs` — which crawls every route, reads what the browser actually painted, fails a control type over its declared variant budget — counting *states* (disabled, current, inside a `data-state` row) separately, so implementing the state matrix never reads as drift — presses Tab to catch controls with no visible focus state, and renders one tile per rendered variant so drift is visible as well as counted. Load when routed to system.
|
||||
- `references/direct.md` — the cinematic register: screenplay contract, scene-sheets, dramaturgy, assembly order. Load when routed to direct.
|
||||
- `references/assets.md` — the media crew and routing (in Hermes: `image_generate` for all image generation and edits, `terminal` for ffmpeg/node; upstream's video/score CLI routing kept as reference), **§0.5 source-vs-generate** (`scripts/source.mjs`: CC0 glTF meshes, HDRIs and PBR materials from Poly Haven, icons, fonts, CC images, stock video — with a licence ledger, because generation cannot make geometry or an IBL and stock video must never be the peak), the consistency trick (edit frame A into frame B), local video via the first→last-frame chain, generated elements/mockups, the ambient score, the degradation ladder, and asset caching. Load during direct phase 1.
|
||||
- `references/assets.md` — the media crew and routing (in Hermes: `image_generate` for all image generation and edits, `terminal` for ffmpeg/node; video via whatever image→video backend the user has), **§0.5 source-vs-generate** (`scripts/source.mjs`: CC0 glTF meshes, HDRIs and PBR materials from Poly Haven, icons, fonts, CC images, stock video — with a licence ledger, because generation cannot make geometry or an IBL and stock video must never be the peak), the consistency trick (edit frame A into frame B), local video via the first→last-frame chain, generated elements/mockups, the ambient score, the degradation ladder, and asset caching. Load during direct phase 1.
|
||||
- `references/scroll-cinema.md` — working code recipes: scroll-scrubbed video, canvas sequences, GSAP+Lenis foundation, CSS scroll-driven animations, text reveals, the two-keyframe WebGL displacement transition, view transitions, ambient audio, and the cinematic transition library (wipe, curtain, letterbox, shutter, depth parallax). Load during assembly.
|
||||
- `references/scroll-flight.md` — the **video-scrub tier**: a photoreal "fly through the world" hero driven by scroll, using the drop-in `templates/scroll-flight-engine.js`. The canonical recipe for scroll-scrubbed *video* (encode-for-scrubbing `-g 8`, encoded-frame posters, SSIM seam gate, chain architecture A/B, iOS/mobile decode hardening, crossfade-vs-seamless seams). Load when the hero should be photoreal footage/AI-video rather than real-time WebGL.
|
||||
- `references/ambient-backgrounds.md` — **quiet** texture for secondary sections and simpler builds (not a hero): a curated 6 editorial/analog effects (paper grain, ledger/blueprint rules, topographic contour, ink tide, sparse dust, one heat-haze shader) + a zero-motion static-mesh default. The governing rule (weaker than the quietest foreground element; one ambient per page), the CSS/SVG-first stack, and the `feTurbulence`-static perf rule. Load when a section needs to not be flat but must NOT compete with copy.
|
||||
- `references/verify.md` — the acceptance pipeline: slopscan → screenshot journey → motion/perf/audio QA (FPS at DPR 2 on a production build, long-tasks, audio-gate, reduced-motion, for Tier-1 scenes) → numeric rubric → **reference diff** (your frame beside the reference that set the direction, with `scripts/chromadiff.mjs` measuring the colour drift a model never sees in itself) → QA sign-off. Load at phase 3.
|
||||
|
||||
## Verification is part of the build
|
||||
|
||||
The page is not done when the code compiles. It is done when:
|
||||
|
||||
1. `node scripts/slopscan.mjs <src-dir>` exits 0 (fails are fixed, not suppressed — `/* auteur-allow: RULE_ID -- reason */` exists for deliberate choices and demands a real reason);
|
||||
2. `node scripts/shoot.mjs <url>` has produced screenshot journeys at 390 / 768 / 1440 and you have **looked at every frame** — text overflow, blank scenes, broken reveals, layout collapse are found by eyes, not by grep;
|
||||
3. the numeric rubric in `references/verify.md` passes (contrast, LCP, CLS, reduced-motion journey, scene variety);
|
||||
4. for direct register: `CINEMA-QA.md` (from templates) is filled with PASS on every row.
|
||||
|
||||
If any gate fails — fix and re-run. Report results honestly: "slopscan clean, 21 screenshots reviewed, LCP 1.9s" beats "looks great".
|
||||
|
||||
## Working relationship with other skills
|
||||
|
||||
Auteur *builds*; it does not re-polish foreign UI. If the user has an existing interface that needs refinement, run a separate UI-critique pass (e.g. `vision_analyze` on screenshots plus the sibling design skills). Upstream paired auteur with an 'impeccable' critique skill (not vendored here); auteur's verify gate and an outside critique measure different things and coexist happily.
|
||||
|
||||
## Weak-model note
|
||||
### Weak-model note
|
||||
|
||||
If you are a smaller model executing this skill: follow the tables and numbers literally, fill every template field, run every gate command, and do not improvise beyond the reference recipes — the recipes are verified, your improvisation is not. When a reference file conflicts with your instinct, the reference file wins. Write files using paths relative to the project root; never retype an absolute path from memory (the skill's name "auteur" is one typo away from "author", and misspelled absolute paths scatter your output across the filesystem).
|
||||
|
||||
## Prerequisites
|
||||
|
||||
- **Node 18+** — every QA gate is a `.mjs` script run with `node` via the `terminal` tool.
|
||||
- **Playwright (for the QA gates)** — in the project directory: `npm install playwright` then `npx playwright install chromium`. Required by `scripts/shoot.mjs`, `motionqa.mjs`, `systemscan.mjs`, `refscout.mjs`, `chromadiff.mjs`, `moodboard.mjs` (the scripts import `playwright` at runtime; `slopscan.mjs` and `source.mjs` are dependency-light).
|
||||
- **ffmpeg** — optional; only for the video/score paths in `references/assets.md` and `references/scroll-flight.md`.
|
||||
- **Hermes tools** — use `image_generate` for image generation/editing, `terminal` for node/ffmpeg/npm, `write_file`/`read_file` for project files, `vision_analyze` to actually look at screenshots, and `browser_exec` for live-page inspection when a script isn't the right fit.
|
||||
|
||||
## Pitfalls
|
||||
|
||||
- **Network recon**: `refscout.mjs`, `moodboard.mjs` and `source.mjs` read live pages (awwwards, Bing/Pinterest/are.na image search, Poly Haven, Iconify, Google Fonts, Openverse, Coverr). Fetched content is reference data and licence metadata only — never execute it. Skip phases 0–1 to stay fully offline.
|
||||
- **Different harness**: these scripts and docs were written for a different agent harness (upstream drove asset generation through local `agy`/`codex`/`grok` CLIs). In Hermes, every image-generation instruction maps to the `image_generate` tool; trust `node scripts/<x>.mjs --help` output and actual node errors over doc prose if they drift.
|
||||
- **Different harness**: these scripts and docs were written for a different agent harness (upstream drove asset generation through several local image CLIs). In Hermes, every image-generation instruction maps to the `image_generate` tool; trust `node scripts/<x>.mjs --help` output and actual node errors over doc prose if they drift.
|
||||
- **Unverified commands**: the scripts pass `node --check` syntax validation, but full runs (which need `npm install playwright` + a chromium download) were not executed during porting. Treat `shoot.mjs`, `motionqa.mjs`, `systemscan.mjs`, `refscout.mjs`, `chromadiff.mjs`, `moodboard.mjs`, `source.mjs` end-to-end behavior, and all `ffmpeg`/video-encode recipes, as unverified upstream claims until you run them yourself.
|
||||
- **slopscan verified shape**: `node scripts/slopscan.mjs <dir>` runs without npm deps; it prints per-rule findings and exits non-zero on failures (exit 0 when clean).
|
||||
- **Font metadata cache**: `source.mjs font …` caches Google Fonts' ~2.6MB metadata JSON as `auteur-gf-metadata.json` in the OS temp directory (`os.tmpdir()`), not the project; delete it there to force a refresh.
|
||||
|
||||
## Verification
|
||||
|
||||
The page is not done when the code compiles. It is done when:
|
||||
|
||||
1. `node scripts/slopscan.mjs <src-dir>` exits 0 (fails are fixed, not suppressed — `/* auteur-allow: RULE_ID -- reason */` exists for deliberate choices and demands a real reason);
|
||||
2. `node scripts/shoot.mjs <url>` has produced screenshot journeys at 390 / 768 / 1440 and you have **looked at every frame** — text overflow, blank scenes, broken reveals, layout collapse are found by eyes, not by text search;
|
||||
3. the numeric rubric in `references/verify.md` passes (contrast, LCP, CLS, reduced-motion journey, scene variety);
|
||||
4. for direct register: `CINEMA-QA.md` (from templates) is filled with PASS on every row.
|
||||
|
||||
If any gate fails — fix and re-run. Report results honestly: "slopscan clean, 21 screenshots reviewed, LCP 1.9s" beats "looks great".
|
||||
|
||||
@@ -1,4 +1,4 @@
|
||||
> **Hermes adaptation note:** upstream auteur generated assets through local agent CLIs (`agy`, `codex`, `grok`). In Hermes, read every such invocation as a call to the built-in `image_generate` tool with the same prompt (then move the returned file into the project's `assets/gen/` path), use the `terminal` tool for `ffmpeg`/`node`/`npx`, and `browser_exec` or Playwright-via-terminal for screenshot loops. The per-CLI routing/strength tables below are upstream reference material — the taste guidance transfers, the CLI names do not.
|
||||
> **Hermes adaptation note:** upstream auteur routed asset generation to several local image CLIs by strength. In Hermes there is one route: the built-in `image_generate` tool for every still and edit (move the returned file into the project's `assets/gen/` path), the `terminal` tool for `ffmpeg`/`node`/`npx`, and `browser_exec` or Playwright-via-terminal for screenshot loops. The taste guidance below transfers; the per-CLI shootout tables were dropped in the port.
|
||||
|
||||
# assets.md — producing visual assets
|
||||
|
||||
@@ -6,36 +6,25 @@ The storyboard's `asset:` lines are a shot list. This file turns them into files
|
||||
|
||||
## 0. The crew — probe once, route by strength
|
||||
|
||||
**Probe with a real round trip, never with `--version`.** These are subscription CLIs and the failure
|
||||
that actually happens is expired auth, not a missing binary — measured: all three passed `--version`
|
||||
and all three were dead. Worse, they fail dishonestly: **`grok` exits 0 while printing "Not signed
|
||||
in"** and **`agy` exits 2 while printing nothing at all**. Read the OUTPUT, not the exit code.
|
||||
**Probe with a real round trip, never with a version flag.** The failure that actually happens is an unconfigured or expired backend, not a missing binary. Before the shoot, run one tiny `image_generate` call and confirm a file or URL actually came back; run `ffmpeg -version` in `terminal` for the encode leg. Read the OUTPUT, not the exit code.
|
||||
|
||||
```bash
|
||||
agy -p "reply with the single word: ok" # Gemini: fast image gen + edit, writes straight to a path
|
||||
codex exec --skip-git-repo-check "reply with: ok" # gpt-image: highest fidelity
|
||||
grok -p "reply with the single word: ok" # grok-4.5: gen + edit + real video, one consistent engine
|
||||
ffmpeg -version # not a subscription; --version is fine here
|
||||
```
|
||||
A route that does not return a real asset is unavailable. Note it as unavailable in the asset plan and route around it — §0.5 and §4 — rather than discovering it mid-shoot.
|
||||
|
||||
A tool that does not answer `ok` is unavailable, whatever its version says. Note it as unavailable in
|
||||
the asset plan and route around it — §0.5 and §4 — rather than discovering it mid-shoot.
|
||||
MiniMax music (ambient score) needs `MINIMAX_API_KEY` — skip the audio leg if unset.
|
||||
|
||||
Grok reaches **grok-4.5** only through the non-EU proxy (`ALL_PROXY="$GROK_PROXY" HTTPS_PROXY="$GROK_PROXY"`); without it grok still gens/edits/videos on grok-build. MiniMax music (ambient score) needs `MINIMAX_API_KEY` — skip the audio leg if unset.
|
||||
**Route each asset to its strength:**
|
||||
|
||||
**Route each asset to its strength** (locked by a shootout, 2026-07):
|
||||
|
||||
| Asset | Tool | Why |
|
||||
| Asset | Route | Why |
|
||||
|---|---|---|
|
||||
| Hero / brand-critical stills — peak scene, abstract hero background (needs clean negative space for text), product mockup / UI screen, premium transparent element or icon | **codex** | quality king across every type tested: cleanest UI render, most negative space, best material realism. Weaknesses: palette drifts warm (weak on teal-shadow / cool briefs), and it cannot make video. It *can* edit an existing frame — via stdin only, see §2 |
|
||||
| Any scene that becomes VIDEO or needs a consistent A→B edit pair; exact brand-COLOR adherence | **grok-4.5** | one engine does gen + edit + video → zero scene drift across the A→B→clip pipeline; best palette adherence when codex drifts warm |
|
||||
| Volume & CONTEXT — lifestyle/environmental shots (room, hands, props, in-situ), bulk backgrounds, fast iteration | **agy** | fast, natural environmental context, writes direct to file |
|
||||
| Real video | **grok** `image_to_video` (6 or 10s) | animate an approved keyframe |
|
||||
| Hero / brand-critical stills — peak scene, abstract hero background (needs clean negative space for text), product mockup / UI screen, premium transparent element or icon | `image_generate` | one frame first, approve, then batch style-locked to the anchor. Watch for warm palette drift on cool briefs — anchor the palette in the prompt |
|
||||
| Any scene that becomes VIDEO or needs a consistent A→B edit pair; exact brand-COLOR adherence | `image_generate` edit of frame A (§2) | editing keeps the world intact; a second generation never does |
|
||||
| Volume & CONTEXT — lifestyle/environmental shots (room, hands, props, in-situ), bulk backgrounds, fast iteration | `image_generate` | cheap iteration; keep the anchor still as the style reference |
|
||||
| Real video | an image→video model on an approved keyframe (6–10s) — user-supplied backend or browser tool | animate an approved keyframe; Hermes ships no native video tool |
|
||||
| Ambient score | **MiniMax** music | one loopable bed matched to the commit-sheet mood |
|
||||
|
||||
All three do transparent PNG (alpha): codex crispest, grok close, agy usable-but-softer. **Match the asset's background to the page** — generate the subject on the SAME ground the page uses (white-on-white, or true alpha) so it melts into the layout with no visible frame; a photographic rectangle floating on a flat page is an instant slop tell.
|
||||
Transparent PNG (alpha) depends on the configured backend. **Match the asset's background to the page** — generate the subject on the SAME ground the page uses (white-on-white, or true alpha) so it melts into the layout with no visible frame; a photographic rectangle floating on a flat page is an instant slop tell.
|
||||
|
||||
Missing a tool → don't fake it: descend the ladder (§4), ask the user for assets, or pivot to type-led/CSS scenes (a great film can be shot entirely in typography). Note what's available in the asset plan.
|
||||
Missing a route → don't fake it: descend the ladder (§4), ask the user for assets, or pivot to type-led/CSS scenes (a great film can be shot entirely in typography). Note what's available in the asset plan.
|
||||
|
||||
## 0.5 Source before you generate — the routing decision
|
||||
|
||||
@@ -88,7 +77,7 @@ Name the ~20 symbols you actually use in `entry.js` rather than `export * from '
|
||||
| a typeface | **source** — Google Fonts | and check it against ban #8/#18 before falling in love |
|
||||
| **the peak scene keyframe** | **generate** | it has to be this brand's world and nobody else's — this is the whole point |
|
||||
| the hero video | **generate** (§3) | the wow moment cannot be a clip three thousand pages already use |
|
||||
| an environmental / lifestyle still | **generate** (agy) | unless the brief needs a *real, identifiable* place |
|
||||
| an environmental / lifestyle still | **generate** (`image_generate`) | unless the brief needs a *real, identifiable* place |
|
||||
| a documentary photo of a real thing or place | **source** — Openverse | generation invents; if it must be true, it must be photographed |
|
||||
| an ambient background loop or video texture | **source ok** — Coverr | supporting layer only |
|
||||
|
||||
@@ -120,43 +109,23 @@ Build the prompt FROM the scene-sheet — `subject` + `camera` + `lighting` are
|
||||
|
||||
> "⟨subject⟩, ⟨camera: low-angle close shot / orbital view / macro detail⟩, ⟨lighting: hard rim light at dusk / soft studio / neon-soaked⟩, color palette anchored on ⟨primary OKLCH → describe as human color⟩, photographic, no text, no watermark, 16:9"
|
||||
|
||||
**agy (fast, direct to file):**
|
||||
```bash
|
||||
agy -p "Generate an image: <prompt>. Save to <ABSOLUTE-PATH>/assets/gen/s3-peak-a.png"
|
||||
```
|
||||
Always give an absolute path; verify the file actually landed on disk (agy occasionally reports success without writing — re-run once if missing).
|
||||
**Generate with `image_generate`:** pass the prompt above, then move the returned file to `assets/gen/s3-peak-a.png` (project-relative). Verify the file actually landed on disk before building on it — re-run once if missing.
|
||||
|
||||
**codex (higher quality, for the peak scene / brand-critical frames):**
|
||||
```bash
|
||||
codex exec --skip-git-repo-check "Generate an image: <prompt>"
|
||||
```
|
||||
codex cannot write into your project (read-only sandbox). Pick up the newest PNG from its output store and copy it yourself — PowerShell:
|
||||
```powershell
|
||||
Get-ChildItem "$env:USERPROFILE\.codex\generated_images" -Recurse -Filter *.png |
|
||||
Sort-Object LastWriteTime -Descending | Select-Object -First 1 |
|
||||
Copy-Item -Destination "assets/gen/s3-peak-a.png"
|
||||
```
|
||||
|
||||
Default split: agy for volume and iteration speed; codex for the peak scene and anything the viewer will stare at.
|
||||
Default split: cheap iteration for volume; spend the retries on the peak scene and anything the viewer will stare at.
|
||||
|
||||
## 2. The consistency trick: frame B is an EDIT of frame A, never a second generation
|
||||
|
||||
Two independent generations of "the same scene" are never the same scene — lighting, geometry and lens drift. Editing frame A into frame B keeps the world intact and is what makes the two-keyframe cinema moves (displacement morph, before/after scrub) look like camera work instead of a jump cut.
|
||||
|
||||
**agy edit — word the change HARSHLY.** agy ignores soft phrasing ("replace X with Y" often returns the original). Use the REQUIRED CHANGE pattern:
|
||||
```bash
|
||||
agy -p "Load the image <ABS>/assets/gen/s3-peak-a.png and edit it. REQUIRED CHANGE: the laptop is now open, screen glowing, and the room lights have dimmed. KEEP IDENTICAL: camera angle, framing, composition, every other object, lighting direction, color grade. Save to <ABS>/assets/gen/s3-peak-b.png"
|
||||
```
|
||||
**Edit — word the change HARSHLY.** Image models ignore soft phrasing ("replace X with Y" often returns the original). Call `image_generate` with frame A as the input image and the REQUIRED CHANGE pattern:
|
||||
|
||||
**codex edit — prompt via stdin only** (a positional prompt together with `-i` fails with "No prompt provided"):
|
||||
```bash
|
||||
printf '%s' "REQUIRED CHANGE: ... KEEP IDENTICAL: camera, composition, lighting." | codex exec --skip-git-repo-check -i assets/gen/s3-peak-a.png -
|
||||
```
|
||||
…then pick up from `generated_images` as above.
|
||||
> "REQUIRED CHANGE: the laptop is now open, screen glowing, and the room lights have dimmed. KEEP IDENTICAL: camera angle, framing, composition, every other object, lighting direction, color grade."
|
||||
|
||||
**Verify the pair eyes-on before building on it:** open A and B side by side. Same camera? Same composition? Only the intended state changed? Small texture drift is fine — the displacement transition tolerates it (it *hides* mid-morph mush). A camera/framing shift is a FAIL: re-edit with harder KEEP IDENTICAL wording, then try codex, then descend the ladder.
|
||||
…then move the result to `assets/gen/s3-peak-b.png`.
|
||||
|
||||
**Retry policy:** any generation/edit gets ONE sharpened retry on the same tool, then ONE attempt on the other tool, then descend the ladder. Do not burn ten generations chasing a frame — reshape the scene instead.
|
||||
**Verify the pair eyes-on before building on it:** open A and B side by side. Same camera? Same composition? Only the intended state changed? Small texture drift is fine — the displacement transition tolerates it (it *hides* mid-morph mush). A camera/framing shift is a FAIL: re-edit with harder KEEP IDENTICAL wording, then descend the ladder.
|
||||
|
||||
**Retry policy:** any generation/edit gets ONE sharpened retry, then descend the ladder. Do not burn ten generations chasing a frame — reshape the scene instead.
|
||||
|
||||
### N-frame chains (for the scroll-cinema state-machine engine)
|
||||
|
||||
@@ -164,7 +133,7 @@ Extend the A→B pair to a chain: **A→B→C→D…, each an EDIT of the previo
|
||||
the whole world stays photographically consistent while it ages / opens / transforms / gets crowded. 4–6
|
||||
frames covers most stories. Verify each link same-camera before editing the next; keep the chain in scroll
|
||||
order (`s1-a … s1-d`). This chain IS the input to scroll-cinema's Tier-1 scrubber — do the whole chain in
|
||||
ONE grok session so the image model never drifts.
|
||||
ONE sitting with the same backend so the image model never drifts.
|
||||
|
||||
## 2.5 Depth maps (for the 2.5D composite / rack-focus)
|
||||
|
||||
@@ -184,18 +153,13 @@ Alternatives: a **Blender Z-pass** when the scene is a 3D render (Blender CLI;
|
||||
grayscale "depth-style" version (fast, imperfect — ok for subtle pointer-parallax, NOT for rack-focus).
|
||||
Depth is a cached master like any still. Feed color + depth to scroll-cinema §3 (2.5D composite).
|
||||
|
||||
## 3. Video — now local via grok
|
||||
## 3. Video — animate an approved keyframe
|
||||
|
||||
Grok animates an approved keyframe: `grok image_to_video` (6 or 10 seconds). Run it through the proxy for grok-4.5, save into `assets/gen/`, then optimize (§5). **Spend video like the motion budget spends attention** — one hero clip + at most a couple supporting; a video that isn't the wow peak is usually a still that should have stayed a still.
|
||||
An image→video model animates an approved keyframe (6 or 10 seconds); Hermes has no native video tool, so use whatever image→video backend the user has (or a browser tool via `browser_exec`), save into `assets/gen/`, then optimize (§5). **Spend video like the motion budget spends attention** — one hero clip + at most a couple supporting; a video that isn't the wow peak is usually a still that should have stayed a still.
|
||||
|
||||
```bash
|
||||
ALL_PROXY="$GROK_PROXY" HTTPS_PROXY="$GROK_PROXY" grok -m grok-4.5 --yolo -p \
|
||||
"image_to_video on <ABS>/assets/gen/s1-hero-a.png: slow push-in, rising steam, 6s. Save the mp4 to <ABS>/assets/gen/s1-hero.mp4"
|
||||
```
|
||||
|
||||
**Directed A→B state change (before/after, "first+last frame").** ⚠️ grok has NO true first+last-frame interpolator (checked 2026-07): `image_to_video` animates ONE source frame with no end frame; `reference_to_video` takes 2–7 images but treats them as style/content *references*, not strict start/end keyframes — an A+B reference clip is organic drift, not a controlled morph. Routes, best first:
|
||||
**Directed A→B state change (before/after, "first+last frame").** ⚠️ Most image→video models have NO true first+last-frame interpolator: image-to-video animates ONE source frame with no end frame; reference-to-video modes treat extra images as style/content *references*, not strict start/end keyframes — an A+B reference clip is organic drift, not a controlled morph. Routes, best first:
|
||||
- **Controlled, on the web (preferred):** the WebGL displacement morph between frame A and frame B (scroll-cinema.md) — exact, scroll-scrubbable, no video model, and it's the skill's signature move anyway. This is the real answer to "we have two frames and want the transition".
|
||||
- **Organic video:** `reference_to_video` with A+B as references for a loose transition, or `image_to_video` on A for pure motion (push-in, steam, drift) — endpoints not guaranteed.
|
||||
- **Organic video:** reference-to-video with A+B as references for a loose transition, or image-to-video on A for pure motion (push-in, steam, drift) — endpoints not guaranteed.
|
||||
- **A TRUE controlled first→last VIDEO** (hard requirement) still means browser Kling (first+last mode) / Runway / Veo: package frame A + frame B + the motion prompt for the user, continue other scenes, drop the clip in when it arrives.
|
||||
|
||||
**Going past 10s — chain segments.** Clips cap at 6–10s: `image_to_video` frame A, generate/edit the next state, animate that, `ffmpeg` concat. Each segment starts on the previous last frame so the seams hide.
|
||||
@@ -256,8 +220,8 @@ Budgets (verify.md re-checks): hero video ≤2MB · poster ≤300KB · sequence
|
||||
|
||||
The crew also produces the small stuff — but every generated element must survive slopscan; a generated gradient/texture that's just decoration is banned like any other. Spend it, then make it earn its place.
|
||||
|
||||
- **Textures / grain / noise / abstract shapes** → agy (fast, transparent where possible). Use as CSS `background`, `mask-image`, or a low-opacity overlay. Generate once, cache, reuse.
|
||||
- **UI mockups in a scene** (device frame + screen, product-in-hand) → codex or grok for the still; Remotion when the mockup must move.
|
||||
- **Textures / grain / noise / abstract shapes** → `image_generate` (transparent where the backend allows). Use as CSS `background`, `mask-image`, or a low-opacity overlay. Generate once, cache, reuse.
|
||||
- **UI mockups in a scene** (device frame + screen, product-in-hand) → `image_generate` for the still; Remotion when the mockup must move.
|
||||
- **Hero mockup gate** (phase 0) can now be a *generated* frame, not only a hand-built HTML screen — one throwaway, screenshotted, approved before the real build.
|
||||
- **Iconography / brand marks** → generate a set, then hand-pick: generated icon sets drift in weight/style, so treat them as sketches to redraw in SVG, not final assets.
|
||||
|
||||
|
||||
@@ -1,4 +1,4 @@
|
||||
> **Hermes adaptation note:** upstream auteur generated assets through local agent CLIs (`agy`, `codex`, `grok`). In Hermes, read every such invocation as a call to the built-in `image_generate` tool with the same prompt (then move the returned file into the project's `assets/gen/` path), use the `terminal` tool for `ffmpeg`/`node`/`npx`, and `browser_exec` or Playwright-via-terminal for screenshot loops. The per-CLI routing/strength tables below are upstream reference material — the taste guidance transfers, the CLI names do not.
|
||||
> **Hermes adaptation note:** upstream auteur generated assets through several local image CLIs. In Hermes, read every generation instruction as a call to the built-in `image_generate` tool with the same prompt (then move the returned file into the project's `assets/gen/` path), use the `terminal` tool for `ffmpeg`/`node`/`npx`, and `browser_exec` or Playwright-via-terminal for screenshot loops. The per-CLI routing/strength tables below are upstream reference material — the taste guidance transfers, the CLI names do not.
|
||||
|
||||
# direct.md — the cinematic register
|
||||
|
||||
@@ -65,10 +65,10 @@ media: the director's shot spec for this scene's asset (→ the asset pla
|
||||
· route: SOURCE or GENERATE — assets.md §0.5 decides. Geometry, IBL lighting and tiling
|
||||
materials are SOURCE (source.mjs, CC0); the peak keyframe and the hero video are
|
||||
always GENERATE; stock video is never the peak
|
||||
· tool: codex (peak photoreal) | grok-4.5 (color hero + ANYTHING that becomes video) | agy (volume/elements)
|
||||
· tool: image_generate (stills, A→B edits) | image→video backend (anything that becomes video)
|
||||
| source.mjs hdri|model|texture|icon|image|video
|
||||
· frame prompt: the literal keyframe prompt = subject + camera + lighting + palette anchor (write it now)
|
||||
· motion prompt: (video/morph only) what moves — e.g. "grok image_to_video on frame A: slow push-in + steam, 6s". Controlled A→B state change = WebGL displacement morph (two frames), NOT a video model
|
||||
· motion prompt: (video/morph only) what moves — e.g. "image-to-video on frame A: slow push-in + steam, 6s". Controlled A→B state change = WebGL displacement morph (two frames), NOT a video model
|
||||
· score: (peak/ambient scenes only) mood/tempo for MiniMax music, or "none"
|
||||
fallback: what this scene is when WebGL/video/motion is unavailable (static frame + one line)
|
||||
```
|
||||
|
||||
@@ -30,7 +30,7 @@ node scripts/refscout.mjs --from awwwards --limit 8
|
||||
node scripts/refscout.mjs --from awwwards:scrolling --limit 6
|
||||
node scripts/refscout.mjs --from awwwards --search "coffee" --limit 5
|
||||
|
||||
# or profile sites you already know / found with WebSearch
|
||||
# or profile sites you already know / found with web_search
|
||||
node scripts/refscout.mjs https://a.com https://b.com --shots 4
|
||||
```
|
||||
|
||||
@@ -39,7 +39,7 @@ Output lands in `design/refs/`: `REFERENCES.md` (read it), `refs.json`, `shots/*
|
||||
**Finding candidates.** `--from awwwards` is the only harvester wired in, because it is the only
|
||||
gallery whose listing *and* detail pages render reliably headless and expose the outbound site
|
||||
URL. For everything else — godly.website, curated.design, minimal.gallery, land-book, siteinspire,
|
||||
thefwa, lapa.ninja, mobbin — use WebSearch to find the write-ups, then pass the site URLs to
|
||||
thefwa, lapa.ninja, mobbin — use `web_search` to find the write-ups, then pass the site URLs to
|
||||
refscout positionally. Searching for the *page about* a site is more reliable than scraping the
|
||||
gallery that lists it.
|
||||
|
||||
@@ -198,7 +198,7 @@ and none of them ships:
|
||||
|
||||
| symptom | what it means | do this |
|
||||
|---|---|---|
|
||||
| `awwwards listing returned no cards` | the gallery markup moved | fall back to WebSearch + positional URLs |
|
||||
| `awwwards listing returned no cards` | the gallery markup moved | fall back to `web_search` + positional URLs |
|
||||
| most entries are `NO CAPTURE` | network or a broad bot wall | profile fewer, better-known sites; lean on the shots and your own eyes |
|
||||
| `pinterest returned 0` | login wall, normal | ignore, bing covered it |
|
||||
| the sheet is 20 variants of one image | single narrow query | re-run with 2–3 queries on different axes |
|
||||
|
||||
@@ -847,7 +847,7 @@ Rules: rAF-throttle (one transform write per frame, never per event); clamp so t
|
||||
## State-machine cinema — A→B→C morph + audio-reactive (Tier-1 engine)
|
||||
|
||||
The signature 2026 move: ONE world that transforms through a chain of scene-consistent frames
|
||||
(assets.md §2 — grok edit-chain A→B→C…N), scrubbed by scroll AND driven by an audio track, both feeding
|
||||
(assets.md §2 — the edit-chain A→B→C…N), scrubbed by scroll AND driven by an audio track, both feeding
|
||||
the SAME `uMix`/`uEnergy` uniforms so picture and sound move as one. Extends the two-frame displacement
|
||||
to N frames. One persistent WebGL context; swap textures, never remount.
|
||||
|
||||
@@ -864,7 +864,7 @@ the current position with a noise-driven wipe, so it reads as a morph, not a cro
|
||||
```js
|
||||
import * as THREE from 'three'
|
||||
import Lenis from 'lenis'
|
||||
const FRAMES = ['/gen/s1-a.webp','/gen/s1-b.webp','/gen/s1-c.webp','/gen/s1-d.webp'] // grok edit-chain, in order
|
||||
const FRAMES = ['/gen/s1-a.webp','/gen/s1-b.webp','/gen/s1-c.webp','/gen/s1-d.webp'] // edit-chain, in order
|
||||
const canvas = document.getElementById('c')
|
||||
const renderer = new THREE.WebGLRenderer({ canvas, antialias:true })
|
||||
renderer.setPixelRatio(Math.min(devicePixelRatio, 2))
|
||||
|
||||
@@ -1,4 +1,4 @@
|
||||
> **Hermes adaptation note:** upstream auteur generated assets through local agent CLIs (`agy`, `codex`, `grok`). In Hermes, read every such invocation as a call to the built-in `image_generate` tool with the same prompt (then move the returned file into the project's `assets/gen/` path), use the `terminal` tool for `ffmpeg`/`node`/`npx`, and `browser_exec` or Playwright-via-terminal for screenshot loops. The per-CLI routing/strength tables below are upstream reference material — the taste guidance transfers, the CLI names do not.
|
||||
> **Hermes adaptation note:** upstream auteur generated assets through several local image CLIs. In Hermes, read every generation instruction as a call to the built-in `image_generate` tool with the same prompt (then move the returned file into the project's `assets/gen/` path), use the `terminal` tool for `ffmpeg`/`node`/`npx`, and `browser_exec` or Playwright-via-terminal for screenshot loops. The per-CLI routing/strength tables below are upstream reference material — the taste guidance transfers, the CLI names do not.
|
||||
|
||||
# scroll-flight — photoreal scroll-scrubbed video ("fly through the world")
|
||||
|
||||
@@ -24,8 +24,8 @@ They compose: a WebGL hero can hand off into a scrubbed-video mid-section.
|
||||
|
||||
## ⚠️ AI clips barely move the camera — the scroll-dolly does the travelling
|
||||
|
||||
The single most important thing to know here. **`grok image_to_video` (and
|
||||
most image→video models) animate a still *ambiently* — light shimmers, water
|
||||
The single most important thing to know here. **image→video models (all the
|
||||
common ones) animate a still *ambiently* — light shimmers, water
|
||||
drifts, particles float — but they do NOT fly the camera through the scene.**
|
||||
Verified on disk: a clip's first and last frame are near-identical. So a scene
|
||||
built only from a raw AI clip reads as *a slightly-moving photo*, not a journey.
|
||||
@@ -58,11 +58,11 @@ never reveal. *That* is what manufactures forward/descent travel; the clip's
|
||||
1. **Scene stills** — one anchor still first, get art-direction approval, then
|
||||
batch the rest **style-locked to the anchor** (pass the approved still as the
|
||||
style reference). A style miss caught on the anchor costs 1 gen, not N.
|
||||
Sources: `codex`/`agy` (Gemini)/`grok` image gen — see `assets.md`.
|
||||
2. **Dive clips** — animate each still into a short camera push-in (grok
|
||||
`image_to_video`, or any image→video model). One clip per scene.
|
||||
Source: `image_generate` — see `assets.md`.
|
||||
2. **Dive clips** — animate each still into a short camera push-in (any
|
||||
image→video model). One clip per scene.
|
||||
3. **Seams** — two ways, pick by what your video model can do:
|
||||
- **Crossfade seams (default, grok/start-image-only models).** Leave
|
||||
- **Crossfade seams (default, start-image-only models).** Leave
|
||||
`connectors` empty/`null`; the engine crossfades directly between adjacent
|
||||
dives. Ship-safe, always works, reads clean. This is auteur's baseline.
|
||||
- **Seamless flight (only with an end-image model — Higgsfield seedance /
|
||||
@@ -73,8 +73,8 @@ never reveal. *That* is what manufactures forward/descent travel; the clip's
|
||||
5. **Posters** — extract each encoded clip's first frame (§ below).
|
||||
6. **Wire** the engine config; run the motion + slopscan gates.
|
||||
|
||||
> auteur has no Higgsfield account by default; grok `image_to_video` is the
|
||||
> baseline. So the honest default is **crossfade-seam** photoreal scrub — still
|
||||
> auteur has no Higgsfield account by default; a start-image-only
|
||||
> image→video model is the baseline. So the honest default is **crossfade-seam** photoreal scrub — still
|
||||
> Apple-tier. Fully-seamless chaining is an upgrade you unlock only with an
|
||||
> end-image-capable video model.
|
||||
|
||||
@@ -188,5 +188,5 @@ it only when video-scrub genuinely stutters on the target hardware.
|
||||
---
|
||||
|
||||
*Technique & engine adapted from **scroll-world** by cyw
|
||||
(github.com/cth9191/scroll-world), MIT. auteur pairs it with grok/Gemini asset
|
||||
generation and its own slopscan / motionqa gates.*
|
||||
(github.com/cth9191/scroll-world), MIT. auteur pairs it with `image_generate` asset
|
||||
production and its own slopscan / motionqa gates.*
|
||||
|
||||
@@ -279,7 +279,7 @@ async function fingerprint(ctx, url) {
|
||||
// --- gallery harvesting -------------------------------------------------------
|
||||
// awwwards is the one gallery whose listing AND detail pages both render reliably headless
|
||||
// and expose the outbound site URL. Other galleries either need JS we can't wait out or bury
|
||||
// the real URL behind affiliate redirects — for those, find URLs with WebSearch and pass them
|
||||
// the real URL behind affiliate redirects — for those, find URLs with web_search and pass them
|
||||
// positionally.
|
||||
async function harvestAwwwards(ctx, tag, text, n) {
|
||||
const page = await ctx.newPage();
|
||||
@@ -322,7 +322,7 @@ let targets = urls.map(u => ({ site: u, name: null, via: 'direct' }));
|
||||
if (from) {
|
||||
const [gallery, tag] = from.split(':');
|
||||
if (gallery === 'awwwards') targets = targets.concat(await harvestAwwwards(ctx, tag, search, limit));
|
||||
else console.error(`[refscout] unknown gallery "${gallery}" — supported: awwwards. Pass URLs directly instead (find them with WebSearch).`);
|
||||
else console.error(`[refscout] unknown gallery "${gallery}" — supported: awwwards. Pass URLs directly instead (find them with web_search).`);
|
||||
}
|
||||
targets = targets.slice(0, limit);
|
||||
|
||||
|
||||
@@ -36,6 +36,7 @@
|
||||
import { mkdir, writeFile, readFile } from 'fs/promises';
|
||||
import { existsSync } from 'fs';
|
||||
import { resolve, join, dirname } from 'path';
|
||||
import { tmpdir } from 'os';
|
||||
|
||||
const args = process.argv.slice(2);
|
||||
const KINDS = ['hdri', 'model', 'texture', 'icon', 'font', 'image', 'video'];
|
||||
@@ -172,7 +173,7 @@ async function icons() {
|
||||
const GF_BANNED = ['Inter', 'Space Grotesk', 'Instrument Serif', 'Playfair Display'];
|
||||
async function fonts() {
|
||||
// Not inside --out: that just relocates 1.5MB from your cwd into your shipped assets tree.
|
||||
const cacheFile = join(process.env.TEMP || process.env.TMPDIR || '.', 'auteur-gf-metadata.json');
|
||||
const cacheFile = join(tmpdir(), 'auteur-gf-metadata.json');
|
||||
let meta;
|
||||
if (existsSync(cacheFile)) meta = JSON.parse(await readFile(cacheFile, 'utf8'));
|
||||
else {
|
||||
|
||||
@@ -40,7 +40,7 @@
|
||||
- **copy:** H: "…" / sub: "…" <!-- real words, not lorem -->
|
||||
- **media:** <!-- the director's shot spec — route via assets.md §0.5 (source) and §0 (generate) -->
|
||||
- type: <!-- still | A→B morph | video | sequence | element/texture | 3D model | HDRI | none (type-led) -->
|
||||
- route: <!-- SOURCE (source.mjs hdri|model|texture|icon|image|video) or GENERATE (codex | grok-4.5 | agy) — assets.md §0.5 decides -->
|
||||
- route: <!-- SOURCE (source.mjs hdri|model|texture|icon|image|video) or GENERATE (image_generate; image→video backend for clips) — assets.md §0.5 decides -->
|
||||
- frame prompt: <!-- the literal keyframe prompt = subject + camera + lighting + palette anchor. Write it NOW, not in phase 1. -->
|
||||
- motion prompt: <!-- video/morph only. A controlled A→B state change is the WebGL displacement morph, NOT a video model. -->
|
||||
- score: <!-- peak/ambient scenes only: mood/tempo for MiniMax music, or "none" -->
|
||||
|
||||
@@ -3,7 +3,7 @@
|
||||
----------------------------------------------------------------------------
|
||||
Portable scroll-scrubbed camera-flight engine. Photoreal AI-video "fly through
|
||||
the world" — auteur's video-scrub tier, complementary to its real-time WebGL
|
||||
recipes. Asset-source-agnostic: feed it ANY .mp4 clips (grok image_to_video,
|
||||
recipes. Asset-source-agnostic: feed it ANY .mp4 clips (image→video models,
|
||||
Gemini, real footage). See references/scroll-flight.md for the full recipe.
|
||||
|
||||
VENDORED from scroll-world (github.com/cth9191/scroll-world), MIT (c) 2026 cyw.
|
||||
|
||||
Reference in New Issue
Block a user