@gentbajko/slopify 3.1.0 → 3.1.1

package/dist/patch-notes/3.1.1.md

@@ -0,0 +1,18 @@
+# Slopify 3.1.1
+
+Released 30 September 2026.
+
+3.1.1 lets the shorts draw figures the way the video does.
+
+## Shorts
+
+- The shorts' image prompt (**Shorts → Image prompt**) takes `{{Appearance}}`, like the video's image and thumbnail prompts. Each short gets the subject's researched look and those of the figures its clip names, so a figure looks the same in the shorts as in the video.
+- A `{{Scene}}` line in the shorts' image prompt is left out: each short's images already show what is said at that moment.
+
+## Help
+
+- The **Prompts** page of the wiki explains the two keywords Slopify fills itself, `{{Scene}}` and `{{Appearance}}`: what each becomes, where it works, and an example.
+
+## Upgrading from 3.1.0
+
+- **Nothing becomes outdated.** Only a shorts image prompt that names `{{Appearance}}` changes, the next time its shorts are made.

package/dist/patch-notes/index.json

@@ -1,4 +1,10 @@
 [
+  {
+    "id": "3.1.1",
+    "title": "Slopify 3.1.1",
+    "version": "3.1.1",
+    "date": "2026-09-30"
+  },
   {
     "id": "3.1.0",
     "title": "Slopify 3.1.0",

package/dist/slices/rebuild/recipe-appearance.js

@@ -4,7 +4,7 @@ import { recipe, selectedReference, } from "./recipe-model.js";
 import { llmInput, llmInputFingerprint } from "./recipe-text.js";
 export const imageAppearanceKey = "images:appearance";
 // The prompt bodies a run draws pictures from: the generated images', the thumbnail's when it
-// is drawn from a Library prompt, and the establishing image's.
+// is drawn from a Library prompt, the establishing image's and the shorts' image prompt.
 function pictureBodies(context) {
     const { config, content } = context;
     const images = config.sources.images === "generate"
@@ -27,6 +27,9 @@ function pictureBodies(context) {
         ...(config.sources.images === "generate" && config.reference?.source === "prompt"
             ? [content.promptTemplates.referencePrompt ?? config.rendered.referencePrompt]
             : []),
+        ...(config.shorts?.enabled === true && config.shorts.imagePrompt?.trim()
+            ? [content.promptTemplates.shortsImage ?? config.rendered.shortsImage]
+            : []),
     ];
 }
 export function imageAppearance(context, text) {

package/dist/slices/rebuild/recipe-build.js

@@ -43,7 +43,7 @@ export function buildRecipes(context) {
         ...exports,
         ...youtube,
         ...(reference === undefined ? [] : [reference]),
-        ...shortsRecipes(context, exports, drawnFrom, audio.cards ?? [], masterPlan(context, audio.levels, "video")),
+        ...shortsRecipes(context, exports, drawnFrom, audio.cards ?? [], masterPlan(context, audio.levels, "video"), appearance),
         ...thumbnail,
         ...documentRecipes(context, text, thumbnail),
         ...visualAssets(context, visualRecipes(context.config, context.content, audio.mediaFingerprint, captions?.fingerprint ?? null, (images) => editPlan(context, exports, youtube, images, audio.cards ?? []), drawnFrom, timing === undefined ? null : resourceIdentity(context, timing), withLines(masterPlan(context, audio.levels, "video"), linePlan(context, timing)), scenes, appearance)),

package/dist/slices/rebuild/recipe-shorts.js

@@ -1,9 +1,11 @@
 import { z } from "zod";
 import { fingerprint } from "../../kernel/runner/work.js";
 import { usesShorts } from "../admission/rules.js";
+import { withoutScene } from "../images/scenes.js";
 import { effectiveClips, pickedShortsOf } from "../shorts/clips.js";
 import { defaultShortsImagePrompt, musicVolumeOf, shortImageCount, shortsSpeedOf, } from "../shorts/model.js";
 import { imagePromptMessages } from "../shorts/prompts.js";
+import { lookWait, withLooks } from "./recipe-appearance.js";
 import { castFor } from "./recipe-cast.js";
 import { noMaster } from "./recipe-loudness.js";
 import { recipe, resourceIdentity, selectedReference, } from "./recipe-model.js";
@@ -28,7 +30,9 @@ reference,
 // "Show tables and figures on screen": the cards a short shows where it describes one.
 cards = [], 
 // Level the volume: the levelled narration and the master (`recipe-loudness.ts`).
-master = noMaster) {
+master = noMaster, 
+// How the figures look, for a shorts image prompt with `{{Appearance}}` (`recipe-appearance.ts`).
+appearance) {
     const { config } = context;
     const shorts = config.shorts;
     const timing = exports.find((value) => value.key === "subtitles:timing");
@@ -58,8 +62,10 @@ master = noMaster) {
     { kind: "provider" });
     const picked = savedPicked(context, pick);
     const images = config.images ?? { provider: "", model: "" };
+    // A shorts image prompt may name `{{Scene}}` like the video's: the prompts call already writes
+    // each image's moment from what the clip says, so that line is left out here.
     const style = shorts.imagePrompt?.trim()
-        ? renderedPrompt(context, "shortsImage")
+        ? withoutScene(renderedPrompt(context, "shortsImage"))
         : defaultShortsImagePrompt;
     if (picked === undefined)
         return [
@@ -91,9 +97,22 @@ master = noMaster) {
         const prefix = `shorts:${String(clip.number)}`;
         const token = clipToken(context, clip);
         const count = shortImageCount(clip.end - clip.start, config.imageSeconds);
+        // The looks of the figures the clip names, and always the subject's: a short stands for
+        // the video. Until they are looked up the clip's prompts wait for them.
+        const looked = withLooks(style, appearance, clip.text, { subjectAlways: true });
+        if (looked === null) {
+            const wait = lookWait(style, appearance);
+            recipes.push(recipe(context, `${prefix}:prompts`, "video", {
+                kind: "deferred",
+                version: 1,
+                operation: "shorts",
+                template: [...wait.template, style, clip.text, count],
+            }, [pick.key, ...wait.dependsOn], { token }));
+            continue;
+        }
         const prompts = {
             ...recipe(context, `${prefix}:prompts`, "video", llmInput(context, imagePromptMessages({
-                style,
+                style: looked,
                 videoTitle: config.title,
                 shortTitle: clip.title,
                 text: clip.text,

package/dist/tutorials/Play-Images.md

@@ -124,7 +124,7 @@ The style, composition and everything else in the prompt stay as written; only t
 
 ## Looks looked up: {{Appearance}}
 
-An image model draws a named figure from whatever it half remembers, so the same character can look different in every image. Put `{{Appearance}}` in an image, thumbnail or establishing prompt, for example a line `Looks: {{Appearance}}`, and Slopify looks the looks up instead.
+An image model draws a named figure from whatever it half remembers, so the same character can look different in every image. Put `{{Appearance}}` in an image, thumbnail, establishing or shorts image prompt, for example a line `Looks: {{Appearance}}`, and Slopify looks the looks up instead.
 
 - Once the article exists, one call to the project's AI model **searches the web** for how the video's subject and every named character in the article look: build, face, clothing, colours and the marks that make them recognisable. Where depictions disagree, it describes the most iconic one.
 - With **Scenes from the article** on, each picture gets the looks of the figures its scene names (up to three besides the subject), so a scene of a place or of someone else isn't redrawn with the subject in it; a scene that names no one leaves the looks line out. A thumbnail always gets the subject's look. Without scenes, every picture gets the subject's look.

package/dist/tutorials/Prompts.md

@@ -50,7 +50,7 @@ When the body already has text, Slopify asks "Replace this prompt body?" first.
 
 ## Keywords (slots)
 
-Every `{{keyword}}` in the body becomes one field on Play.
+Every `{{keyword}}` in the body becomes one field on Play, except the two Slopify fills itself (below).
 
 - You type a keyword's value once per video, and it is filled into every prompt, intro and outro that names it.
 - Names are case-sensitive: `{{Topic}}` and `{{topic}}` are two fields.
@@ -63,6 +63,30 @@ Mistakes the editor refuses, with the line and column:
 - a keyword with no name,
 - a keyword placed inside another keyword.
 
+### Keywords Slopify fills: {{Scene}} and {{Appearance}}
+
+Two keywords never become fields. The project's AI model writes them during the run, once the article exists, and the picture waits for them:
+
+| Keyword | Filled with | Works in | Needs |
+| --- | --- | --- | --- |
+| `{{Scene}}` | This picture's moment from the article: for images, in order from the article's opening to its ending; for thumbnails, its most striking moment, a different one for each of three. | Image and thumbnail prompts | Images → **Scenes from the article** on |
+| `{{Appearance}}` | How the figures look, looked up on the web (most iconic depiction): the ones the picture's scene names, or the video's subject; a thumbnail or a short always gets the subject. | Image, thumbnail, establishing and shorts image prompts | Nothing: the keyword turns it on |
+
+Put each on a line of its own with a label, for example:
+
+```
+Scene: {{Scene}}
+
+Looks: {{Appearance}}
+```
+
+- An image prompt without `{{Scene}}` still gets its scene, after its first paragraph, while Scenes from the article is on. A thumbnail prompt without it gets none.
+- With Scenes from the article off, a line holding `{{Scene}}` is left out. A `{{Appearance}}` line whose scene names no one is left out too.
+- Each costs one more AI call per video (the looks call searches the web).
+- In the shorts' image prompt (**Shorts → Image prompt**), `{{Appearance}}` gives each short the subject's look and those of the figures its clip names. A `{{Scene}}` line is left out there: each short's images already show what is said at that moment.
+
+See [Play Images](Play-Images#scenes-from-the-article) for how the scenes and looks are chosen.
+
 ## Mark an image prompt as photorealistic
 
 YouTube asks whether a video contains realistic-looking AI scenes. Slopify answers for you from your marks.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gentbajko/slopify",
-  "version": "3.1.0",
+  "version": "3.1.1",
   "description": "Turn a prompt into a finished video, locally, with your own provider keys",
   "license": "Apache-2.0",
   "author": "Gent Bajko",