@gentbajko/slopify 2.3.1 → 2.5.0

package/dist/adapters/image/replicate.js

@@ -4,6 +4,7 @@ import { providerError } from "../../kernel/ports/model.js";
 import { httpFailure, missingKey, noImage, providerSaid, refusedImage, unreadable, } from "../explain.js";
 import { retryAfter } from "../retry-after.js";
 import { downloadImage } from "./bytes.js";
+import { dataUri, downloadVideo } from "./video.js";
 // The HTTP gateway adapter for Replicate: `fetch`, the injected clock and the downloader
 // beside this file, no SDK. The `replicate` package wraps exactly the two requests below, so
 // the platform's own `fetch` covers it.
@@ -40,6 +41,13 @@ export function replicateImage(deps) {
         id: "replicate",
         models: () => Promise.resolve(replicateModels),
         generate: async (req) => {
+            // The Replicate models listed take no input image to draw from (the FLUX image inputs
+            // copy the picture's composition), so an establishing image is refused, not dropped.
+            if (req.reference !== undefined)
+                throw providerError({
+                    kind: "unsupported",
+                    message: "Replicate's image models can't use an establishing image as a reference. Choose Codex CLI, OpenAI, Google or fal.ai under Images → Provider on Play or in Edit project → Providers, or set Establishing image to Off in the Images section.",
+                });
             const response = await deps.fetch(`${replicateBase}/models/${req.model}/predictions`, {
                 method: "POST",
                 signal: req.signal,
@@ -71,8 +79,46 @@ export function replicateImage(deps) {
                 signal: req.signal,
             });
         },
+        // The same prediction, polled the same way; a clip outlives `Prefer: wait` every time, so
+        // it is not asked for.
+        animate: async (req) => {
+            const response = await deps.fetch(`${replicateBase}/models/${req.model}/predictions`, {
+                method: "POST",
+                signal: req.signal,
+                headers: { ...auth(deps), "Content-Type": "application/json" },
+                body: JSON.stringify({
+                    input: {
+                        prompt: req.prompt,
+                        [Object.hasOwn(videoImageField, req.model)
+                            ? (videoImageField[req.model] ?? "image")
+                            : "image"]: dataUri(req.image),
+                        duration: Math.round(req.seconds),
+                        ...(Object.hasOwn(videoResolution, req.model) ? { resolution: "720p" } : {}),
+                    },
+                }),
+            });
+            if (!response.ok)
+                throw await failure(response, "video clip request");
+            const url = await settle(deps, parse(await response.text()), req.signal);
+            return await downloadVideo({
+                fetch: deps.fetch,
+                provider: "Replicate",
+                url,
+                signal: req.signal,
+            });
+        },
     };
 }
+// Replicate's image-to-video models name the still differently: Kling calls it
+// `start_image`, the others `image`. Wan and Seedance default to larger frames than the
+// render needs, so they are asked for 720p, which is what the catalogue prices.
+const videoImageField = {
+    "kwaivgi/kling-v2.5-turbo-pro": "start_image",
+};
+const videoResolution = {
+    "wan-video/wan-2.5-i2v": true,
+    "bytedance/seedance-1-pro-fast": true,
+};
 // `Prefer: wait` answers `starting` when the model outlived its 60 s; Replicate's guidance
 // is to poll `urls.get` until the prediction settles.
 async function settle(deps, first, signal) {
@@ -144,7 +190,7 @@ function kindOf(status) {
     // error contract first, which is not this adapter's to widen.
     return "other";
 }
-async function failure(response) {
+async function failure(response, subject = "image request") {
     const text = await response.text().catch(() => "");
     const parsed = errorBody.safeParse(safeJson(text));
     // The provider's own words, through the redactor - an error body may quote the key back.
@@ -157,7 +203,7 @@ async function failure(response) {
             provider: "Replicate",
             status: response.status,
             detail: message || response.statusText,
-            subject: "image request",
+            subject,
         }),
         ...(retryAfterMs === undefined ? {} : { retryAfterMs }),
     });

package/dist/adapters/image/video.js

@@ -0,0 +1,28 @@
+import { providerError } from "../../kernel/ports/model.js";
+import { describeBytes } from "./bytes.js";
+// What the two image-to-video adapters share: the still goes up inline as a data URI (the
+// render sends a small JPEG, `slices/rebuild/runtime-animate.ts`), and the clip comes back
+// as a link that is downloaded inside the attempt, like an image.
+export function dataUri(image) {
+    return `data:${image.mime};base64,${Buffer.from(image.bytes).toString("base64")}`;
+}
+// An MP4 (or MOV) names its box type at byte 4.
+export function sniffVideo(bytes) {
+    return bytes.length > 12 && Buffer.from(bytes.slice(4, 8)).toString("latin1") === "ftyp";
+}
+export async function downloadVideo(download) {
+    const response = await download.fetch(download.url, { signal: download.signal });
+    if (!response.ok)
+        // The link is not quoted back: a delivery link carries its own signature.
+        throw providerError({
+            kind: "other",
+            message: `${download.provider} made the video clip, but Slopify could not download it (error ${String(response.status)}). Check your internet connection, then use Retry stage.`,
+        });
+    const bytes = new Uint8Array(await response.arrayBuffer());
+    if (!sniffVideo(bytes))
+        throw providerError({
+            kind: "other",
+            message: `${download.provider} sent back something that is not an MP4 video clip (${describeBytes(bytes)}). Use Retry stage; if it keeps happening, choose another image-to-video model under Animate images in Edit project → Inputs → Look.`,
+        });
+    return { bytes, mime: "video/mp4" };
+}

package/dist/adapters/llm/codex-models.js

@@ -1,6 +1,7 @@
 import { homedir } from "node:os";
 import { join } from "node:path";
 import { z } from "zod";
+import { thinkingModes as knownEfforts } from "../../kernel/ports/llm.js";
 import { readCatalogueFile } from "./catalogue-files.js";
 import { codexModelName, nodeCodexServerModels } from "./codex-server-models.js";
 const safeText = z
@@ -40,7 +41,7 @@ export async function nodeCodexModels(env = process.env, binary = "codex", timeo
             if (!unique.has(item.slug)) {
                 const thinkingModes = item.supported_reasoning_levels
                     ?.map((level) => level.effort)
-                    .filter((level) => ["off", "low", "medium", "high", "xhigh"].includes(level));
+                    .filter((level) => knownEfforts.includes(level));
                 unique.set(item.slug, {
                     id: item.slug,
                     name: codexModelName(item.display_name ?? item.slug),

package/dist/adapters/llm/codex-server-models.js

@@ -4,6 +4,7 @@ import { tmpdir } from "node:os";
 import { join } from "node:path";
 import { z } from "zod";
 import { cliCommand } from "../../kernel/cli-command.js";
+import { thinkingModes as knownEfforts } from "../../kernel/ports/llm.js";
 const text = z
     .string()
     .trim()
@@ -102,7 +103,7 @@ export async function nodeCodexServerModels(binary, env, timeoutMs = 15_000) {
                             continue;
                         const thinkingModes = row.supportedReasoningEfforts
                             ?.map((item) => item.reasoningEffort)
-                            .filter((effort) => ["off", "low", "medium", "high", "xhigh"].includes(effort));
+                            .filter((effort) => knownEfforts.includes(effort));
                         models.set(row.model, {
                             id: row.model,
                             name: codexModelName(row.displayName),

package/dist/adapters/llm/run-cli.js

@@ -87,12 +87,24 @@ export function nodeRunCli(binary, args, signal, options) {
             child.stderr?.destroy();
         }
     };
-    const abort = () => terminate();
+    // An abort asks politely, then forces after the same grace stopCliRun gives. A reader
+    // waiting on stdout only reaches stopCliRun once stdout closes, so a CLI that traps
+    // SIGTERM and keeps its pipe open would otherwise outlive the run that was cancelled.
+    let forced;
+    const abort = () => {
+        terminate();
+        forced ??= setTimeout(() => terminate(true), cliTerminationGraceMs);
+        forced.unref();
+    };
     if (signal.aborted)
         abort();
     else
         signal.addEventListener("abort", abort, { once: true });
-    void ended.then(() => signal.removeEventListener("abort", abort));
+    void ended.then(() => {
+        signal.removeEventListener("abort", abort);
+        if (forced !== undefined)
+            clearTimeout(forced);
+    });
     return {
         ...(inputWritten === undefined ? {} : { inputWritten }),
         pid: child.pid,

package/dist/assets/fonts/SOURCE.txt

@@ -2,6 +2,12 @@ Barlow-Regular.ttf
 Source: https://raw.githubusercontent.com/google/fonts/6cdf01867df0813c2390f90dff7dc66c87f14cf7/ofl/barlow/Barlow-Regular.ttf
 SHA-256: 95aa02c7c43096e0dd44d787ba6216864a67157e402adab59b35572e0c1577ea
 
+Barlow-Bold.ttf
+Source: https://raw.githubusercontent.com/google/fonts/6cdf01867df0813c2390f90dff7dc66c87f14cf7/ofl/barlow/Barlow-Bold.ttf
+SHA-256: 84e6a4d61e7c3e21f3c50ea6a4f7e5303a3467864c038be6ea3759bab8d547f9
+Git blob: 2f20cbe667ca5cd871410dc50dc4b1a30c1d6267 (matches the google/fonts tree at that commit)
+Licence: SIL Open Font License 1.1, OFL.txt below (the same licence and authors as Barlow-Regular)
+
 OFL.txt
 Source: https://raw.githubusercontent.com/google/fonts/6cdf01867df0813c2390f90dff7dc66c87f14cf7/ofl/barlow/OFL.txt
 SHA-256: 186d750eb496a4c17a76385f82be6aea2ac1cf2de074a811d63786cf374ea73f

package/dist/assets/models.yaml

@@ -2,7 +2,7 @@
 # A private copy in the Slopify data directory reloads automatically.
 # Omitted pricing means unknown, never free. Provider concurrency is capped at five.
 schemaVersion: 1
-updatedAt: 2026-09-24
+updatedAt: 2026-09-26
 providers:
   openrouter:
     maxConcurrent: 3
@@ -141,6 +141,7 @@ image:
     source: https://developers.openai.com/api/docs/guides/image-generation
     keywords:
       - image
+      - reference
     pricing:
       note: Automatic quality and generated image tokens determine the charge; a fixed
         image price is unavailable.
@@ -154,6 +155,7 @@ image:
     source: https://developers.openai.com/api/docs/guides/image-generation
     keywords:
       - image
+      - reference
     pricing:
       note: Automatic quality and generated image tokens determine the charge; a fixed
         image price is unavailable.
@@ -164,6 +166,7 @@ image:
     source: https://developers.openai.com/api/docs/guides/image-generation
     keywords:
       - image
+      - reference
     pricing:
       note: Automatic quality and generated image tokens determine the charge; a fixed
         image price is unavailable.
@@ -174,6 +177,7 @@ image:
     source: https://ai.google.dev/gemini-api/docs/pricing
     keywords:
       - image
+      - reference
     pricing:
       perImage: 0.101
       note: 2K image output price; prompt, thinking and other input tokens are
@@ -187,6 +191,7 @@ image:
     source: https://ai.google.dev/gemini-api/docs/models
     keywords:
       - image
+      - reference
     pricing:
       note: Price depends on your account or generated output; estimate unavailable.
     image:
@@ -198,6 +203,7 @@ image:
     source: https://fal.ai/models/fal-ai/flux-2
     keywords:
       - image
+      - reference
     pricing:
       note: Price depends on your account or generated output; estimate unavailable.
     image: *a4
@@ -207,6 +213,7 @@ image:
     source: https://fal.ai/models/fal-ai/nano-banana-2
     keywords:
       - image
+      - reference
     pricing:
       note: Price depends on your account or generated output; estimate unavailable.
     image: *a4
@@ -237,6 +244,87 @@ image:
     pricing:
       note: Price depends on your account or generated output; estimate unavailable.
     image: *a4
+  # Image-to-video models, marked by the `video` keyword: Animate images turns a still into a
+  # 5-second clip with one of these. perImage is the price of one 5-second clip.
+  - provider: fal
+    id: fal-ai/kling-video/v2.5-turbo/pro/image-to-video
+    name: Kling 2.5 Turbo Pro (image to video)
+    source: https://fal.ai/models/fal-ai/kling-video/v2.5-turbo/pro/image-to-video
+    keywords:
+      - video
+    pricing:
+      perImage: 0.35
+      note: Public fal.ai rate verified 2026-09-26, one 5-second clip.
+    image:
+      aspectRatios:
+        - 16:9
+        - 9:16
+  - provider: fal
+    id: fal-ai/wan-25-preview/image-to-video
+    name: Wan 2.5 (image to video, 720p)
+    source: https://fal.ai/models/fal-ai/wan-25-preview/image-to-video
+    keywords:
+      - video
+    pricing:
+      perImage: 0.5
+      note: Public fal.ai rate verified 2026-09-26, $0.10 a second at 720p, one 5-second clip.
+    image:
+      aspectRatios:
+        - 16:9
+        - 9:16
+  - provider: fal
+    id: fal-ai/bytedance/seedance/v1/pro/fast/image-to-video
+    name: Seedance 1.0 Pro Fast (image to video, 720p)
+    source: https://fal.ai/models/fal-ai/bytedance/seedance/v1/pro/fast/image-to-video
+    keywords:
+      - video
+    pricing:
+      perImage: 0.12
+      note: fal.ai bills by video tokens; about $0.12 for a 5-second 720p clip (verified
+        2026-09-26), more for larger frames.
+    image:
+      aspectRatios:
+        - 16:9
+        - 9:16
+  - provider: replicate
+    id: kwaivgi/kling-v2.5-turbo-pro
+    name: Kling 2.5 Turbo Pro (image to video)
+    source: https://replicate.com/kwaivgi/kling-v2.5-turbo-pro
+    keywords:
+      - video
+    pricing:
+      perImage: 0.35
+      note: Public Replicate rate verified 2026-09-26, $0.07 a second, one 5-second clip.
+    image:
+      aspectRatios:
+        - 16:9
+        - 9:16
+  - provider: replicate
+    id: wan-video/wan-2.5-i2v
+    name: Wan 2.5 (image to video, 720p)
+    source: https://replicate.com/wan-video/wan-2.5-i2v
+    keywords:
+      - video
+    pricing:
+      perImage: 0.5
+      note: Public Replicate rate verified 2026-09-26, $0.10 a second at 720p, one 5-second clip.
+    image:
+      aspectRatios:
+        - 16:9
+        - 9:16
+  - provider: replicate
+    id: bytedance/seedance-1-pro-fast
+    name: Seedance 1 Pro Fast (image to video, 720p)
+    source: https://replicate.com/bytedance/seedance-1-pro-fast
+    keywords:
+      - video
+    pricing:
+      perImage: 0.125
+      note: Public Replicate rate verified 2026-09-26, $0.025 a second at 720p, one 5-second clip.
+    image:
+      aspectRatios:
+        - 16:9
+        - 9:16
 tts:
   - provider: elevenlabs
     id: eleven_v3

package/dist/catalog/registry.js

@@ -1,5 +1,7 @@
+import { thinkingModes } from "../kernel/ports/llm.js";
 import { providerError } from "../kernel/ports/model.js";
 import { isLocalCliProvider } from "../slices/settings/model.js";
+import { referenceRefusal, takesReferenceImage } from "./schema.js";
 export function curateRegistry(registry, catalogue) {
     const requireModel = (provider, family, id) => {
         const models = catalogue.models(provider, family);
@@ -67,7 +69,7 @@ export function curateRegistry(registry, catalogue) {
                     ...m,
                     ...("llm" in m && m.llm.thinking
                         ? {
-                            thinkingModes: Object.keys(m.llm.thinking).filter((mode) => ["off", "low", "medium", "high", "xhigh"].includes(mode)),
+                            thinkingModes: Object.keys(m.llm.thinking).filter((mode) => thinkingModes.includes(mode)),
                         }
                         : {}),
                 })),
@@ -100,10 +102,19 @@ export function curateRegistry(registry, catalogue) {
                     ...port,
                     models: () => port.models(),
                     generate: async (request) => {
-                        if (!(await port.models()).some((model) => model.id === request.model))
+                        const selected = (await port.models()).find((model) => model.id === request.model);
+                        if (selected === undefined)
                             throw providerError({
                                 kind: "unsupported",
-                                message: "Image generation is not available in the Codex CLI on your computer. Update Codex CLI, or choose another image provider in the Edit tab under Edit project, then use Resume.",
+                                message: request.model === "codex-imagegen"
+                                    ? "Image generation is not available in the Codex CLI on your computer. Update Codex CLI, or choose another image provider in the Edit tab under Edit project, then use Resume."
+                                    : "The chosen Codex model for images is not in the Codex CLI's model list on your computer. Update Codex CLI, or pick another model for images: in the Edit tab, choose Edit project and change it under Providers, then use Resume.",
+                            });
+                        if (request.thinking !== undefined &&
+                            !selected.thinkingModes?.includes(request.thinking))
+                            throw providerError({
+                                kind: "unsupported",
+                                message: "The chosen Codex model does not support this effort for images. In the Edit tab, choose Edit project and change the images' Effort under Providers, then use Resume.",
                             });
                         return port.generate(request);
                     },
@@ -119,6 +130,16 @@ export function curateRegistry(registry, catalogue) {
                             kind: "unsupported",
                             message: "The chosen image model cannot make images in this video's shape. In the Edit tab, choose Edit project and change it under Providers, then use Resume.",
                         });
+                    if (request.thinking !== undefined)
+                        throw providerError({
+                            kind: "unsupported",
+                            message: "The chosen image model has no effort setting. In the Edit tab, choose Edit project and set the images' Effort back to Model default under Providers, then use Resume.",
+                        });
+                    if (request.reference !== undefined && !takesReferenceImage(model))
+                        throw providerError({
+                            kind: "unsupported",
+                            message: referenceRefusal,
+                        });
                     return port.generate(request);
                 },
             };

package/dist/catalog/runtime-models.js

@@ -6,7 +6,7 @@ export async function checkRuntimeModel(modelsFor, provider, family, id, thinkin
         const model = (await modelsFor(provider, family)).find((row) => row.id === id);
         if (model === undefined)
             return "missing";
-        if (family === "llm" && thinking !== undefined && !model.thinkingModes?.includes(thinking))
+        if (family !== "tts" && thinking !== undefined && !model.thinkingModes?.includes(thinking))
             return "thinking";
         return "available";
     }

package/dist/catalog/schema.js

@@ -1,4 +1,5 @@
 import { z } from "zod";
+import { thinkingModes } from "../kernel/ports/llm.js";
 import { providers } from "../slices/settings/model.js";
 const money = z.number().finite().nonnegative();
 const base = z.object({
@@ -29,7 +30,7 @@ export const llmModelSchema = base
         maxOutputTokens: z.number().int().positive().optional(),
         webSearch: z.boolean().default(false),
         thinking: z
-            .partialRecord(z.enum(["off", "low", "medium", "high", "xhigh"]), z
+            .partialRecord(z.enum(thinkingModes), z
             .object({
             budget: z.number().int().min(-1).optional(),
             level: z.enum(["minimal", "low", "medium", "high"]).optional(),
@@ -89,3 +90,21 @@ export const catalogueSchema = z
                 });
         }
 });
+// Image-to-video models live in the image list, under the image providers that run them,
+// marked by the `video` keyword; their perImage is the price of one clip. A catalogue that
+// older Slopify versions read keeps its shape, and the image pickers leave these out.
+export function isVideoModel(model) {
+    return model.keywords.includes("video");
+}
+// An image model that also takes an input image - so it can draw with the project's
+// establishing image as its visual reference - carries the `reference` keyword, the same way
+// and for the same reason the image-to-video models carry `video`.
+export function takesReferenceImage(model) {
+    return model.keywords.includes("reference");
+}
+// What the run says when the chosen image model cannot take the establishing image.
+export const referenceRefusal = "The chosen image model can't use an establishing image as a reference, so the images can't be drawn to match it. Choose an image model that can (Codex CLI, OpenAI, Google, or a fal.ai model listed with reference) under Images → Model on Play or in Edit project → Providers, or set Establishing image to Off in the Images section.";
+// The image-to-video models a provider offers, as Animate images lists them.
+export function videoModelsOf(catalogue, provider) {
+    return catalogue.image.filter((m) => m.provider === provider && m.enabled && !m.deprecated && isVideoModel(m));
+}

package/dist/catalog/store.js

@@ -2,7 +2,7 @@ import { existsSync, mkdirSync, readFileSync, renameSync, statSync, writeFileSyn
 import { join } from "node:path";
 import { parse } from "yaml";
 import { isLocalCliProvider } from "../slices/settings/model.js";
-import { catalogueSchema } from "./schema.js";
+import { catalogueSchema, isVideoModel } from "./schema.js";
 export const catalogueSource = "https://raw.githubusercontent.com/GentBajko/slopify/main/packages/app/src/assets/models.yaml";
 export function parseCatalogue(text) {
     if (Buffer.byteLength(text) > 1024 * 1024)
@@ -86,7 +86,7 @@ export function createCatalogueStore(deps) {
     }
     return {
         read,
-        models: (provider, family) => read()[family].filter((m) => m.provider === provider && m.enabled && !m.deprecated),
+        models: (provider, family) => read()[family].filter((m) => m.provider === provider && m.enabled && !m.deprecated && !isVideoModel(m)),
         status: () => ({ updatedAt: read().updatedAt, path, warning, source: catalogueSource }),
         refresh: () => {
             refreshing ??= refresh().finally(() => {

package/dist/catalog/validate.js

@@ -1,5 +1,7 @@
-import { usesNarrationPreparation, usesYoutubeDescription, } from "../slices/admission/rules.js";
+import { usesNarrationPreparation, usesReference, usesShorts, usesYoutubeDescription, } from "../slices/admission/rules.js";
 import { isLocalCliProvider } from "../slices/settings/model.js";
+import { usesAnimation } from "../slices/video/edit-settings.js";
+import { referenceRefusal, takesReferenceImage, videoModelsOf } from "./schema.js";
 export function modelFields(draft, catalogue) {
     if (!catalogue)
         return [];
@@ -10,7 +12,8 @@ export function modelFields(draft, catalogue) {
         draft.intro?.mode === "llm" ||
         draft.outro?.mode === "llm" ||
         usesNarrationPreparation(draft) ||
-        usesYoutubeDescription(draft);
+        usesYoutubeDescription(draft) ||
+        usesShorts(draft);
     const checks = [
         { field: "llm", family: "llm", choice: draft.llm, needed: needLlm },
         {
@@ -24,7 +27,8 @@ export function modelFields(draft, catalogue) {
             family: "image",
             choice: draft.images,
             needed: draft.sources.images === "generate" ||
-                ["from_prompt", "prompt_by_llm"].includes(draft.sources.thumbnail),
+                ["from_prompt", "prompt_by_llm"].includes(draft.sources.thumbnail) ||
+                usesShorts(draft),
         },
     ];
     for (const { field, family, choice, needed } of checks) {
@@ -41,16 +45,55 @@ export function modelFields(draft, catalogue) {
                 field,
                 message: "This model does not support that thinking setting. Choose another.",
             });
+        else if ("image" in model && choice.thinking)
+            fields.push({
+                field: "images.thinking",
+                message: "This image model has no effort setting. Set Effort back to Model default.",
+            });
         else if ("llm" in model && draft.sources.research === "generate" && !model.llm.webSearch)
             fields.push({
                 field,
                 message: "This model cannot search the web for research. Choose another model or turn Research off.",
             });
-        else if ("image" in model && !model.image.aspectRatios.includes(draft.format))
+        else if ("image" in model &&
+            (draft.sources.images === "generate" ||
+                ["from_prompt", "prompt_by_llm"].includes(draft.sources.thumbnail)) &&
+            !model.image.aspectRatios.includes(draft.format))
             fields.push({
                 field,
                 message: "This image model cannot make images in this video's shape. Choose another model.",
             });
+        // The shorts are always vertical, whatever shape the video is.
+        else if ("image" in model && usesShorts(draft) && !model.image.aspectRatios.includes("9:16"))
+            fields.push({
+                field,
+                message: "This image model cannot make the vertical (9:16) images Shorts need. Choose another image model, or turn Shorts off.",
+            });
+    }
+    // The establishing image needs a model that takes an input image; the catalogue says which.
+    // The Codex CLI always can, so only the keyed providers are looked up.
+    if (usesReference(draft) &&
+        draft.images !== undefined &&
+        !isLocalCliProvider(draft.images.provider)) {
+        const images = draft.images;
+        const model = catalogue.models(images.provider, "image").find((m) => m.id === images.model);
+        if (model !== undefined && !takesReferenceImage(model))
+            fields.push({ field: "reference.source", message: referenceRefusal });
+    }
+    // Animate images runs on the image provider's image-to-video models.
+    if (usesAnimation(draft) && draft.images !== undefined) {
+        const model = draft.videoEdit?.animateModel ?? "";
+        const offered = videoModelsOf(catalogue.read(), draft.images.provider);
+        if (offered.length === 0)
+            fields.push({
+                field: "videoEdit.animateModel",
+                message: "This image provider can't animate images. Choose fal.ai or Replicate as the image provider (on Play's Images rail, or in Edit project → Providers), or turn Animate images off.",
+            });
+        else if (model.trim() !== "" && !offered.some((one) => one.id === model))
+            fields.push({
+                field: "videoEdit.animateModel",
+                message: "This image-to-video model is no longer in Slopify's model list. Choose another under Animate images.",
+            });
     }
     return fields;
 }

package/dist/edge/docker-launch.js

@@ -62,6 +62,10 @@ async function main() {
         console.log(`Project files on this machine: ${result.projects}`);
         if (result.recovery)
             console.log(`Recovery volume retained: ${result.recovery}. Previous containers remain stopped.`);
+        for (const name of result.pruned.removed)
+            console.log(`Removed older recovery volume: ${name}`);
+        for (const problem of result.pruned.problems)
+            console.warn(problem);
     }
     finally {
         process.off("SIGINT", abort);

package/dist/edge/docker-projects/engine.js

@@ -4,6 +4,24 @@ import { setTimeout as delay } from "node:timers/promises";
 import { z } from "zod";
 import { containerSchema, digestSchema, } from "./state.js";
 import { contains } from "./tree.js";
+const uuid = /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/;
+// A recovery volume says which update, installation and container made it. Earlier launchers
+// set only the transaction label; `prune.ts` asks those for the update's own record instead.
+export function recoveryLabels(j) {
+    return {
+        "io.slopify.transaction": j.id,
+        "io.slopify.installation": j.installation,
+        "io.slopify.container": j.name,
+    };
+}
+// The id a recovery volume of `volume` carries in its name, or null for any other volume.
+export function recoveryId(volume, name) {
+    const prefix = `${volume}-recovery-`;
+    if (!name.startsWith(prefix))
+        return null;
+    const id = name.slice(prefix.length);
+    return uuid.test(id) ? id : null;
+}
 // Docker's exit code alone says nothing, so its last error line decides the advice: the daemon
 // being stopped and the socket being off-limits are by far the most common first-run failures.
 export function dockerFailure(args, result) {
@@ -294,9 +312,47 @@ export function dockerEngine(runner, signal, env) {
             const all = (await command(["volume", "ls", "--format", "{{.Name}}"])).split(/\s+/);
             if (all.includes(j.backup))
                 throw new Error("Recovery volume already exists; reconcile the journal before retrying.");
-            await command(["volume", "create", "--label", `io.slopify.transaction=${j.id}`, j.backup]);
+            await command([
+                "volume",
+                "create",
+                ...Object.entries(recoveryLabels(j)).flatMap(([key, value]) => [
+                    "--label",
+                    `${key}=${value}`,
+                ]),
+                j.backup,
+            ]);
             return digestSchema.parse(await helper(j.image, [...source(j.volume, null), ...mount("volume", j.backup, "/backup")], "snapshot"));
         },
+        recoveryVolumes: async (volume) => {
+            const names = (await command([
+                "volume",
+                "ls",
+                "--filter",
+                "label=io.slopify.transaction",
+                "--format",
+                "{{.Name}}",
+            ]))
+                .split(/\s+/)
+                .filter((name) => recoveryId(volume, name) !== null);
+            const found = [];
+            for (const name of names) {
+                const labels = z
+                    .record(z.string(), z.string())
+                    .nullable()
+                    .parse(JSON.parse(await command(["volume", "inspect", "--format", "{{json .Labels}}", name])));
+                found.push({
+                    name,
+                    transaction: labels?.["io.slopify.transaction"] ?? null,
+                    installation: labels?.["io.slopify.installation"] ?? null,
+                    container: labels?.["io.slopify.container"] ?? null,
+                });
+            }
+            return found;
+        },
+        // Never forced: Docker refuses a volume a container still mounts, and that refusal stands.
+        removeVolume: async (name) => {
+            await command(["volume", "rm", name]);
+        },
         restore: async (j) => {
             if (!j.backupDigest)
                 throw new Error("No verified private snapshot; automatic restore is unsafe.");