@docyrus/animaguru 0.29.1 → 0.30.0

package/README.md

@@ -311,6 +311,23 @@ animaguru assets generate PROJECT_ID --type video --model fal/flux-3-max \
 
 `--keyframes` takes 1–10 `seconds,SHA` pairs, separated by semicolons, of this project's images or sprite art. Each is reached at its second, within `--seconds`; the clip has 24 frames a second and each keyframe needs a frame of its own. `--aspect-ratio auto` (default) follows the first keyframe, or give a ratio. FLUX 3 Max always works through the prompt before filming: `--prompt-expansion quality` asks for its thorough, slower level, any other value its light one. It takes `--seed`. Its dry runs are local estimates.
 
+CLI 0.30.0 adds HeyGen Video 1's other workflows (`heygen/heygen-video-1`). Besides image-to-video, it writes videos from text (`--workflow text-to-video`, with `--aspect-ratio`) and keeps the subjects of saved files (`--workflow reference-to-video`), for 5–15 whole seconds at 480p (default), 768p, 1080p or 2K. It makes its own sound, takes no `--soundtrack`, and takes `--seed`.
+
+```sh
+# A 30-second scene from text; clips over 15 seconds need prompt rewriting.
+animaguru assets generate PROJECT_ID --type video --model heygen/heygen-video-1 \
+  --workflow text-to-video --name "Potter" \
+  --prompt "A potter shapes a clay bowl on a wheel in a sunlit studio. One continuous shot, soft workshop sounds, no music" \
+  --seconds 30 --prompt-expansion balanced --resolution 1080p --aspect-ratio 16:9 --dry-run
+# Keep a product and a place from saved images.
+animaguru assets generate PROJECT_ID --type video --model heygen/heygen-video-1 \
+  --workflow reference-to-video --name "Harness check" \
+  --prompt "A supervisor wearing the harness in Image 1 stands still beside the line in Image 2 and speaks to camera" \
+  --references HARNESS_SHA,FACTORY_SHA --seconds 10 --resolution 768p --dry-run
+```
+
+From an image or text, HeyGen's clips run to 30 seconds while it rewrites the prompt: `--prompt-expansion balanced` asks for its fastest rewriting and `quality` its fuller one. Every second of a clip over 15 seconds costs its long-clip rate. At 1080p and 2K, text-to-video and reference-to-video are 16:9 or 9:16; from an image, the clip keeps the image's shape. `reference-to-video` takes up to nine `--references` images, three `--reference-videos` (HeyGen reads the first 5 seconds of each) and three `--audio-references`, twelve files in all, named Image 1, Video 1, Audio 1 in the prompt. It takes no `--reference`, `--end-reference` or `--middle-reference`, and `--aspect-ratio auto` (default) follows the first image or clip. HeyGen takes images of up to 5 MB, and clips and audio of up to 16 MB. Up to 5 seconds of each reference clip, and each reference image after the second, add to the cost; audio references are free.
+
 ### Vectors
 
 CLI 0.27.0 adds vectors: editable SVG files drawn by Ideogram on fal.ai (`fal/ideogram-vector`), for logos, icons, mascots and flat illustrations that stay sharp at any size and can be animated shape by shape.

package/dist/animaguru.js

@@ -19761,7 +19761,7 @@ var require_ignore2 = __commonJS({
 // cli/package.json
 var package_default = {
   name: "@docyrus/animaguru",
-  version: "0.29.1",
+  version: "0.30.0",
   description: "Agent-friendly AnimaGuru CLI built with CrustJS",
   type: "module",
   bin: {
@@ -45042,6 +45042,9 @@ var MAX_CAMERA_KEYFRAMES = 12;
 var MAX_CAMERA_TURNS = 32;
 var MAX_IMAGE_KEYFRAMES = 10;
 var KEYFRAME_FPS = 24;
+var HEYGEN_RESOLUTIONS = ["480p", "768p", "1080p", "2k"];
+var HEYGEN_WIDE = { "1080p": ["16:9", "9:16"], "2k": ["16:9", "9:16"] };
+var HEYGEN_LONG = { max: 30, creditsPerSecond: { "480p": 4.5, "768p": 7.5, "1080p": 10.5, "2k": 10.5 } };
 var H3_MAX = "minimax/h3-max";
 var FLUX_3_MAX = "blackforestlabs/flux-3-max";
 var h3Endpoints = (route) => ({ "image-to-video": `${route}/image-to-video`, "text-to-video": `${route}/text-to-video`, "extend-video": `${route}/extend-video` });
@@ -45061,11 +45064,18 @@ var SPRITE_VIDEO_MODELS = [
     id: "heygen/heygen-video-1",
     label: "HeyGen Video 1",
     provider: "heygen",
-    workflows: ["image-to-video"],
+    workflows: ["image-to-video", "text-to-video", "reference-to-video"],
     resolutions: ["480p", "768p"],
+    workflowOptions: {
+      "image-to-video": { resolutions: HEYGEN_RESOLUTIONS, long: HEYGEN_LONG },
+      "text-to-video": { resolutions: HEYGEN_RESOLUTIONS, long: HEYGEN_LONG, shapes: HEYGEN_WIDE },
+      "reference-to-video": { resolutions: HEYGEN_RESOLUTIONS, shapes: HEYGEN_WIDE, referencesOnly: true, creditsPerSecond: { "480p": 7, "768p": 10, "1080p": 18, "2k": 18 } }
+    },
     modes: ["cost"],
     lastFrame: false,
-    estimatedCreditsPerSecond: { "480p": { cost: 1 }, "768p": { cost: 1.5 } },
+    seed: true,
+    promptExpansion: PROMPT_EXPANSION_MODES,
+    estimatedCreditsPerSecond: { "480p": { cost: 2 }, "768p": { cost: 3.6 }, "1080p": { cost: 10 }, "2k": { cost: 10 } },
     seconds: { min: 5, max: 15 }
   },
   {
@@ -45145,21 +45155,29 @@ function spriteVideoModel(id2) {
 function videoResolutions(model, workflow = "image-to-video") {
   return model.workflowOptions?.[workflow]?.resolutions ?? model.resolutions;
 }
-function videoSeconds(model, workflow = "image-to-video") {
-  return model.workflowOptions?.[workflow]?.seconds ?? model.seconds;
+function videoSeconds(model, workflow = "image-to-video", rewrites = false) {
+  const options = model.workflowOptions?.[workflow];
+  const range = options?.seconds ?? model.seconds;
+  return rewrites && options?.long ? { ...range, max: options.long.max } : range;
 }
+var longVideo = (model, workflow, seconds2) => Boolean(model.workflowOptions?.[workflow]?.long) && Math.round(seconds2) > videoSeconds(model, workflow).max;
 var INCLUDED_REFERENCE_TOKENS = 4096;
 var imageReferenceTokens = (size) => size && size[0] > 0 && size[1] > 0 ? 32 * Math.ceil(32 * Math.max(...size) / Math.min(...size)) : 1824;
 var clipReferenceTokens = (seconds2, resolution) => Math.ceil(Math.max(0, seconds2) * (resolution === "480p" ? 2700 : 6900));
 var audioReferenceTokens = (seconds2) => Math.ceil(Math.max(0, seconds2) * 80);
 var DEFAULT_GENERATED_REFERENCES = 2;
+var HEYGEN_REFERENCE_CLIP_SECONDS = 5;
+var HEYGEN_FREE_REFERENCE_IMAGES = 2;
 function billedVideoSeconds(model, workflow, seconds2) {
   const s = videoSeconds(model, workflow).fractional ? seconds2 : Math.round(seconds2);
   if (workflow === "3d-to-video") return Math.max(5, s);
   return workflow === "lip-sync" && s > 15 ? s * 1.2 : s;
 }
 function videoCreditsPerSecond(model, p) {
-  const override = model.workflowOptions?.[p.workflow ?? "image-to-video"]?.creditsPerSecond?.[p.resolution];
+  const workflow = p.workflow ?? "image-to-video";
+  const options = model.workflowOptions?.[workflow];
+  const long = p.seconds !== void 0 && longVideo(model, workflow, p.seconds) ? options?.long?.creditsPerSecond[p.resolution] : void 0;
+  const override = long ?? options?.creditsPerSecond?.[p.resolution];
   if (override !== void 0) return override;
   const byMode = model.estimatedCreditsPerSecond[p.resolution] ?? model.estimatedCreditsPerSecond[model.resolutions[0]];
   return byMode[p.mode] ?? byMode[model.modes[0]] ?? 0;
@@ -45167,7 +45185,7 @@ function videoCreditsPerSecond(model, p) {
 function spriteVideoEstimate(model, p, extras2 = {}) {
   const seconds2 = billedVideoSeconds(model, p.workflow ?? "image-to-video", p.seconds);
   const beyond = Math.max(0, (extras2.referenceTokens ?? 0) - INCLUDED_REFERENCE_TOKENS) * 2e-3 + (extras2.generatedReferences ?? 0) * 3;
-  return Math.max(1, Math.ceil((videoCreditsPerSecond(model, p) * seconds2 + beyond) * UNITS_PER_CREDIT));
+  return Math.max(1, Math.ceil((videoCreditsPerSecond(model, p) * (seconds2 + (extras2.referenceSeconds ?? 0)) + beyond) * UNITS_PER_CREDIT));
 }
 
 // shared/sprites/generation-models.ts
@@ -47101,14 +47119,19 @@ function videoChecks(p, ctx) {
   if (!model.workflows.includes(p.workflow)) return issue2("workflow", "Choose a way of making video supported by the video model.");
   const w = p.workflow;
   const extend2 = w === "extend-video";
-  const range = videoSeconds(model, w);
+  const rewrites = p.promptExpansion !== "disabled";
+  const range = videoSeconds(model, w, rewrites);
   const seconds2 = p.seconds ?? defaultSeconds(model, w);
   const resolution = p.resolution ?? videoResolutions(model, w)[0];
+  const options = model.workflowOptions?.[w];
   const prompt = PROMPT[w];
   if (prompt === "none" && p.prompt) issue2("prompt", `${w} takes no prompt; leave it out.`);
   if (!prompt && p.prompt.length < 3) issue2("prompt", "Describe the video in at least 3 characters.");
   if ((w === "recast" || w === "3d-to-video") && p.prompt.length > 2e3) issue2("prompt", `${w} takes a prompt of at most 2,000 characters.`);
-  if (!lengthFromMedia(w) && (seconds2 < range.min || seconds2 > range.max || !range.fractional && !Number.isInteger(seconds2))) issue2("seconds", "Choose a duration supported by the video model.");
+  if (!lengthFromMedia(w) && (seconds2 < range.min || seconds2 > range.max || !range.fractional && !Number.isInteger(seconds2))) {
+    if (!rewrites && options?.long && Number.isInteger(seconds2) && seconds2 > range.max && seconds2 <= options.long.max) issue2("seconds", `Set promptExpansion to balanced or quality for a video over ${range.max} seconds.`);
+    else issue2("seconds", "Choose a duration supported by the video model.");
+  }
   if (!videoResolutions(model, w).includes(resolution)) issue2("resolution", "Choose a resolution supported by the video model.");
   if (p.mode && !model.modes.includes(p.mode)) issue2("mode", "Choose a mode supported by the video model.");
   for (const [field, workflows] of Object.entries(TAKES)) {
@@ -47131,16 +47154,22 @@ function videoChecks(p, ctx) {
   if (p.damage && p.style !== "vhs") issue2("damage", "Only the VHS style takes tape damage.");
   if (p.soundtrack && !model.audio) issue2("soundtrack", "This video model can't use a soundtrack.");
   if (p.soundtrack && w !== "lip-sync" && seconds2 < 2) issue2("seconds", "A video with a soundtrack must be at least 2 seconds long.");
-  if (p.audioReferences.length && !model.audio) issue2("audioReferences", "Only an extension takes audio references.");
+  if (p.audioReferences.length && !model.audio && w !== "reference-to-video") issue2("audioReferences", "Only an extension takes audio references.");
   if (p.promptExpansion !== "disabled" && !model.promptExpansion?.includes(p.promptExpansion)) issue2("promptExpansion", "This video model doesn't rewrite prompts.");
   if (p.seed !== void 0 && !model.seed) issue2("seed", "This video model doesn't take a seed.");
   if (p.seed !== void 0 && p.seed >= 2 ** 31 && (w === "lip-sync" || w === "insert-video")) issue2("seed", `${w} takes a seed below 2,147,483,648.`);
   const references = p.references ?? [];
   const unique = (shas2) => new Set(shas2).size === shas2.length;
   if (!unique(references) || !unique(p.referenceVideos ?? []) || !unique(p.audioReferences)) issue2("references", "Choose each reference once.");
+  const shapes = options?.shapes?.[resolution];
+  if (shapes && !shapes.includes(p.aspectRatio ?? (w === "text-to-video" ? "16:9" : "auto"))) issue2("aspectRatio", `At ${resolution === "2k" ? "2K" : resolution}, ${w} with this video model is ${shapes.join(" or ")}; choose one as the aspect ratio.`);
   if (w === "reference-to-video") {
+    if (options?.referencesOnly) {
+      for (const field of ["reference", "endReference", "middleReference"]) if (p[field]) issue2(field, "This video model's reference-to-video takes references only, without a first, last or middle frame.");
+    }
     if (references.length + (p.referenceVideos?.length ?? 0) + p.audioReferences.length > MAX_REFERENCE_FILES) issue2("references", `reference-to-video takes at most ${MAX_REFERENCE_FILES} reference files in all.`);
-    if (!references.length && !p.referenceVideos?.length && !p.audioReferences.length && !p.reference && !p.endReference) issue2("references", "reference-to-video needs reference images, videos or audio, or a first or last frame.");
+    const frames = !options?.referencesOnly && (p.reference || p.endReference);
+    if (!references.length && !p.referenceVideos?.length && !p.audioReferences.length && !frames) issue2("references", `reference-to-video needs reference images, videos or audio${options?.referencesOnly ? "" : ", or a first or last frame"}.`);
     if (Boolean(p.middleReference) !== (p.middleSeconds !== void 0)) issue2("middleSeconds", "A middle image needs middleSeconds, and middleSeconds a middle image.");
     if (p.middleReference && (!p.reference || !p.endReference)) issue2("middleReference", "A middle image needs the starting and ending images too.");
     if (p.middleReference && !["480p", "768p"].includes(resolution)) issue2("middleReference", "A middle image needs 480p or 768p.");
@@ -47291,6 +47320,12 @@ var imageEstimate = (p) => Math.ceil((130 + 40 * p.references.length) * spriteIm
 var vectorEstimate = (p) => p.workflow === "image-to-vector" ? 100 : p.quality === "low" ? 130 : p.quality === "high" ? p.references.length ? 320 : 200 : 160;
 function videoExtras(p, facts = {}) {
   const w = p.workflow;
+  if (spriteVideoModel(p.model).provider === "heygen") {
+    if (w !== "reference-to-video") return {};
+    const clips3 = (p.referenceVideos ?? []).reduce((sum, sha2) => sum + Math.min(HEYGEN_REFERENCE_CLIP_SECONDS, facts.seconds?.(sha2) ?? HEYGEN_REFERENCE_CLIP_SECONDS), 0);
+    const seconds2 = clips3 + Math.max(0, (p.references?.length ?? 0) - HEYGEN_FREE_REFERENCE_IMAGES);
+    return seconds2 ? { referenceSeconds: seconds2 } : {};
+  }
   if (w !== "reference-to-video" && w !== "insert-video" && w !== "3d-to-video") return {};
   const images = (p.references ?? []).reduce((sum, sha2) => sum + imageReferenceTokens(facts.size?.(sha2)), 0);
   if (w === "3d-to-video") return { referenceTokens: images + clipReferenceTokens(p.seconds, p.resolution), generatedReferences: p.references?.length ? 0 : p.generatedReferences ?? DEFAULT_GENERATED_REFERENCES };
@@ -50946,15 +50981,15 @@ Confirm code: ${device.user_code}`);
     { name: "transparent", type: "boolean", description: "Image: transparent background" },
     { name: "references", type: "string", description: "Comma-separated blob SHAs of this project's images or sprite art, named Image 1, Image 2\u2026 in the prompt. Image: up to 4 to follow, such as a character sheet. Vector from text: up to 5, the first leading. Video: subjects or looks to keep (reference-to-video and insert-video up to 9, 3d-to-video up to 8), or recast's new people, one photo each (up to 4)" },
     { name: "mask", type: "string", description: "Image: edit only these rectangles of Image 1 (the first --references image), as x,y,width,height fractions of its size from the top left, up to 8 separated by semicolons; the rest is kept" },
-    { name: "workflow", type: "string", choices: [...VIDEO_WORKFLOWS, ...VECTOR_WORKFLOWS], description: "Video: image-to-video (default), text-to-video or extend-video; H3 Max also reference-to-video, lip-sync, camera-controls, insert-video, recast, relight, 3d-to-video and style, FLUX 3 Max keyframes-to-video, as the model's videoProfiles allow. Vector: text-to-vector (default) or image-to-vector, which traces --source" },
+    { name: "workflow", type: "string", choices: [...VIDEO_WORKFLOWS, ...VECTOR_WORKFLOWS], description: "Video: image-to-video (default), text-to-video or extend-video; H3 Max also reference-to-video, lip-sync, camera-controls, insert-video, recast, relight, 3d-to-video and style, FLUX 3 Max keyframes-to-video, HeyGen Video 1 reference-to-video, as the model's videoProfiles allow. Vector: text-to-vector (default) or image-to-vector, which traces --source" },
     { name: "reference", type: "string", description: "Video: starting image blob SHA from this project's assets or sprites; the speaking face for lip-sync, the image the camera moves around for camera-controls. Model: the image or sprite art to build it from (image-to-3D); without one the prompt describes it" },
     { name: "end-reference", type: "string", description: "Video: ending image blob SHA; models with endFrameOnly accept it without --reference" },
     { name: "middle-reference", type: "string", description: "Reference-to-video: image blob SHA reached at --middle-seconds; needs --reference and --end-reference, 480p or 768p" },
     { name: "middle-seconds", type: "number", description: "Reference-to-video: when the middle image is reached, in seconds" },
     { name: "source", type: "string", description: "Blob SHA of a video asset in this project: the clip to extend, insert a scene into, recast or relight, or the 3D render 3d-to-video turns into footage. Vector image-to-vector: the image or sprite art to trace" },
     { name: "soundtrack", type: "string", description: "Video from an image or text: blob SHA of a music, speech or SFX asset pinned as the soundtrack (2 s or longer). Lip-sync: the speech to say (5 s\u201315 min), which sets the length" },
-    { name: "audio-references", type: "string", description: "Extension or reference-to-video: up to 3 comma-separated audio asset blob SHAs (2\u201315 s each, 15 s total) guiding the sound; refer to them as Audio 1, Audio 2\u2026 in the prompt" },
-    { name: "reference-videos", type: "string", description: "Reference-to-video or insert-video: up to 3 comma-separated video asset blob SHAs (2\u201315 s each, 15 s total) guiding motion, named Video 1, Video 2\u2026 in the prompt" },
+    { name: "audio-references", type: "string", description: "Extension or reference-to-video: up to 3 comma-separated audio asset blob SHAs (on fal 2\u201315 s each, 15 s total) guiding the sound; refer to them as Audio 1, Audio 2\u2026 in the prompt" },
+    { name: "reference-videos", type: "string", description: "Reference-to-video or insert-video: up to 3 comma-separated video asset blob SHAs (on fal 2\u201315 s each, 15 s total; HeyGen reads the first 5 s of each) guiding motion, named Video 1, Video 2\u2026 in the prompt" },
     { name: "camera", type: "string", description: 'Camera-controls: 2\u201312 keyframes time,azimuth,elevation,distance (time 0\u20131, angles in degrees, elevation \u221290\u201390) separated by semicolons, e.g. "0,0,0,1;1,90,15,1.2"' },
     { name: "keyframes", type: "string", description: 'Keyframes-to-video (FLUX 3 Max): 1\u201310 seconds,image-SHA pairs separated by semicolons, e.g. "0,SHA;2.5,SHA;5,SHA"; each image is reached at its second, on a frame of its own of the 24 fps clip, within --seconds' },
     { name: "insert-at", type: "number", description: "Insert-video: seconds into --source where the new scene starts (1.625\u201360)" },
@@ -50967,10 +51002,10 @@ Confirm code: ${device.user_code}`);
     { name: "generated-references", type: "number", description: "3d-to-video without --references: how many reference images it may draw, 1\u20138 (default 2)" },
     { name: "style", type: "string", choices: VIDEO_STYLES, description: "Style workflow: the preset look, at 768p for 5\u201315 whole seconds" },
     { name: "damage", type: "string", choices: VHS_DAMAGE_LEVELS, description: "VHS style: tape damage (fal's default medium)" },
-    { name: "seconds", type: "number", description: "Video or SFX duration in seconds (default 5); H3 takes fractions (6.5), FLUX 3 Max 5\u201320 whole seconds; extensions add this many seconds, insert-video's scene is 5\u201313; lip-sync, recast, relight and 3d-to-video take their input's length; SFX supports 0.5\u201330" },
+    { name: "seconds", type: "number", description: "Video or SFX duration in seconds (default 5); H3 takes fractions (6.5), FLUX 3 Max 5\u201320 whole seconds, HeyGen Video 1 5\u201315 (up to 30 from an image or text with --prompt-expansion balanced or quality); extensions add this many seconds, insert-video's scene is 5\u201313; lip-sync, recast, relight and 3d-to-video take their input's length; SFX supports 0.5\u201330" },
     { name: "resolution", type: "string", choices: VIDEO_RESOLUTIONS, description: "Video resolution supported by the selected model and workflow (default its first)" },
     { name: "mode", type: "string", choices: ["cost", "speed", "quality"], description: "Video generation mode supported by the selected model (default its first)" },
-    { name: "prompt-expansion", type: "string", choices: PROMPT_EXPANSION_MODES, description: "Video: how much the model may rewrite the prompt (default disabled); extensions treat any other value as on. FLUX 3 Max always reasons about the prompt: quality thoroughly, any other value lightly" },
+    { name: "prompt-expansion", type: "string", choices: PROMPT_EXPANSION_MODES, description: "Video: how much the model may rewrite the prompt (default disabled); extensions treat any other value as on. FLUX 3 Max always reasons about the prompt: quality thoroughly, any other value lightly. HeyGen Video 1 needs balanced or quality for clips over 15 s" },
     { name: "seed", type: "number", description: "Video or vector from text: seed for repeatable results, where the model takes one" },
     { name: "extend-output", type: "string", choices: EXTEND_OUTPUTS, description: "Extension: extended (default, source plus new footage) or continuation (a 1.625 s overlap, then the new footage)" },
     { name: "voice", type: "string", description: "Gemini voice name or ElevenLabs voice ID (defaults: Kore / George)" },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@docyrus/animaguru",
-  "version": "0.29.1",
+  "version": "0.30.0",
   "description": "Agent-friendly AnimaGuru CLI built with CrustJS",
   "type": "module",
   "bin": {

package/skills/animaguru/SKILL.md

@@ -62,7 +62,7 @@ animaguru assets generate PROJECT_ID --type speech --name "Welcome" \
   --prompt "Welcome, traveler." --voice Kore --delivery "Warm and welcoming" --yes
 ```
 
-Images use the staff-managed route automatically. Use `--model` to select a model for video, music, speech or sound effects; discover the choices with `catalog` and the server's configured subset with `assets list`. Asset defaults are per type, independently of sprite project defaults. Image backgrounds are retained unless `--transparent` is supplied. CLI 0.17.0 adds `--references`, `--mask` and the 16:9 and 9:16 image shapes. `--references SHA,SHA` (up to four) makes an image follow this project's saved images, uploads or sprite art. To keep a recurring character consistent, generate a character sheet once and reference its `file` SHA in every later image; refer to references in the prompt as Image 1, Image 2 and so on, and say what to take from each. `--aspect-ratio` is 1:1, 3:2, 2:3, 16:9 or 9:16; use 16:9 or 9:16 for video scenes. `--mask x,y,width,height[;…]` (up to eight rectangles, in fractions of Image 1 from the top left) edits only those parts of Image 1, the first reference, and keeps the rest and Image 1's shape. Look at the image before choosing the rectangles. Videos require an image asset's `file` SHA from `assets list`, or a character's `art` SHA from `projects get`, in the same project. Validate video options against the selected model's `videoProfiles` in `catalog`; omitted video settings use five seconds and that provider's first resolution/mode. MiniMax H3 Max and H3 Max Turbo (`fal/minimax-h3-max`, `fal/minimax-h3-max-turbo`) also take `--workflow text-to-video` (with `--aspect-ratio`) and `--workflow extend-video` (with `--source VIDEO_SHA`, 2K resolution, `--extend-output`, and up to three comma-separated `--audio-references`). From images, they accept `--end-reference` with or without `--reference` and a `--soundtrack AUDIO_SHA`. All workflows take `--prompt-expansion` and `--seed`. These need CLI 0.8.0 or newer. CLI 0.18.0 adds H3 Max's (not Turbo's) `--workflow reference-to-video` (keep a character across clips from up to nine `--references`, three `--reference-videos` and three `--audio-references`, named Image 1, Video 1, Audio 1 in the prompt; optional `--reference`, `--end-reference`, `--middle-reference` with `--middle-seconds`), `lip-sync` (the face in `--reference` says `--soundtrack` speech of 5 s–15 min; no `--prompt`; `--transcribe`), `camera-controls` (`--camera "time,azimuth,elevation,distance;…"`, 2–12 keyframes around `--reference`; without a prompt only the camera moves), `insert-video` (a 5–13 s scene inside `--source` from `--insert-at` to `--resume-at`, guided by `--references`/`--reference-videos`), `recast` (people in `--references` replace those in `--source`; pass `--subject-consent` only after each person agreed), `relight` (`--source` relit from a `--lighting` sphere image; no prompt), `3d-to-video` (a 3D render `--source` into footage, from `--references` or `--generated-references`) and `style` (`--style vhs|retro-toon-70s|low-poly|hand-drawn|16bit-pixel`, `--damage` for VHS, 768p, 5–15 whole seconds). Lip-sync, recast, relight and 3d-to-video take their input's length; their dry runs, like reference-to-video's and insert-video's, are server estimates because the price depends on the files. H3 takes fractional `--seconds` (6.5). CLI 0.21.0 adds FLUX 3 Max (`fal/flux-3-max`): image-to-video (with an optional `--end-reference`) and text-to-video for 5–20 whole seconds at 720p or 1080p with its own sound and no `--soundtrack`, and `--workflow keyframes-to-video`, which passes through 1–10 `--keyframes "SECONDS,SHA;…"` images, each reached at its second within `--seconds`. It always reasons about the prompt: `--prompt-expansion quality` thoroughly, other values lightly. For Gemini speech, `--prompt` is spoken verbatim and `--delivery` is a separate instruction. Music uses Lyria; the Gemini TTS models are for speech.
+Images use the staff-managed route automatically. Use `--model` to select a model for video, music, speech or sound effects; discover the choices with `catalog` and the server's configured subset with `assets list`. Asset defaults are per type, independently of sprite project defaults. Image backgrounds are retained unless `--transparent` is supplied. CLI 0.17.0 adds `--references`, `--mask` and the 16:9 and 9:16 image shapes. `--references SHA,SHA` (up to four) makes an image follow this project's saved images, uploads or sprite art. To keep a recurring character consistent, generate a character sheet once and reference its `file` SHA in every later image; refer to references in the prompt as Image 1, Image 2 and so on, and say what to take from each. `--aspect-ratio` is 1:1, 3:2, 2:3, 16:9 or 9:16; use 16:9 or 9:16 for video scenes. `--mask x,y,width,height[;…]` (up to eight rectangles, in fractions of Image 1 from the top left) edits only those parts of Image 1, the first reference, and keeps the rest and Image 1's shape. Look at the image before choosing the rectangles. Videos require an image asset's `file` SHA from `assets list`, or a character's `art` SHA from `projects get`, in the same project. Validate video options against the selected model's `videoProfiles` in `catalog`; omitted video settings use five seconds and that provider's first resolution/mode. MiniMax H3 Max and H3 Max Turbo (`fal/minimax-h3-max`, `fal/minimax-h3-max-turbo`) also take `--workflow text-to-video` (with `--aspect-ratio`) and `--workflow extend-video` (with `--source VIDEO_SHA`, 2K resolution, `--extend-output`, and up to three comma-separated `--audio-references`). From images, they accept `--end-reference` with or without `--reference` and a `--soundtrack AUDIO_SHA`. All workflows take `--prompt-expansion` and `--seed`. These need CLI 0.8.0 or newer. CLI 0.18.0 adds H3 Max's (not Turbo's) `--workflow reference-to-video` (keep a character across clips from up to nine `--references`, three `--reference-videos` and three `--audio-references`, named Image 1, Video 1, Audio 1 in the prompt; optional `--reference`, `--end-reference`, `--middle-reference` with `--middle-seconds`), `lip-sync` (the face in `--reference` says `--soundtrack` speech of 5 s–15 min; no `--prompt`; `--transcribe`), `camera-controls` (`--camera "time,azimuth,elevation,distance;…"`, 2–12 keyframes around `--reference`; without a prompt only the camera moves), `insert-video` (a 5–13 s scene inside `--source` from `--insert-at` to `--resume-at`, guided by `--references`/`--reference-videos`), `recast` (people in `--references` replace those in `--source`; pass `--subject-consent` only after each person agreed), `relight` (`--source` relit from a `--lighting` sphere image; no prompt), `3d-to-video` (a 3D render `--source` into footage, from `--references` or `--generated-references`) and `style` (`--style vhs|retro-toon-70s|low-poly|hand-drawn|16bit-pixel`, `--damage` for VHS, 768p, 5–15 whole seconds). Lip-sync, recast, relight and 3d-to-video take their input's length; their dry runs, like reference-to-video's and insert-video's, are server estimates because the price depends on the files. H3 takes fractional `--seconds` (6.5). CLI 0.21.0 adds FLUX 3 Max (`fal/flux-3-max`): image-to-video (with an optional `--end-reference`) and text-to-video for 5–20 whole seconds at 720p or 1080p with its own sound and no `--soundtrack`, and `--workflow keyframes-to-video`, which passes through 1–10 `--keyframes "SECONDS,SHA;…"` images, each reached at its second within `--seconds`. It always reasons about the prompt: `--prompt-expansion quality` thoroughly, other values lightly. CLI 0.30.0 adds HeyGen Video 1's (`heygen/heygen-video-1`) `--workflow text-to-video` (with `--aspect-ratio`) and `reference-to-video` (up to nine `--references`, three `--reference-videos`, of which it reads the first 5 s, and three `--audio-references`, named Image 1, Video 1, Audio 1 in the prompt; no `--reference`, `--end-reference` or `--middle-reference`), 1080p and 2K (16:9 or 9:16 from text or references), and `--seed`. From an image or text it runs to 30 whole seconds with `--prompt-expansion balanced` or `quality`, and every second past 15 costs more. For Gemini speech, `--prompt` is spoken verbatim and `--delivery` is a separate instruction. Music uses Lyria; the Gemini TTS models are for speech.
 
 For ElevenLabs speech, choose `eleven_v4` or `eleven_v4_turbo`, pass an ElevenLabs voice ID with `--voice` (George is the default), and put audio tags in `--prompt` rather than using `--delivery`. For sound effects, use `--type sfx`; the only model is `eleven_text_to_sound_v2`. SFX supports `--seconds` from 0.5 to 30 (default 5), `--loop`, and `--prompt-influence` from 0 to 1 (default 0.3). These additions require CLI 0.5.0 or newer; check `catalog` before using them with an installed CLI.