@heyamiko/amiko-cli 0.15.1 → 0.15.3

package/README.md

@@ -128,6 +128,7 @@ amiko create image "logo" --aspect 16:9 --model gpt-image-2     # widescreen, Op
 amiko create image "a sunset" --pay credits --yes               # pay with Amiko account credits (unified accounts)
 amiko create video "a whale in space" --resolution 768P --seconds 6
 amiko create video "animate cover" --first-frame https://.../cover.png --yes
+amiko create video "a slow pan over dunes at dusk" --model MiniMax-H3 --seconds 12 --resolution 2K --aspect 21:9 --yes
 amiko create video "dance clip" --model dreamina-seedance-2-0-fast-260128 \
   --reference-image https://.../ref1.png --reference-audio https://.../beat.mp3 --generate-audio --yes
 amiko create tts "Hello world" --voice 21m00Tcm4TlvDq8ikWAM
@@ -135,7 +136,7 @@ amiko create tts "Hello world" --provider minimax --model speech-2.8-turbo
 amiko create tts "Hello" --provider minimax --voice English_expressive_narrator --yes
 amiko voice clone ./me.mp3 --provider minimax
 amiko voice clone ./me.mp3 --provider elevenlabs
-amiko create music "lo-fi chill beat" --duration 30000
+amiko create music "lo-fi chill beat" --duration 30          # seconds
 amiko create music --lyrics "..." --instrumental               # instrumental
 amiko create sfx "thunder and heavy rain" --duration 5
 amiko create image "banner" --raw                               # raw JSON
@@ -163,21 +164,23 @@ Modes: `image`, `video`, `tts`, `music`, `sfx` (mirrors the platform Create Stud
 
 | Flag | Default | Description |
 |------|---------|-------------|
-| `--model <model>` | smart | Video model (`MiniMax-Hailuo-*`, `dreamina-seedance-2-0-fast-260128`, …). Default: `MiniMax-Hailuo-02` for prompt-only (T2V), `MiniMax-Hailuo-2.3-Fast` when `--first-frame` is given (I2V) |
-| `--resolution <res>` | `768P` | `512P`, `720P`, `768P`, `1080P` |
-| `--seconds <n>` | `6` | Clip length: `6` or `10` |
-| `--aspect <ratio>` | — | Aspect ratio (provider-dependent) |
+| `--model <model>` | smart | Video model (`MiniMax-Hailuo-*`, `MiniMax-H3`, `dreamina-seedance-2-0-fast-260128`, …). Default: `MiniMax-Hailuo-02` for prompt-only (T2V), `MiniMax-Hailuo-2.3-Fast` when `--first-frame` is given (I2V) |
+| `--resolution <res>` | `768P` | `512P`, `720P`, `768P`, `1080P`; `MiniMax-H3` only `768P` or `2K` |
+| `--seconds <n>` | `6` | Clip length: `6`/`10` (Hailuo), `4`/`6`/`8` (Veo), `4`/`6`/`8`/`10` (Grok, Seedance), any whole number `4`–`15` (`MiniMax-H3`) |
+| `--aspect <ratio>` | — | Aspect ratio (provider-dependent). `MiniMax-H3` text-to-video needs a concrete ratio (`21:9`, `16:9`, `4:3`, `1:1`, `3:4`, `9:16`) and defaults to `16:9` when omitted |
 | `--first-frame <url>` | — | Image-to-video: first frame URL or data URI |
 | `--last-frame <url>` | — | First-and-last-frame video: ending frame URL or data URI |
-| `--reference-image <url>` | — | Seedance multimodal reference image (repeatable, max 9) |
-| `--reference-video <url>` | — | Seedance multimodal reference video (repeatable, max 3) |
-| `--reference-audio <url>` | — | Seedance multimodal reference audio (repeatable, max 3) |
-| `--subject-reference <url>` | — | MiniMax S2V subject/character reference image |
+| `--reference-image <url>` | — | Seedance / `MiniMax-H3` reference image (repeatable, max 9; H3 surcharges images past the 5th) |
+| `--reference-video <url>` | — | Seedance / `MiniMax-H3` reference video (repeatable, max 3; H3 bills the input seconds too) |
+| `--reference-audio <url>` | — | Seedance / `MiniMax-H3` reference audio (repeatable, max 3; H3 needs an image or video alongside) |
+| `--subject-reference <url>` | — | MiniMax S2V / H3 subject/character reference image |
 | `--generate-audio` | — | Seedance: generate audio in the output |
 | `--no-generate-audio` | — | Seedance: output silent video |
 
 Reference flags accept URLs or data URIs (not local paths). Typical I2V flow: `create image` → `create status <jobId>` for the image URL → `create video --first-frame <url>`.
 
+**MiniMax-H3** (aliases `h3`, `hailuo-03`) is MiniMax's v2 video API and follows its own rules: a prompt is required in every mode (at most 7000 characters, checked locally), clips run any whole number of seconds from 4 to 15, only `768P` and `2K` exist, and it is billed **per second** (plus reference-video input seconds and a per-image surcharge past 5 reference images), so the quote scales with `--seconds`. It does text-to-video, first/last-frame image-to-video, and reference-to-video (`--reference-image`/`--reference-video`/`--reference-audio`), but a frame and a reference cannot be combined in one job, reference audio needs an image or video alongside it, and a mixed reference set is capped at 12 files. The CLI checks all of that before the quote and approval prompt, so a bad combination never costs a confirmation round-trip.
+
 **`amiko create tts <text>`**
 
 | Flag | Default | Description |
@@ -203,7 +206,7 @@ Examples: `amiko create tts "Hi" --provider minimax --yes` · `amiko create tts
 |------|---------|-------------|
 | `--model <model>` | `music-2.6` | `music_v1`, `music-2.6`, `music-cover` |
 | `--lyrics <lyrics>` | — | Lyrics (optional) |
-| `--duration <ms>` | — | Length in milliseconds |
+| `--duration <seconds>` | — | Track length in seconds, 1–300. `music-3.0` bills per second, so this sets the price |
 | `--instrumental` | — | Instrumental (no vocals) |
 
 **`amiko create sfx <text>`**
@@ -636,6 +639,12 @@ npm publish
 
 ## Changelog
 
+### 0.15.3
+
+- **`amiko create video --model MiniMax-H3`** (aliases `h3`, `hailuo-03`): MiniMax's v2 video model, already live in mpp-service and amiko-web, is now reachable from the CLI. H3 breaks every local rule the video command enforced for Hailuo, so it gets its own pre-flight: `--seconds` accepts any whole number 4–15 (was hard-coded `6|10`), `--resolution` must be `768P` or `2K` (case-insensitive, canonicalized on the wire), and text-to-video needs a concrete `--aspect` — the CLI defaults it to `16:9` and prints a dim note, because the server only rejects the missing ratio *after* the owner has approved the spend. Frame/reference runs are left alone (H3 infers the ratio from the input). All checks run before the quote so an invalid job never costs an approval round-trip.
+- **`--seconds` for non-H3 models widened from `6|10` to `4|6|8|10`**, matching amiko-web's own gate. The old check rejected `8`, which Veo and Seedance accept, and `4`, which Veo/Grok/Seedance accept; mpp still enforces the per-provider subset (Hailuo remains `6|10`). The error now points at `--model MiniMax-H3` for anything else.
+- Reference flags (`--reference-image/video/audio`, `--subject-reference`) document that they drive H3 reference-to-video too, and the CLI enforces H3's input-shape rules locally: a first/last frame cannot be mixed with reference inputs, reference audio needs an image or video alongside it, and at most 12 reference files in total.
+
 ### 0.14.2
 
 - **`--media drive:<docId>`** (new source for `post create` and `post comment`): attach a media file already in the twin's Drive — an upload, or a Create Studio generation, which `saveCreationToDrive` mirrors into the "Create Studio Files" folder. The id comes from `amiko drive list --json`. This closes the gap where the only way to post an *audio or video* file the twin already owned was to have kept the original `amiko create` URL: `--media` takes local paths for images only, and a Drive signed URL expires.