mirror of
https://github.com/superdesigndev/treg.git
synced 2026-10-02 03:24:35 +08:00
Merge pull request #562 from superdesigndev/codex/minimax-tts
feat(catalog): add MiniMax voice generation
This commit is contained in:
Taus
File diff suppressed because one or more lines are too long
@@ -137,6 +137,7 @@ sources:
|
||||
- src/treg/catalog/examples/minimax.video-gen.result.retrieve.json
|
||||
- src/treg/catalog/examples/minimax.video-gen.from_image.json
|
||||
- src/treg/catalog/examples/minimax.video-gen.task.status.json
|
||||
- src/treg/catalog/examples/minimax.voice-gen.voices.list.json
|
||||
- src/treg/catalog/openrouter.yaml
|
||||
- src/treg/catalog/openrouter.extended.yaml
|
||||
- src/treg/catalog/examples/openrouter.x.alibaba-wan-3-0.json
|
||||
@@ -575,8 +576,8 @@ The core AIGC generation rows pin `domain: models` too and carry PER-MODEL capab
|
||||
(`video-gen.hailuo.from_text`, proposed in their provider files) rather than the job-level
|
||||
`video-gen.from_text` family. Generation models are not interchangeable - a merged row comparing
|
||||
Hailuo with Wan or Seedance is a false comparison - so the job-level capabilities are deliberately
|
||||
memberless, reserved for hand-picked models (see capabilities.yaml). Both AI generation pages
|
||||
therefore render as ONE flat model wall; the same model reachable over several routes (MiniMax
|
||||
memberless, reserved for hand-picked models (see capabilities.yaml). The AI generation modality
|
||||
pages therefore render as flat model walls; the same model reachable over several routes (MiniMax
|
||||
direct, OpenRouter, Replicate all serve Hailuo) sits adjacent under model-led names, which is the
|
||||
comparison that actually means something. The per-model capability is the join key that lets those
|
||||
routes merge onto one row if that comparison is later curated. reAPI and PiAPI are the first pair
|
||||
@@ -607,6 +608,7 @@ platforms:
|
||||
web: "The web at large (backlinks, authority, traffic)"
|
||||
video-gen: {label: "Video generation", category: "AI generation"}
|
||||
image-gen: {label: "Image generation", category: "AI generation"}
|
||||
voice-gen: {label: "Voice generation", category: "AI generation"}
|
||||
```
|
||||
|
||||
Rules:
|
||||
@@ -617,8 +619,8 @@ Rules:
|
||||
this file. The validator accepts a capability that is either global or proposed in the same file.
|
||||
- Under `AI generation`, platform means the generated-media modality rather than a system that owns
|
||||
the data. The frozen vocabulary is `video-gen.from_text`, `video-gen.from_image`,
|
||||
`video-gen.task.status`, `image-gen.from_text`, and `image-gen.edit`; text-to-video and
|
||||
image-to-video stay separate because their required inputs and prices differ.
|
||||
`video-gen.task.status`, `image-gen.from_text`, `image-gen.edit`, and `voice-gen.from_text`;
|
||||
text-to-video and image-to-video stay separate because their required inputs and prices differ.
|
||||
|
||||
### `<service>.yaml`
|
||||
|
||||
@@ -673,7 +675,7 @@ endpoints:
|
||||
|
||||
### Async descriptors
|
||||
|
||||
`catalog_store._normalize` sets `cache: forbidden` for `image-gen` and `video-gen` endpoints,
|
||||
`catalog_store._normalize` sets `cache: forbidden` for `image-gen`, `video-gen`, and `voice-gen` endpoints,
|
||||
including synchronous generation, task/result utilities and generated extended rows. Their `kind`
|
||||
is unchanged. These requests must reach the provider, not replay shared-account task ids or media
|
||||
from an identical prompt. Other platforms retain their declared/default cache policy.
|
||||
@@ -783,6 +785,17 @@ the terminal values `Success`/`Fail`, then pass the returned `file_id` to
|
||||
`GET /v1/files/retrieve`. The v2 generation path serves the H3 family and is not a protocol upgrade
|
||||
for the Hailuo models in this listing.
|
||||
|
||||
MiniMax also supplies the first `voice-gen` rows through the same provider connection. Speech 2.8
|
||||
HD and Turbo are separate model rows over `POST /v1/t2a_v2`, each fixing its model plus
|
||||
`stream: false` and `output_format: url`; this keeps the response bounded and returns a 24-hour
|
||||
audio URL. The route is synchronous and billed per input character, so `_body_text_characters`
|
||||
scales the reserve from the provider-facing `text` value. The two rows remain marked `skipped`
|
||||
until a deliberate paid verification call is authorized; documentation provenance is enough for
|
||||
platform eligibility, but is not presented as live route evidence. The Voice generation Actions
|
||||
shelf exposes `minimax.voice-gen.voices.list` so callers can discover valid system voice IDs. Its
|
||||
request is fixed to `voice_type: system`: account-specific cloned and generated voices must not
|
||||
cross team boundaries when treg's shared MiniMax connection is used.
|
||||
|
||||
reAPI answers every submission with a bare `{id, status}` and reports the charge on the poll body
|
||||
(`usage.credits`, 1 credit = $0.001); video rows keep the file-level descriptor (`output.video_urls`)
|
||||
and image rows replace it whole for `output.image_urls`. PiAPI wraps its task routes in
|
||||
@@ -1046,7 +1059,11 @@ that slice: it bills one whole credit (~$0.0245) for one email or ten (observed
|
||||
10 emails"); `Catalog.advertised_usd` is what `catalog_search` / `catalog_get` put on
|
||||
`usd_per_call`. Settlement still reads `usd` and the derived email-count rule — display only.
|
||||
Akta bills 1.5 credits per 50 reviews the same `per` way. Without `per`, every one of those had
|
||||
to be either wrong or rounded into prose.
|
||||
to be either wrong or rounded into prose. A scalar `unit: character` is request-priced rather than
|
||||
page-priced: `_body_text_characters` counts the top-level JSON `text` string and multiplies the
|
||||
normalized per-character USD rate. Invalid JSON or a missing/empty string reserves one character,
|
||||
never zero; the normal request/envelope checks decide whether the provider served anything and
|
||||
`per_success` releases a rejected call.
|
||||
|
||||
**Three kinds of denomination convert, and they convert differently:**
|
||||
|
||||
|
||||
@@ -648,10 +648,13 @@ platform without opening it, and every field comes off the `/catalog/platforms`
|
||||
provider with no published rate stays silent. Note that `price_from` arrives as `null` *or* as an
|
||||
empty `{}`, and the empty object has to be normalised to null first — being truthy, it otherwise
|
||||
short-circuits the auth-kind branch and silently costs an OAuth-only platform its "free with your
|
||||
account".
|
||||
account". A grouped scalar rate such as MiniMax TTS supplies `display_usd` plus `display_unit`, so
|
||||
the card says `$0.60 / 10000 characters` instead of rounding the normalized per-character rate
|
||||
down to an unreadable number.
|
||||
|
||||
**Prices are unified USD.** Every price the marketplace displays — the card footer, the capability card's
|
||||
"from", and the per-endpoint cost chip — is the **server's computed `usd`** field on `cost` / `price_from`,
|
||||
"from", and the per-endpoint cost chip — comes from the server's price object on `cost` / `price_from`:
|
||||
normally its computed **`usd`**, or its equivalent grouped **`display_usd` / `display_unit`** pair,
|
||||
formatted by `usdNum`: two significant figures under a dollar (`$0.015`, `$0.00015`), cents at or above one.
|
||||
The FX table lives in the catalog (`fx.yaml`) so a rate refresh re-prices every surface at once, and the
|
||||
dashboard carries **no** conversion constant of its own — one here would drift from the CLI the moment the
|
||||
|
||||
@@ -36,10 +36,12 @@ One skill, three personas:
|
||||
- **consumer** — discover + call tools with no credentials locally. Teaches the agent-native
|
||||
**URL-passthrough** first: take the real upstream URL and prefix it with `{BASE}/call/`
|
||||
+ the `X-Treg-Token` header; `treg call <tool> <path>` is the CLI shorthand. Its
|
||||
"generate a video or an image" task teaches the async shape: `--await` for the CLI, the
|
||||
`X-Treg-Async` descriptor and lazy 30-60 s polling for MCP agents, the shell-timeout warning
|
||||
(video takes 1-5 minutes), reserve-then-settle money with refunds on failure, and expiring
|
||||
result URLs that treg never stores.
|
||||
generation task distinguishes synchronous MiniMax voice generation from asynchronous video and
|
||||
image generation. Voice callers discover current system voice IDs through
|
||||
`minimax.voice-gen.voices.list`, then call the HD or Turbo endpoint for a temporary audio URL.
|
||||
Video/image callers learn `--await`, the `X-Treg-Async` descriptor and lazy 30-60 s polling for
|
||||
MCP agents, the shell-timeout warning (video takes 1-5 minutes), reserve-then-settle money with
|
||||
releases on failure, and expiring result URLs that treg never stores.
|
||||
- **creator** — turn a local skill into a shared tool: `treg secret add`, `treg tool add` (single-key or
|
||||
`--bind` multi-credential), the `treg skill scaffold → push` bundle flow, and `treg oauth connect` for
|
||||
browser-consent tokens. Documents the two OAuth modes (auto-refresh vs manual) and the four auth shapes.
|
||||
|
||||
@@ -188,11 +188,11 @@ Notes:
|
||||
verify call each would have caught.
|
||||
- An endpoint with no published price is refused rather than served free; connect your own key.
|
||||
|
||||
## Task - generate a video or an image
|
||||
## Task - generate video, images, or voice
|
||||
|
||||
Generation models live in the catalog under the `video-gen` and `image-gen` platforms, one row per
|
||||
model per route (MiniMax direct, Replicate, OpenRouter), so the same model on two routes sits next
|
||||
to itself with both prices. Models are not interchangeable - you pick one; treg does not choose.
|
||||
Generation models live in the catalog under the `video-gen`, `image-gen`, and `voice-gen` platforms,
|
||||
one row per model per route, so the same model on two routes sits next to itself with both prices.
|
||||
Models are not interchangeable - you pick one; treg does not choose.
|
||||
|
||||
```bash
|
||||
treg catalog search "text to video" # every model, with prices
|
||||
@@ -200,9 +200,17 @@ treg catalog get minimax.video-gen.h3.generate # native params, model enum
|
||||
treg call minimax.video-gen.h3.generate --await --timeout 900 --data '{"model":"MiniMax-H3-Max",
|
||||
"content":[{"type":"text","text":"A paper boat drifts across a quiet pond at sunrise."}],
|
||||
"resolution":"480P","duration":5,"ratio":"16:9"}'
|
||||
treg call minimax.voice-gen.voices.list --data '{"voice_type":"system"}'
|
||||
treg catalog get minimax.voice-gen.speech-2-8-turbo
|
||||
treg call minimax.voice-gen.speech-2-8-turbo --data '{"model":"speech-2.8-turbo",
|
||||
"text":"A calm voice can make a complex idea feel simple.","stream":false,"output_format":"url",
|
||||
"voice_setting":{"voice_id":"English_expressive_narrator","speed":1,"vol":1,"pitch":0}}'
|
||||
```
|
||||
How it works:
|
||||
- **A generation call is an async task.** The submission returns a task id at once; `--await` polls
|
||||
- **Voice generation is synchronous.** MiniMax returns JSON containing a 24-hour audio URL. The
|
||||
catalog route fixes `stream:false` and `output_format:"url"`; use the voice-list action to discover
|
||||
valid system voice IDs, then choose HD or Turbo by endpoint id.
|
||||
- **A video or image generation call is an async task.** The submission returns a task id at once; `--await` polls
|
||||
the provider until it finishes and prints the **final response only** on stdout. stderr carries the
|
||||
task id, a resumable `treg call …` command (Ctrl-C loses the wait, never the task or the money),
|
||||
progress, and the result URL. Exit 0 = done, 2 = the provider failed the task, 3 = timed out
|
||||
|
||||
@@ -191,11 +191,11 @@ Notes:
|
||||
verify call each would have caught.
|
||||
- An endpoint with no published price is refused rather than served free; connect your own key.
|
||||
|
||||
## Task - generate a video or an image
|
||||
## Task - generate video, images, or voice
|
||||
|
||||
Generation models live in the catalog under the `video-gen` and `image-gen` platforms, one row per
|
||||
model per route (MiniMax direct, Replicate, OpenRouter), so the same model on two routes sits next
|
||||
to itself with both prices. Models are not interchangeable - you pick one; treg does not choose.
|
||||
Generation models live in the catalog under the `video-gen`, `image-gen`, and `voice-gen` platforms,
|
||||
one row per model per route, so the same model on two routes sits next to itself with both prices.
|
||||
Models are not interchangeable - you pick one; treg does not choose.
|
||||
|
||||
```bash
|
||||
treg catalog search "text to video" # every model, with prices
|
||||
@@ -203,9 +203,17 @@ treg catalog get minimax.video-gen.h3.generate # native params, model enum
|
||||
treg call minimax.video-gen.h3.generate --await --timeout 900 --data '{"model":"MiniMax-H3-Max",
|
||||
"content":[{"type":"text","text":"A paper boat drifts across a quiet pond at sunrise."}],
|
||||
"resolution":"480P","duration":5,"ratio":"16:9"}'
|
||||
treg call minimax.voice-gen.voices.list --data '{"voice_type":"system"}'
|
||||
treg catalog get minimax.voice-gen.speech-2-8-turbo
|
||||
treg call minimax.voice-gen.speech-2-8-turbo --data '{"model":"speech-2.8-turbo",
|
||||
"text":"A calm voice can make a complex idea feel simple.","stream":false,"output_format":"url",
|
||||
"voice_setting":{"voice_id":"English_expressive_narrator","speed":1,"vol":1,"pitch":0}}'
|
||||
```
|
||||
How it works:
|
||||
- **A generation call is an async task.** The submission returns a task id at once; `--await` polls
|
||||
- **Voice generation is synchronous.** MiniMax returns JSON containing a 24-hour audio URL. The
|
||||
catalog route fixes `stream:false` and `output_format:"url"`; use the voice-list action to discover
|
||||
valid system voice IDs, then choose HD or Turbo by endpoint id.
|
||||
- **A video or image generation call is an async task.** The submission returns a task id at once; `--await` polls
|
||||
the provider until it finishes and prints the **final response only** on stdout. stderr carries the
|
||||
task id, a resumable `treg call …` command (Ctrl-C loses the wait, never the task or the money),
|
||||
progress, and the result URL. Exit 0 = done, 2 = the provider failed the task, 3 = timed out
|
||||
|
||||
@@ -172,11 +172,11 @@ Notes:
|
||||
verify call each would have caught.
|
||||
- An endpoint with no published price is refused rather than served free; connect your own key.
|
||||
|
||||
## Task - generate a video or an image
|
||||
## Task - generate video, images, or voice
|
||||
|
||||
Generation models live in the catalog under the `video-gen` and `image-gen` platforms, one row per
|
||||
model per route (MiniMax direct, Replicate, OpenRouter), so the same model on two routes sits next
|
||||
to itself with both prices. Models are not interchangeable - you pick one; treg does not choose.
|
||||
Generation models live in the catalog under the `video-gen`, `image-gen`, and `voice-gen` platforms,
|
||||
one row per model per route, so the same model on two routes sits next to itself with both prices.
|
||||
Models are not interchangeable - you pick one; treg does not choose.
|
||||
|
||||
```bash
|
||||
treg catalog search "text to video" # every model, with prices
|
||||
@@ -184,9 +184,17 @@ treg catalog get minimax.video-gen.h3.generate # native params, model enum
|
||||
treg call minimax.video-gen.h3.generate --await --timeout 900 --data '{"model":"MiniMax-H3-Max",
|
||||
"content":[{"type":"text","text":"A paper boat drifts across a quiet pond at sunrise."}],
|
||||
"resolution":"480P","duration":5,"ratio":"16:9"}'
|
||||
treg call minimax.voice-gen.voices.list --data '{"voice_type":"system"}'
|
||||
treg catalog get minimax.voice-gen.speech-2-8-turbo
|
||||
treg call minimax.voice-gen.speech-2-8-turbo --data '{"model":"speech-2.8-turbo",
|
||||
"text":"A calm voice can make a complex idea feel simple.","stream":false,"output_format":"url",
|
||||
"voice_setting":{"voice_id":"English_expressive_narrator","speed":1,"vol":1,"pitch":0}}'
|
||||
```
|
||||
How it works:
|
||||
- **A generation call is an async task.** The submission returns a task id at once; `--await` polls
|
||||
- **Voice generation is synchronous.** MiniMax returns JSON containing a 24-hour audio URL. The
|
||||
catalog route fixes `stream:false` and `output_format:"url"`; use the voice-list action to discover
|
||||
valid system voice IDs, then choose HD or Turbo by endpoint id.
|
||||
- **A video or image generation call is an async task.** The submission returns a task id at once; `--await` polls
|
||||
the provider until it finishes and prints the **final response only** on stdout. stderr carries the
|
||||
task id, a resumable `treg call …` command (Ctrl-C loses the wait, never the task or the money),
|
||||
progress, and the result URL. Exit 0 = done, 2 = the provider failed the task, 3 = timed out
|
||||
|
||||
@@ -186,11 +186,11 @@ Notes:
|
||||
verify call each would have caught.
|
||||
- An endpoint with no published price is refused rather than served free; connect your own key.
|
||||
|
||||
## Task - generate a video or an image
|
||||
## Task - generate video, images, or voice
|
||||
|
||||
Generation models live in the catalog under the `video-gen` and `image-gen` platforms, one row per
|
||||
model per route (MiniMax direct, Replicate, OpenRouter), so the same model on two routes sits next
|
||||
to itself with both prices. Models are not interchangeable - you pick one; treg does not choose.
|
||||
Generation models live in the catalog under the `video-gen`, `image-gen`, and `voice-gen` platforms,
|
||||
one row per model per route, so the same model on two routes sits next to itself with both prices.
|
||||
Models are not interchangeable - you pick one; treg does not choose.
|
||||
|
||||
```bash
|
||||
treg catalog search "text to video" # every model, with prices
|
||||
@@ -198,9 +198,17 @@ treg catalog get minimax.video-gen.h3.generate # native params, model enum
|
||||
treg call minimax.video-gen.h3.generate --await --timeout 900 --data '{"model":"MiniMax-H3-Max",
|
||||
"content":[{"type":"text","text":"A paper boat drifts across a quiet pond at sunrise."}],
|
||||
"resolution":"480P","duration":5,"ratio":"16:9"}'
|
||||
treg call minimax.voice-gen.voices.list --data '{"voice_type":"system"}'
|
||||
treg catalog get minimax.voice-gen.speech-2-8-turbo
|
||||
treg call minimax.voice-gen.speech-2-8-turbo --data '{"model":"speech-2.8-turbo",
|
||||
"text":"A calm voice can make a complex idea feel simple.","stream":false,"output_format":"url",
|
||||
"voice_setting":{"voice_id":"English_expressive_narrator","speed":1,"vol":1,"pitch":0}}'
|
||||
```
|
||||
How it works:
|
||||
- **A generation call is an async task.** The submission returns a task id at once; `--await` polls
|
||||
- **Voice generation is synchronous.** MiniMax returns JSON containing a 24-hour audio URL. The
|
||||
catalog route fixes `stream:false` and `output_format:"url"`; use the voice-list action to discover
|
||||
valid system voice IDs, then choose HD or Turbo by endpoint id.
|
||||
- **A video or image generation call is an async task.** The submission returns a task id at once; `--await` polls
|
||||
the provider until it finishes and prints the **final response only** on stdout. stderr carries the
|
||||
task id, a resumable `treg call …` command (Ctrl-C loses the wait, never the task or the money),
|
||||
progress, and the result URL. Exit 0 = done, 2 = the provider failed the task, 3 = timed out
|
||||
|
||||
+13
-5
@@ -170,11 +170,11 @@ Notes:
|
||||
verify call each would have caught.
|
||||
- An endpoint with no published price is refused rather than served free; connect your own key.
|
||||
|
||||
## Task - generate a video or an image
|
||||
## Task - generate video, images, or voice
|
||||
|
||||
Generation models live in the catalog under the `video-gen` and `image-gen` platforms, one row per
|
||||
model per route (MiniMax direct, Replicate, OpenRouter), so the same model on two routes sits next
|
||||
to itself with both prices. Models are not interchangeable - you pick one; treg does not choose.
|
||||
Generation models live in the catalog under the `video-gen`, `image-gen`, and `voice-gen` platforms,
|
||||
one row per model per route, so the same model on two routes sits next to itself with both prices.
|
||||
Models are not interchangeable - you pick one; treg does not choose.
|
||||
|
||||
```bash
|
||||
treg catalog search "text to video" # every model, with prices
|
||||
@@ -182,9 +182,17 @@ treg catalog get minimax.video-gen.h3.generate # native params, model enum
|
||||
treg call minimax.video-gen.h3.generate --await --timeout 900 --data '{"model":"MiniMax-H3-Max",
|
||||
"content":[{"type":"text","text":"A paper boat drifts across a quiet pond at sunrise."}],
|
||||
"resolution":"480P","duration":5,"ratio":"16:9"}'
|
||||
treg call minimax.voice-gen.voices.list --data '{"voice_type":"system"}'
|
||||
treg catalog get minimax.voice-gen.speech-2-8-turbo
|
||||
treg call minimax.voice-gen.speech-2-8-turbo --data '{"model":"speech-2.8-turbo",
|
||||
"text":"A calm voice can make a complex idea feel simple.","stream":false,"output_format":"url",
|
||||
"voice_setting":{"voice_id":"English_expressive_narrator","speed":1,"vol":1,"pitch":0}}'
|
||||
```
|
||||
How it works:
|
||||
- **A generation call is an async task.** The submission returns a task id at once; `--await` polls
|
||||
- **Voice generation is synchronous.** MiniMax returns JSON containing a 24-hour audio URL. The
|
||||
catalog route fixes `stream:false` and `output_format:"url"`; use the voice-list action to discover
|
||||
valid system voice IDs, then choose HD or Turbo by endpoint id.
|
||||
- **A video or image generation call is an async task.** The submission returns a task id at once; `--await` polls
|
||||
the provider until it finishes and prints the **final response only** on stdout. stderr carries the
|
||||
task id, a resumable `treg call …` command (Ctrl-C loses the wait, never the task or the money),
|
||||
progress, and the result URL. Exit 0 = done, 2 = the provider failed the task, 3 = timed out
|
||||
|
||||
@@ -465,6 +465,23 @@ def _body_limit(body: bytes) -> int | None:
|
||||
return items
|
||||
|
||||
|
||||
def _body_text_characters(body: bytes) -> int:
|
||||
"""Count the provider-facing ``text`` field for character-priced generation calls.
|
||||
|
||||
The catalog price is already normalized to USD per character. Invalid JSON or a missing text
|
||||
field reserves one unit rather than zero; platform request validation/provider rejection still
|
||||
decides whether the call is relayed or charged.
|
||||
"""
|
||||
if body:
|
||||
try:
|
||||
document = json.loads(body)
|
||||
except (ValueError, UnicodeDecodeError):
|
||||
document = None
|
||||
if isinstance(document, dict) and isinstance(document.get("text"), str):
|
||||
return max(1, len(document["text"]))
|
||||
return 1
|
||||
|
||||
|
||||
def _platform_estimate_micro(cost: dict, query, body: bytes = b"") -> int:
|
||||
"""What one call is expected to cost the platform, in RAW micro-USD (no margin — ledger.reserve
|
||||
applies that). Rounds UP: a fraction of a micro-dollar is not representable and must not round to
|
||||
@@ -473,7 +490,9 @@ def _platform_estimate_micro(cost: dict, query, body: bytes = b"") -> int:
|
||||
if usd is None:
|
||||
return 0
|
||||
n = 1
|
||||
if cost.get("type") in ("per_result", "quota_rows") and cost.get("unit") in _ENTITY_UNITS:
|
||||
if cost.get("unit") == "character":
|
||||
n = _body_text_characters(body)
|
||||
elif cost.get("type") in ("per_result", "quota_rows") and cost.get("unit") in _ENTITY_UNITS:
|
||||
# Priced per INPUT entity, not per returned row: the page-size default below has no
|
||||
# meaning here and billed one-target calls 20x (seranking summary, serpstat overview —
|
||||
# 2026-09-05). The request names how many entities it asks about.
|
||||
|
||||
@@ -30,6 +30,8 @@ aliases:
|
||||
i2v: [image-to-video]
|
||||
texttovideo: [text-to-video]
|
||||
imagetovideo: [image-to-video]
|
||||
tts: [text-to-speech]
|
||||
texttospeech: [text-to-speech]
|
||||
# sales vocabulary — the catalog says "people" / "contact"; agents say "leads" / "prospects"
|
||||
# (logged miss 2026-08-28: `find leads` matched only the provider names leadsforge/predictleads)
|
||||
lead: [leads]
|
||||
|
||||
@@ -28,6 +28,7 @@ platforms:
|
||||
# --- AI generation: generated media modalities ------------------------------------------------
|
||||
video-gen: {label: "Video generation", category: "AI generation", summary: "Text-to-video and image-to-video across models, with prices side by side."}
|
||||
image-gen: {label: "Image generation", category: "AI generation", summary: "Text-to-image and prompt-based image editing across models."}
|
||||
voice-gen: {label: "Voice generation", category: "AI generation", summary: "Text-to-speech across voice models, with prices side by side."}
|
||||
|
||||
# --- Social: global social & video -----------------------------------------------------------
|
||||
tiktok: {label: "TikTok", category: "Social", featured: 1, summary: "Profiles, videos, comments, hashtags and search — plus your own account via OAuth."}
|
||||
@@ -135,6 +136,7 @@ capabilities:
|
||||
video-gen.task.status: "Check an async video generation task"
|
||||
image-gen.from_text: "Generate an image from a text prompt"
|
||||
image-gen.edit: "Edit or restyle an image with a prompt"
|
||||
voice-gen.from_text: "Generate speech from text"
|
||||
|
||||
creators.search: "Find influencers & creators by niche, country, followers & engagement on Instagram, TikTok, YouTube"
|
||||
# --- SEO / search -----------------------------------------------------------------------------
|
||||
|
||||
@@ -0,0 +1,24 @@
|
||||
{
|
||||
"system_voice": [
|
||||
{
|
||||
"voice_id": "English_expressive_narrator",
|
||||
"description": [
|
||||
"An expressive adult male voice with a British accent, perfect for engaging audiobook narration."
|
||||
],
|
||||
"voice_name": "Expressive Narrator",
|
||||
"created_time": "2025-01-01"
|
||||
},
|
||||
{
|
||||
"voice_id": "English_radiant_girl",
|
||||
"description": [
|
||||
"A radiant and lively young adult female voice with a general American accent, full of energy and brightness."
|
||||
],
|
||||
"voice_name": "Radiant Girl",
|
||||
"created_time": "2025-01-01"
|
||||
}
|
||||
],
|
||||
"base_resp": {
|
||||
"status_code": 0,
|
||||
"status_msg": "success"
|
||||
}
|
||||
}
|
||||
@@ -3,7 +3,7 @@ source:
|
||||
docs: https://platform.minimax.io/docs/api-reference/api-overview
|
||||
curated: 2026-09-02
|
||||
pricing_url: https://platform.minimax.io/docs/guides/pricing-paygo
|
||||
limits: "Video generation is asynchronous; poll every 10 seconds. Generated download URLs expire after about 9 hours."
|
||||
limits: "Video generation is asynchronous; poll every 10 seconds. Video download URLs expire after about 9 hours. Synchronous TTS accepts up to 10,000 characters; generated audio URLs expire after 24 hours."
|
||||
proposed_capabilities:
|
||||
video-gen.result.retrieve: "Retrieve completed video generation content"
|
||||
# Per-model capabilities: each is the join key for one model, so the row stands alone in the
|
||||
@@ -14,6 +14,9 @@ proposed_capabilities:
|
||||
video-gen.hailuo.from_image: "Animate an image into a Hailuo video"
|
||||
video-gen.h3.generate: "Generate an H3 (Hailuo 3) video from text, optionally with first/last frame images"
|
||||
image-gen.image-01.from_text: "Generate images with MiniMax image-01"
|
||||
voice-gen.speech-2-8-hd.generate: "MiniMax Speech 2.8 HD"
|
||||
voice-gen.speech-2-8-turbo.generate: "MiniMax Speech 2.8 Turbo"
|
||||
voice-gen.voices.list: "List available MiniMax voices"
|
||||
async:
|
||||
id_from: task_id
|
||||
poll:
|
||||
@@ -284,3 +287,130 @@ endpoints:
|
||||
confidence: documented
|
||||
note: "Price is per generated image."
|
||||
docs_url: https://platform.minimax.io/docs/api-reference/image-generation-t2i
|
||||
|
||||
- id: minimax.voice-gen.speech-2-8-hd
|
||||
capability: voice-gen.speech-2-8-hd.generate
|
||||
platform: voice-gen
|
||||
domain: models
|
||||
scope: any_account
|
||||
method: POST
|
||||
path: /v1/t2a_v2
|
||||
name: "MiniMax: Speech 2.8 HD"
|
||||
summary: "Generate high-definition speech from up to 10,000 characters of text."
|
||||
async: false
|
||||
platform_request:
|
||||
body.model: speech-2.8-hd
|
||||
body.stream: false
|
||||
body.output_format: url
|
||||
input:
|
||||
body:
|
||||
model: {type: string, required: true, enum: [speech-2.8-hd], example: speech-2.8-hd}
|
||||
text: {type: string, required: true, max: 10000, example: "A calm voice can make a complex idea feel simple.", note: "Fewer than 10,000 characters. Use <#x#> between speakable segments for a pause of x seconds."}
|
||||
stream: {type: boolean, required: true, enum: [false], example: false, note: "This catalog route is non-streaming so the relay returns one bounded JSON response."}
|
||||
output_format: {type: string, required: true, enum: [url], example: url, note: "The returned audio URL is valid for 24 hours."}
|
||||
voice_setting: {type: object, required: true, example: {voice_id: English_expressive_narrator, speed: 1, vol: 1, pitch: 0}, note: "voice_id is required. Optional controls include speed, vol, pitch, emotion, text_normalization and latex_read; see the API reference for ranges."}
|
||||
audio_setting: {type: object, required: false, example: {sample_rate: 32000, bitrate: 128000, format: mp3, channel: 1}, note: "Non-streaming output supports mp3, wav and flac."}
|
||||
language_boost: {type: string, required: false, example: auto, note: "Set auto for automatic language detection, or use one of MiniMax's documented language names."}
|
||||
pronunciation_dict: {type: object, required: false, example: {tone: ["Treg/tree-g"]}, note: "Custom pronunciation pairs in source/target form."}
|
||||
subtitle_enable: {type: boolean, required: false, default: false}
|
||||
subtitle_type: {type: string, required: false, default: sentence, enum: [sentence, word], example: sentence}
|
||||
voice_modify: {type: object, required: false, note: "Optional pitch, intensity, timbre and sound_effects controls."}
|
||||
bodyType: json
|
||||
test_request:
|
||||
body:
|
||||
model: speech-2.8-hd
|
||||
text: "Hello."
|
||||
stream: false
|
||||
output_format: url
|
||||
voice_setting: {voice_id: English_expressive_narrator, speed: 1, vol: 1, pitch: 0}
|
||||
audio_setting: {sample_rate: 32000, bitrate: 128000, format: mp3, channel: 1}
|
||||
expect: {json_path: base_resp.status_code, equals: 0}
|
||||
cost:
|
||||
type: per_success
|
||||
value: 1.00
|
||||
currency: USD
|
||||
per: 10000
|
||||
unit: character
|
||||
display: {unit: characters, grouped: true}
|
||||
source: docs
|
||||
source_url: https://platform.minimax.io/docs/guides/pricing-paygo#audio
|
||||
checked: 2026-09-18
|
||||
confidence: documented
|
||||
note: "MiniMax bills synchronous Speech 2.8 HD by input characters at $100 per million ($1 per 10,000). Failed requests are not charged."
|
||||
skipped: "Not live-called: synthesis spends real MiniMax credits, and no explicit paid verification was authorized for this change."
|
||||
docs_url: https://platform.minimax.io/docs/api-reference/speech-t2a-http
|
||||
|
||||
- id: minimax.voice-gen.speech-2-8-turbo
|
||||
capability: voice-gen.speech-2-8-turbo.generate
|
||||
platform: voice-gen
|
||||
domain: models
|
||||
scope: any_account
|
||||
method: POST
|
||||
path: /v1/t2a_v2
|
||||
name: "MiniMax: Speech 2.8 Turbo"
|
||||
summary: "Generate fast, cost-efficient speech from up to 10,000 characters of text."
|
||||
async: false
|
||||
platform_request:
|
||||
body.model: speech-2.8-turbo
|
||||
body.stream: false
|
||||
body.output_format: url
|
||||
input:
|
||||
body:
|
||||
model: {type: string, required: true, enum: [speech-2.8-turbo], example: speech-2.8-turbo}
|
||||
text: {type: string, required: true, max: 10000, example: "A calm voice can make a complex idea feel simple.", note: "Fewer than 10,000 characters. Use <#x#> between speakable segments for a pause of x seconds."}
|
||||
stream: {type: boolean, required: true, enum: [false], example: false, note: "This catalog route is non-streaming so the relay returns one bounded JSON response."}
|
||||
output_format: {type: string, required: true, enum: [url], example: url, note: "The returned audio URL is valid for 24 hours."}
|
||||
voice_setting: {type: object, required: true, example: {voice_id: English_expressive_narrator, speed: 1, vol: 1, pitch: 0}, note: "voice_id is required. Optional controls include speed, vol, pitch, emotion, text_normalization and latex_read; see the API reference for ranges."}
|
||||
audio_setting: {type: object, required: false, example: {sample_rate: 32000, bitrate: 128000, format: mp3, channel: 1}, note: "Non-streaming output supports mp3, wav and flac."}
|
||||
language_boost: {type: string, required: false, example: auto, note: "Set auto for automatic language detection, or use one of MiniMax's documented language names."}
|
||||
pronunciation_dict: {type: object, required: false, example: {tone: ["Treg/tree-g"]}, note: "Custom pronunciation pairs in source/target form."}
|
||||
subtitle_enable: {type: boolean, required: false, default: false}
|
||||
subtitle_type: {type: string, required: false, default: sentence, enum: [sentence, word], example: sentence}
|
||||
voice_modify: {type: object, required: false, note: "Optional pitch, intensity, timbre and sound_effects controls."}
|
||||
bodyType: json
|
||||
test_request:
|
||||
body:
|
||||
model: speech-2.8-turbo
|
||||
text: "Hello."
|
||||
stream: false
|
||||
output_format: url
|
||||
voice_setting: {voice_id: English_expressive_narrator, speed: 1, vol: 1, pitch: 0}
|
||||
audio_setting: {sample_rate: 32000, bitrate: 128000, format: mp3, channel: 1}
|
||||
expect: {json_path: base_resp.status_code, equals: 0}
|
||||
cost:
|
||||
type: per_success
|
||||
value: 0.60
|
||||
currency: USD
|
||||
per: 10000
|
||||
unit: character
|
||||
display: {unit: characters, grouped: true}
|
||||
source: docs
|
||||
source_url: https://platform.minimax.io/docs/guides/pricing-paygo#audio
|
||||
checked: 2026-09-18
|
||||
confidence: documented
|
||||
note: "MiniMax bills synchronous Speech 2.8 Turbo by input characters at $60 per million ($0.60 per 10,000). Failed requests are not charged."
|
||||
skipped: "Not live-called: synthesis spends real MiniMax credits, and no explicit paid verification was authorized for this change."
|
||||
docs_url: https://platform.minimax.io/docs/api-reference/speech-t2a-http
|
||||
|
||||
- id: minimax.voice-gen.voices.list
|
||||
kind: utility
|
||||
capability: voice-gen.voices.list
|
||||
platform: voice-gen
|
||||
scope: any_account
|
||||
method: POST
|
||||
path: /v1/get_voice
|
||||
name: "List available voices"
|
||||
summary: "List the current MiniMax system voices and their voice IDs."
|
||||
async: false
|
||||
platform_request:
|
||||
body.voice_type: system
|
||||
input:
|
||||
body:
|
||||
voice_type: {type: string, required: true, enum: [system], example: system, note: "This catalog action intentionally exposes system voices only, so account-specific cloned or generated voices never cross team boundaries on treg's shared MiniMax connection."}
|
||||
bodyType: json
|
||||
test_request: {body: {voice_type: system}}
|
||||
expect: {json_path: base_resp.status_code, equals: 0}
|
||||
cost: {type: free, value: 0, currency: USD, unit: call, note: "Voice discovery is a free account-management read."}
|
||||
verified: 2026-09-18
|
||||
example_response: examples/minimax.voice-gen.voices.list.json
|
||||
docs_url: https://platform.minimax.io/docs/api-reference/voice-management-get
|
||||
|
||||
+1
-1
@@ -230,7 +230,7 @@ class Settings(BaseSettings):
|
||||
platform_key_aviato: str = "" # Bearer key; $10 auto-top-up buys 1,000 credits
|
||||
platform_key_exa: str = "" # x-api-key; dollar-metered ($7/1k searches, $1/1k pages); settles from costDollars.total
|
||||
platform_key_cloro: str = "" # Bearer key (sk_live_…); Hobby metered rate $0.0004/credit; settles from X-Credits-Charged
|
||||
platform_key_minimax: str = "" # Bearer key for asynchronous Hailuo generation
|
||||
platform_key_minimax: str = "" # Bearer key for MiniMax voice, image and video generation
|
||||
platform_key_openrouter: str = "" # Bearer key for asynchronous routed generation
|
||||
platform_key_replicate: str = "" # Bearer token for official asynchronous models
|
||||
platform_key_reapi: str = "" # Bearer key; prepaid credits at $0.001, Seedance 2.5 + image models
|
||||
|
||||
@@ -660,7 +660,7 @@ def _normalize(raw: dict, provider: str, directory: Path) -> dict:
|
||||
# inherited from the file header unless the endpoint declares its own.
|
||||
# Generated media and its task/result utilities must never replay shared-account ids.
|
||||
# Enforce this for core and generated extended rows without changing their public kind.
|
||||
"cache": "forbidden" if platform in {"image-gen", "video-gen"} else raw.get("cache"),
|
||||
"cache": "forbidden" if platform in {"image-gen", "video-gen", "voice-gen"} else raw.get("cache"),
|
||||
"verified": str(verified) if verified else None,
|
||||
# {status, means} — a status the provider uses for "asked and answered: no result" (PDL
|
||||
# 404s a person it has no record of). Only endpoints with evidenced miss semantics carry
|
||||
|
||||
@@ -1673,7 +1673,7 @@ MINIMAX = OAuthProvider(
|
||||
auth_uri="", token_uri="", scopes={},
|
||||
client_id_setting="", client_secret_setting="",
|
||||
category="AI generation",
|
||||
summary="Generate images and create videos from text or source images.",
|
||||
summary="Generate voice, images, and videos from text or source images.",
|
||||
base_url="https://api.minimax.io",
|
||||
docs_url="https://platform.minimax.io/docs/api-reference/api-overview",
|
||||
probe_path="/v2/video_generation",
|
||||
|
||||
@@ -4094,7 +4094,7 @@ createApp({
|
||||
// catalog invents tomorrow still gets a shelf, sorted to the end.
|
||||
const order=['Enrichment','SEO/AEO','Social','Advertising','E-commerce','Reviews & Apps','AI generation','Community'];
|
||||
const hints={
|
||||
'AI generation':'video and image models, the same model over several routes priced side by side',
|
||||
'AI generation':'video, image and voice models, the same model over several routes priced side by side',
|
||||
'SEO/AEO':'rankings, keywords and backlinks — what search engines know, and what the answer engines say',
|
||||
'Social':'posts, profiles and comments, straight from the feeds',
|
||||
'Enrichment':'people and company records, resolved from an email or a domain',
|
||||
@@ -5772,7 +5772,10 @@ createApp({
|
||||
platPrice(pl){
|
||||
const raw=pl && pl.price_from;
|
||||
const pf=(raw && Object.keys(raw).length) ? raw : null;
|
||||
const paid=(pf && typeof pf.usd==='number') ? '$'+this.usdNum(pf.usd)+' / '+this.priceUnit(pf.type) : null;
|
||||
const paid=(pf && typeof pf.usd==='number')
|
||||
? '$'+this.usdNum(typeof pf.display_usd==='number' ? pf.display_usd : pf.usd)+' / '+
|
||||
(pf.display_unit || this.priceUnit(pf.type))
|
||||
: null;
|
||||
if(pf && pf.type==='free') return {free:true, text:'free with your account'};
|
||||
// An OAuth integration among the providers means the floor price is $0: the account you
|
||||
// connect IS the licence. Metered providers may serve the same platform (that rate moves to
|
||||
|
||||
+13
-7
@@ -251,13 +251,19 @@ unverified rows and guesses that one $0.006 verify call each would have caught.)
|
||||
Call a specific provider's endpoint when you need a particular one.
|
||||
|
||||
<!--/routed-->
|
||||
**Video and image generation - async calls.** The `video-gen` and `image-gen` platforms list one
|
||||
row per model per route (MiniMax direct, Replicate, OpenRouter); the same model on two routes sits
|
||||
side by side with both prices. Models are not interchangeable - you pick one; treg does not choose.
|
||||
A generation call is an **async task**: the submission returns a task id, and the response carries
|
||||
`X-Treg-Async`, a JSON descriptor saying where to poll (a catalog id plus the parameter that takes
|
||||
the task id, or an allow-listed URL from the submission), which status values mean done or failed,
|
||||
and where the result is (a path in the terminal response, or one more catalog call that fetches it).
|
||||
**Video, image and voice generation.** The `video-gen`, `image-gen`, and `voice-gen` platforms list
|
||||
one row per model per route; the same model on two routes sits side by side with both prices. Models
|
||||
are not interchangeable - you pick one; treg does not choose. MiniMax voice generation is
|
||||
synchronous: call `minimax.voice-gen.speech-2-8-hd` or
|
||||
`minimax.voice-gen.speech-2-8-turbo` with `stream:false` and `output_format:"url"`; the response
|
||||
contains a generated-audio URL valid for 24 hours and the price scales with input characters.
|
||||
Call `minimax.voice-gen.voices.list` with `{"voice_type":"system"}` to discover valid voice IDs.
|
||||
|
||||
**Video and image generation use async calls.** A generation call is an **async task**: the
|
||||
submission returns a task id, and the response carries `X-Treg-Async`, a JSON descriptor saying
|
||||
where to poll (a catalog id plus the parameter that takes the task id, or an allow-listed URL from
|
||||
the submission), which status values mean done or failed, and where the result is (a path in the
|
||||
terminal response, or one more catalog call that fetches it).
|
||||
|
||||
treg catalog get minimax.video-gen.h3.generate # native params, model enum, price table, descriptor
|
||||
treg call minimax.video-gen.h3.generate --await --timeout 900 --data '{...}'
|
||||
|
||||
+13
-5
@@ -151,11 +151,11 @@ Notes:
|
||||
<!--/routed-->
|
||||
- An endpoint with no published price is refused rather than served free; connect your own key.
|
||||
|
||||
## Task - generate a video or an image
|
||||
## Task - generate video, images, or voice
|
||||
|
||||
Generation models live in the catalog under the `video-gen` and `image-gen` platforms, one row per
|
||||
model per route (MiniMax direct, Replicate, OpenRouter), so the same model on two routes sits next
|
||||
to itself with both prices. Models are not interchangeable - you pick one; treg does not choose.
|
||||
Generation models live in the catalog under the `video-gen`, `image-gen`, and `voice-gen` platforms,
|
||||
one row per model per route, so the same model on two routes sits next to itself with both prices.
|
||||
Models are not interchangeable - you pick one; treg does not choose.
|
||||
|
||||
```bash
|
||||
treg catalog search "text to video" # every model, with prices
|
||||
@@ -163,9 +163,17 @@ treg catalog get minimax.video-gen.h3.generate # native params, model enum
|
||||
treg call minimax.video-gen.h3.generate --await --timeout 900 --data '{"model":"MiniMax-H3-Max",
|
||||
"content":[{"type":"text","text":"A paper boat drifts across a quiet pond at sunrise."}],
|
||||
"resolution":"480P","duration":5,"ratio":"16:9"}'
|
||||
treg call minimax.voice-gen.voices.list --data '{"voice_type":"system"}'
|
||||
treg catalog get minimax.voice-gen.speech-2-8-turbo
|
||||
treg call minimax.voice-gen.speech-2-8-turbo --data '{"model":"speech-2.8-turbo",
|
||||
"text":"A calm voice can make a complex idea feel simple.","stream":false,"output_format":"url",
|
||||
"voice_setting":{"voice_id":"English_expressive_narrator","speed":1,"vol":1,"pitch":0}}'
|
||||
```
|
||||
How it works:
|
||||
- **A generation call is an async task.** The submission returns a task id at once; `--await` polls
|
||||
- **Voice generation is synchronous.** MiniMax returns JSON containing a 24-hour audio URL. The
|
||||
catalog route fixes `stream:false` and `output_format:"url"`; use the voice-list action to discover
|
||||
valid system voice IDs, then choose HD or Turbo by endpoint id.
|
||||
- **A video or image generation call is an async task.** The submission returns a task id at once; `--await` polls
|
||||
the provider until it finishes and prints the **final response only** on stdout. stderr carries the
|
||||
task id, a resumable `treg call …` command (Ctrl-C loses the wait, never the task or the money),
|
||||
progress, and the result URL. Exit 0 = done, 2 = the provider failed the task, 3 = timed out
|
||||
|
||||
@@ -37,7 +37,7 @@ EP = "replicate.image-gen.flux-schnell"
|
||||
|
||||
def test_all_generation_catalog_entries_forbid_cache_including_extended():
|
||||
entries = [ep for ep in catalog_store.load().endpoints
|
||||
if ep["platform"] in {"image-gen", "video-gen"}]
|
||||
if ep["platform"] in {"image-gen", "video-gen", "voice-gen"}]
|
||||
assert any(".x." in ep["id"] for ep in entries)
|
||||
assert any(ep["id"] == "minimax.image-gen.from_text" for ep in entries)
|
||||
for ep in entries:
|
||||
|
||||
@@ -891,8 +891,13 @@ def test_ai_generation_taxonomy_and_chinese_alias_tokens_are_loaded():
|
||||
cat = cs.load()
|
||||
assert cat.platforms["video-gen"]["category"] == "AI generation"
|
||||
assert cat.platforms["image-gen"]["category"] == "AI generation"
|
||||
assert cat.platforms["voice-gen"] == {
|
||||
"label": "Voice generation",
|
||||
"category": "AI generation",
|
||||
"summary": "Text-to-speech across voice models, with prices side by side.",
|
||||
}
|
||||
assert {"video-gen.from_text", "video-gen.from_image", "video-gen.task.status",
|
||||
"image-gen.from_text", "image-gen.edit"} <= set(cat.capabilities)
|
||||
"image-gen.from_text", "image-gen.edit", "voice-gen.from_text"} <= set(cat.capabilities)
|
||||
text_to_video_zh = "\u6587\u751f\u89c6\u9891"
|
||||
assert cat.aliases[text_to_video_zh] == ["text-to-video"]
|
||||
assert cs._tokens(f"{text_to_video_zh} text-to-video") == [
|
||||
@@ -940,6 +945,38 @@ async def test_ai_generation_pages_keep_comparisons_curated_and_coverage_in_mode
|
||||
assert {"minimax.image-gen.from_text", "replicate.image-gen.flux-schnell",
|
||||
"reapi.image-gen.gemini-3-pro-image", "piapi.image-gen.gpt-image-2-5"} <= image_ids
|
||||
|
||||
voice = (await clients.get("/catalog/platforms/voice-gen")).json()
|
||||
assert {section["domain"] for section in voice["domains"]} == {"models"}
|
||||
voice_rows = [row for section in voice["domains"] for row in section["rows"]]
|
||||
assert {row["capability"] for row in voice_rows} == {
|
||||
"voice-gen.speech-2-8-hd.generate",
|
||||
"voice-gen.speech-2-8-turbo.generate",
|
||||
}
|
||||
voice_endpoints = [endpoint for row in voice_rows for endpoint in row["endpoints"]]
|
||||
assert {endpoint["id"] for endpoint in voice_endpoints} == {
|
||||
"minimax.voice-gen.speech-2-8-hd",
|
||||
"minimax.voice-gen.speech-2-8-turbo",
|
||||
}
|
||||
assert all(endpoint["provider"] == "minimax" for endpoint in voice_endpoints)
|
||||
catalog = cs.load()
|
||||
assert all(catalog.by_id[endpoint["id"]]["cache"] == "forbidden"
|
||||
for endpoint in voice_endpoints)
|
||||
|
||||
voice_full = (await clients.get(
|
||||
"/catalog/platforms/voice-gen?include_hidden=1")).json()
|
||||
assert voice_full["hidden_count"] == 1
|
||||
action_endpoints = {
|
||||
endpoint["id"]: endpoint
|
||||
for section in voice_full["domains"]
|
||||
for row in section["rows"]
|
||||
for endpoint in row["endpoints"]
|
||||
if endpoint["kind"] == "utility"
|
||||
}
|
||||
assert set(action_endpoints) == {"minimax.voice-gen.voices.list"}
|
||||
assert catalog.by_id["minimax.voice-gen.voices.list"]["platform_request"] == {
|
||||
"body.voice_type": "system"
|
||||
}
|
||||
|
||||
|
||||
def test_a_missing_catalog_directory_is_an_empty_catalog_not_a_crash(tmp_path):
|
||||
cat = cs.load(directory=tmp_path / "nope")
|
||||
|
||||
@@ -190,6 +190,10 @@ def test_the_category_list_comes_from_the_data_not_the_order_list():
|
||||
assert gone not in body, f"{gone} is no longer a catalog category — remove the reference"
|
||||
|
||||
|
||||
def test_ai_generation_shelf_names_all_three_media_modalities():
|
||||
assert "'AI generation':'video, image and voice models" in INDEX
|
||||
|
||||
|
||||
def test_tiles_are_grouped_by_category_on_every_capability_tab():
|
||||
""""All" is not a flat wall of tiles: it keeps the category headings, and a category tab is the
|
||||
same grouping filtered to one — so the page never loses its place."""
|
||||
@@ -309,7 +313,8 @@ def test_the_card_price_is_the_servers_computed_usd():
|
||||
would drift from the CLI the moment the rate table changed."""
|
||||
fn = INDEX[INDEX.index("platPrice(pl){") :][:1400]
|
||||
assert "typeof pf.usd==='number'" in fn
|
||||
assert "'$'+this.usdNum(pf.usd)+' / '+this.priceUnit(pf.type)" in fn
|
||||
assert "typeof pf.display_usd==='number' ? pf.display_usd : pf.usd" in fn
|
||||
assert "pf.display_unit || this.priceUnit(pf.type)" in fn
|
||||
assert "return null; }" in fn # priced but unpublished → say nothing, not "from —"
|
||||
# "from" is a floor: an OAuth provider among the platform's providers makes the floor $0, even
|
||||
# when metered providers publish a rate — that rate demotes to the tooltip.
|
||||
|
||||
@@ -53,6 +53,7 @@ def test_key_providers_appear_in_the_marketplace_listing():
|
||||
assert listing["zerobounce"]["category"] == "Enrichment"
|
||||
assert listing["zerobounce"]["auth_kind"] == "key"
|
||||
assert listing["minimax"]["category"] == "AI generation"
|
||||
assert listing["minimax"]["summary"] == "Generate voice, images, and videos from text or source images."
|
||||
assert listing["openrouter"]["auth_kind"] == "token"
|
||||
assert listing["replicate"]["base_url"] == "https://api.replicate.com/v1"
|
||||
assert "Enrichment" in P.CATEGORY_ORDER
|
||||
|
||||
@@ -1377,6 +1377,17 @@ def test_platform_estimate_normalizes_per_result_pricing():
|
||||
assert call_resolution._platform_estimate_micro({"type": "per_call", "usd": 0.0000005}, {}) == 1
|
||||
|
||||
|
||||
def test_platform_estimate_prices_text_to_speech_by_input_characters():
|
||||
"""MiniMax publishes TTS per character, so the request's text length—not a result-page
|
||||
default or a flat call price—sets the reserve. Unicode code points count as characters."""
|
||||
hd = {"type": "per_success", "unit": "character", "usd": 0.0001}
|
||||
turbo = {"type": "per_success", "unit": "character", "usd": 0.00006}
|
||||
assert call_resolution._platform_estimate_micro(hd, {}, b'{"text":"Hello."}') == 600
|
||||
assert call_resolution._platform_estimate_micro(turbo, {}, ' {"text":"Hi 👋"}'.encode()) == 240
|
||||
assert call_resolution._platform_estimate_micro(hd, {}, b'{"text":""}') == 100
|
||||
assert call_resolution._platform_estimate_micro(hd, {}, b'not-json') == 100
|
||||
|
||||
|
||||
def test_platform_estimate_counts_input_entities_not_a_page():
|
||||
"""A price per TARGET / DOMAIN / KEYWORD is per thing asked about, never per returned row: with
|
||||
no limit param the 20-row page default billed a one-target SE Ranking summary 20x ($0.358 for a
|
||||
|
||||
Reference in New Issue
Block a user