fix(video): seedance-2.5 takes reference clips; per-model reference ceiling

CHANGELOG.md

@@ -1,5 +1,30 @@
 # Changelog
 
+## Unreleased
+
+### Fixed — seedance-2.5 takes reference clips; the clip ceiling is per model
+
+The gateway turned on reference video/audio for seedance-2.5 on 2026-09-26
+(enterprise#297): its registry entry now carries `supportsReferenceMedia: true`
+and "reference video/audio clips up to 30s each (billed per reference
+second)". `blockrun_video` still refused `reference_videos` /
+`reference_audios` on 2.5 by name. It now forwards them, with the same limits
+as the 2.0 family: 1-3 clips of each type, audio only beside a reference image
+or video, never with frame seeds, up to 30 reference images.
+
+The reserve priced every clip at a flat 15.2s. The provider's ceiling is per
+model, and the gateway quotes each clip at its model's ceiling
+(`REFERENCE_CEILING_SECONDS`): 15.2s on seedance-2.0 / 2.0-fast / 2.0-mini,
+30.2s on 2.5. A 2.5 clip would have reserved half its price on the one rail
+where the reserve is the only budget control. One reference video now puts a
+5s 720p 2.5 render at ~$11.07 (~7x its ~$1.58 plain price). 1.5-pro still
+takes no clips, and the estimator now throws instead of borrowing another
+model's ceiling for it.
+
+Rail rules are unchanged: both wallet gateways (`blockrun.ai`,
+`sol.blockrun.ai`) still refuse every reference field with a 400 before
+quoting, so reference media stays account-rail only, 2.5 included.
+
 ## 0.53.1
 
 ### Fixed — seedance-2.5 first-and-last-frame is refused by name on Solana

README.md

@@ -252,9 +252,9 @@ Package managers have shown install size for decades. Almost no MCP server shows
 
 | Profile | Tools | Context |
 |---------|-------|---------|
-| `full` *(default)* | 19 | 14,014 |
+| `full` *(default)* | 19 | 14,046 |
 | `trading` | 8 | 5,411 |
-| `media` | 7 | 6,795 |
+| `media` | 7 | 6,827 |
 | `research` | 5 | 2,755 |
 | `chat` | 3 | 2,082 |
 
@@ -371,7 +371,7 @@ npx -y @blockrun/mcp@latest skills install --to ~/.codex/skills
 |------|-------------|------|
 | `blockrun_chat` | <!-- br:models.chatVisible -->82<!-- /br:models.chatVisible --> LLMs (GPT, Claude, Gemini, DeepSeek, Kimi K3, GLM, NVIDIA free tier, …) with `mode` tier routing | per token |
 | `blockrun_image` | Generate: openai/gpt-image-2, gpt-image-1, google/nano-banana(-2/-pro), xai/grok-imagine-image(-pro), zai/cogview-4, bytedance/seedream-5-pro. Edit: img2img, inpaint, fusion. | $0.015–0.15 |
-| `blockrun_video` | Sora 2 + xAI Grok Imagine Video + ByteDance Seedance 1.5/2.0-mini/2.0-fast/2.0/2.5 (720p + audio; 4K on 2.0, up to 30s on 2.5); RealFace asset → real-person video; [reference images, video and audio](docs/seedance-capabilities.md) on the account rail | $0.053–0.32/sec charged; reference clips billed at a 15.2s ceiling |
+| `blockrun_video` | Sora 2 + xAI Grok Imagine Video + ByteDance Seedance 1.5/2.0-mini/2.0-fast/2.0/2.5 (720p + audio; 4K on 2.0, up to 30s on 2.5); RealFace asset → real-person video; [reference images, video and audio](docs/seedance-capabilities.md) on the account rail | $0.053–0.32/sec charged; reference clips billed at the model's ceiling (15.2s on 2.0, 30.2s on 2.5) |
 | `blockrun_realface` | Enroll a real person (phone liveness) or AI character (Virtual Portrait) as a `ta_xxxx` asset for Seedance 2.0 / 2.0-fast / 2.0-mini video (not 2.5) | free; $0.01 to enroll |
 | `blockrun_music` | MiniMax music generation | per track |
 | `blockrun_speech` | ElevenLabs TTS (Flash/Turbo/Multilingual/v3, 8 voices) + ByteDance Seed Audio (prompt-directed) + cinematic sound effects; free voice listing | $0.05–0.10/1k chars |

docs/mcp-schema-overhead.md

@@ -16,9 +16,9 @@ Written 2026-09-01, verified against `@modelcontextprotocol/sdk` 1.29.0. Numbers
 
 | Profile | Tools | Context |
 |---------|-------|---------|
-| `full` *(default)* | 19 | 14,014 |
+| `full` *(default)* | 19 | 14,046 |
 | `trading` | 8 | 5,411 |
-| `media` | 7 | 6,795 |
+| `media` | 7 | 6,827 |
 | `research` | 5 | 2,755 |
 | `chat` | 3 | 2,082 |
 

docs/seedance-capabilities.md

@@ -32,22 +32,24 @@ below.
 | --- | --- | --- | --- |
 | Seedance 1.5-pro | Yes | No | No |
 | Seedance 2.0 / 2.0-fast / 2.0-mini | Yes | 1–9 | Image + video, image + audio, video + audio, or all three; 1–3 clips of each type |
-| Seedance 2.5 | Yes, except on the Solana wallet rail | 1–30 | No |
+| Seedance 2.5 | Yes, except on the Solana wallet rail | 1–30 | Same combinations as 2.0; 1–3 clips of each type, up to 30s each |
 
 `sol.blockrun.ai` has not yet added 2.5 to its first-and-last-frame list. It
 refuses `last_frame_url` on 2.5 with a `400` before quoting, while
 `blockrun.ai` quotes the same body; 2.0, 2.0-fast, 2.0-mini and 1.5-pro quote
 on both (unsigned probes, 2026-09-29). The tool refuses that one combination on
 Solana by name, so no payment is taken; use Base or the account rail for it.
 
-Seedance 2.5 takes reference **images** but no reference clips — the gateway
-registry carries `supportsReferenceImages: true` with
-`supportsReferenceMedia: false` for it, and the tool's guard matches.
+Seedance 2.5 takes reference **clips** as well as images since 2026-09-26
+(enterprise#297): the gateway registry carries `supportsReferenceMedia: true`
+for it, after its clip ceiling was probed at 30.2s and its per-second rate
+measured. Its clips may run up to 30s each, twice the 2.0 family's, and are
+billed accordingly (below).
 
 `image_url` means a first-frame seed. For a character or style image alongside
 a reference video, use `reference_image_urls`, not `image_url`. Frame seeding
-and reference mode are mutually exclusive. Reference audio on the 2.0 family
-requires at least one reference image or video. Upstream duration, size and
+and reference mode are mutually exclusive. Reference audio requires at least one
+reference image or video. Upstream duration, size and
 content constraints still apply; accepting a URL does not verify the remote
 file.
 
@@ -71,11 +73,20 @@ since `"reference"` is the only value upstream honours.
 Reference **clips** are billed **per reference second**, not per clip — measured
 against token360 on 2026-09-23 (blockrun#730): a reference second costs what an
 output second costs (~21,600 tokens against 21,780), with no per-clip
-component. Audio counts at 0.3x video.
+component. Re-measured on seedance-2.5 on 2026-09-25 (enterprise#297): the same
+rate. Audio counts at 0.3x video.
 
 The caller sends a URL and never declares the clip's length, so the gateway
-quotes **every clip at the model's 15.2s ceiling** and this tool reserves the
-same. Consequences worth knowing before you call:
+quotes **every clip at its model's ceiling** and this tool reserves the same.
+The ceiling is per model, probed from the provider's own validation error:
+
+| Model | Reference-clip ceiling |
+| --- | --- |
+| Seedance 2.0 / 2.0-fast / 2.0-mini | 15.2s |
+| Seedance 2.5 | 30.2s |
+
+A 2.5 clip therefore reserves twice the seconds of a 2.0 clip. Consequences
+worth knowing before you call:
 
 | Request | Reserved |
 | --- | --- |
@@ -85,6 +96,9 @@ same. Consequences worth knowing before you call:
 | seedance-2.0, 5s output, 3 videos + 3 audios | ~$14.54 |
 | seedance-2.0, 5s output at **4K**, 3 videos + 3 audios | ~$130.88 |
 | seedance-2.0-mini, 5s output at **480p**, 3 videos + 3 audios | ~$4.91 |
+| seedance-2.5, 5s output, no references | ~$1.58 |
+| seedance-2.5, 5s output, 1 reference video | ~$11.07 |
+| seedance-2.5, 5s output, 3 videos + 3 audios | ~$38.59 |
 
 A single reference clip therefore costs more than the render it conditions —
 roughly 4x a plain 5s 720p render, and about 24x for three videos plus three