feat: support exact 17k+5 frame durations for vLLM-Omni

docs/guides/user-guide.md

@@ -35,6 +35,8 @@ For a self-hosted chat service, allow at least 8192 tokens of context: the
 application sends `max_tokens: 8192`, so a smaller `--max-model-len` (for
 example vLLM's 4096 default) rejects every request with HTTP 400.
 
+For vLLM-Omni video connections, enter fractional seconds directly, for example `6.583333`. The CLI accepts `opendirector video create --duration 6.583333` together with the other required options. Read `opendirector video create --help` and `opendirector model list` first; `inputRules.fractionalDuration` identifies support, and each output's duration list defines the allowed range. H3 can align output to its supported frame count, so inspect the actual result duration before assembling segments.
+
 For detailed video settings, see [Video model capability configuration (Chinese)](video-model-capabilities.md). When no explicit output specification is provided, use **Auto** or **Service default duration**.
 
 Automatic region selection in images and video matting require a model with segmentation capabilities. Image generation and editing still require the corresponding generation or editing model.

docs/guides/user-guide.zh-CN.md

@@ -47,6 +47,8 @@ HTTP 400。
 
 在设置的“通用”页中可切换主题、调整去除水印选项。水印设置仅影响后续生成的内容，实际效果取决于模型服务是否支持。
 
+vLLM-Omni 视频连接支持直接输入小数秒，例如 `6.583333`。CLI 可使用 `opendirector video create --duration 6.583333` 并填写其余必需参数。先运行 `opendirector video create --help` 和 `opendirector model list`；`inputRules.fractionalDuration` 表示支持小数秒，各输出规格的时长列表定义允许范围。H3 可能按模型支持的帧数对齐，拼接分段前应核对结果的实际时长。
+
 ## 制作第一支短片
 
 建议先完成一个镜头的生成与导出，再扩展到多段素材。

packages/core/src/opendirector/application/creation_generation.py

@@ -29,6 +29,7 @@
 from opendirector.domain.input_rules import input_rules
 from opendirector.domain.minimax_speech import is_minimax_speech
 from opendirector.domain.projects import ProjectError
+from opendirector.domain.rule_facts import duration_matches
 from opendirector.domain.video_output import apply_video_resolution
 from opendirector.ports.gateway import gateway
 from opendirector.integrations.http_settings import MODEL_JOB_CONTEXT
@@ -212,7 +213,7 @@ def _parameters(self, spec, config, node, records):
                 if node.get('videoMode') == 'continuation':
                     raise ProjectError('接续沿用上一段视频的画幅，请将比例设为自动。', 422)
                 caps = rules.get('videoCapabilities') or {}
-                choices = [item for item in caps.get('outputs', []) if seconds is None or seconds in item['durations']]
+                choices = [item for item in caps.get('outputs', []) if seconds is None or duration_matches(seconds, item['durations'], rules.get('fractionalDuration', False))]
                 if not choices:
                     raise ProjectError('指定视频比例前，请在模型接入中配置输出规格。', 422)
                 preferred = next((item for item in choices if item['resolution'] == caps.get('defaultResolution')), choices[0])

packages/core/src/opendirector/cli.py

@@ -434,7 +434,7 @@ def add_material(ctx: typer.Context, asset: Annotated[UUID, typer.Option(help="R
 def video_create(ctx: typer.Context,
                  model: Annotated[str, typer.Option(help="Video model configuration UUID or unique exact model name from opendirector model list.")],
                  prompt: Annotated[str, typer.Option(help="Video generation prompt.")],
-                 duration: Annotated[float, typer.Option(min=0, help="Video duration in seconds.")],
+                 duration: Annotated[float, typer.Option(min=0, help="Video duration in seconds, e.g. 6.583333. Fractional values require model inputRules.fractionalDuration; inspect model list for supported ranges. H3 may align the generated frame count.")],
                  x: Annotated[float, typer.Option(help="Canvas X coordinate. Choose using the current item positions and sizes.")],
                  y: Annotated[float, typer.Option(help="Canvas Y coordinate. Choose using the current item positions and sizes.")],
                  first_frame: Annotated[UUID | None, typer.Option(help="Source image canvas item for first-frame mode.")] = None,

packages/core/src/opendirector/domain/creation/generation.py

@@ -13,7 +13,7 @@
     FASTH3_MEDIA_REFERENCE_LIMIT, FRAME_IMAGE_LIMIT, LYRICS_MAX_LENGTH, MAX_REFERENCES,
     MUSIC_SECONDS_INITIAL, MUSIC_SECONDS_MAX, MUSIC_SECONDS_MIN, PROMPT_MAX_LENGTH,
     REFERENCE_MEDIA_SECONDS, STORED_VOICE_PATTERN, TEXT_DEFAULT_MAX_TOKENS, TEXT_MAX_TOKENS,
-    effective_video_mode, generate_audio_options,
+    duration_matches, effective_video_mode, generate_audio_options,
 )
 from .assets import AssetRef, MediaMetadata
 from .base import CanvasItemVersion, Contract, Digest, GenerationKind, NonNegative, ObjectVersion, PositiveInt, ProjectEntity, Revision, digest, unique
@@ -36,7 +36,7 @@ class ImageParameters(Contract):
 class VideoParameters(Contract):
     generateAudio: bool = Field(default=True, description='Include audio; value must be allowed by model inputRules.generateAudioOptions.')
     videoMode: VideoMode = Field(default='first_frame', description='Video generation input mode.')
-    durationSeconds: Annotated[float, Field(strict=True, ge=1, le=60)] | None = Field(default=None, description='Duration in seconds.')
+    durationSeconds: Annotated[float, Field(strict=True, ge=1, le=60)] | None = Field(default=None, description='Duration in seconds; fractional values require model inputRules.fractionalDuration. Generated frames may be aligned by the model.')
     aspectRatio: str | None = Field(default=None, pattern=r'^[1-9]\d?:[1-9]\d?$', description='Output width-to-height ratio.')
     resolution: str = Field(default='auto', pattern=r'^(?:auto|[A-Z0-9][A-Z0-9_.-]{0,23})$', description='Requested output resolution.')
 
@@ -490,5 +490,5 @@ def validate_generation(spec: GenerationSpec, model: ModelConfigPublic, referenc
     caps = rules.videoCapabilities
     if caps and params.durationSeconds is not None:
         choices = [output for output in caps.outputs if params.resolution == 'auto' or output.resolution == params.resolution]
-        if not any(params.durationSeconds in output.durations for output in choices):
+        if not any(duration_matches(params.durationSeconds, output.durations, rules.fractionalDuration) for output in choices):
             raise ValueError('当前时长与分辨率组合不受支持。')

packages/core/src/opendirector/domain/creation/models.py

@@ -46,6 +46,7 @@ class VoiceLibraryCapabilities(Contract):
 
 
 class InputRules(Contract):
+    fractionalDuration: bool = Field(default=False, exclude_if=lambda value: not value, description="Accept fractional video seconds within each output duration range.")
     profile: str = Field(min_length=1, max_length=100, description='Model capability profile.')
     voiceLibrary: VoiceLibraryCapabilities | None = Field(default=None, exclude_if=lambda value: value is None,
         description='Supported voice registry and adapter requirements; omitted for unsupported models.')

packages/core/src/opendirector/domain/rule_facts.py

@@ -102,6 +102,12 @@ def aspect_ratio_choices(spec):
     return ['auto']
 
 
+def duration_matches(seconds: float, durations: list[int] | tuple[int, ...], fractional: bool = False) -> bool:
+    if fractional:
+        return bool(durations) and min(durations) <= seconds <= max(durations)
+    return seconds in durations
+
+
 def resolution_choices(spec):
     """Sizes offered before submit. Callers report whether an image reference is connected."""
     kind = spec['kind']
@@ -122,7 +128,7 @@ def resolution_choices(spec):
     duration = spec.get('durationSeconds')
     if outputs is not None and duration is not None:
         choices = [label for label in choices
-                   if any(item['resolution'] == label and duration in item['durations'] for item in outputs)]
+                   if any(item['resolution'] == label and duration_matches(duration, item['durations'], spec.get('fractionalDuration', False)) for item in outputs)]
     return ['auto', *choices]
 
 
@@ -336,7 +342,14 @@ def derivation_vectors():
   return ["auto"];
 }
 
+export function durationMatches(seconds: number, durations: readonly number[], fractional = false): boolean {
+  return fractional
+    ? durations.length > 0 && seconds >= Math.min(...durations) && seconds <= Math.max(...durations)
+    : durations.includes(seconds);
+}
+
 export function resolutionChoices(input: {
+  readonly fractionalDuration?: boolean;
   readonly kind: string;
   readonly videoMode?: string | null;
   readonly aspectRatio?: string | null;
@@ -380,7 +393,7 @@ def derivation_vectors():
       ? choices.filter((label) =>
           outputs.some(
             (item) =>
-              item.resolution === label && item.durations.includes(duration),
+              item.resolution === label && durationMatches(duration, item.durations, input.fractionalDuration),
           ),
         )
       : choices;

packages/core/src/opendirector/domain/video_capabilities.py

@@ -5,6 +5,7 @@
 from pydantic import BaseModel, ConfigDict, Field, StrictInt, model_validator
 
 from opendirector.domain.projects import ProjectError
+from opendirector.domain.rule_facts import duration_matches
 from opendirector.domain.minimax_models import MINIMAX_MODELS, minimax_capabilities
 from opendirector.domain.kling_models import kling_capabilities
 from opendirector.domain.bigmodel_video import cogvideo_capabilities, validate_cogvideo_outputs
@@ -97,6 +98,8 @@ def capability_fields(config, rules):
         validate_cogvideo_outputs(capabilities)
     if rules['profile'] == 'kling-official':
         validate_kling_outputs(capabilities, config['modelId'])
+    if config.get('kind') == 'video' and config.get('protocol') == 'vllm_omni':
+        rules['fractionalDuration'] = True
     rules['videoCapabilities'] = capabilities
     rules['outputResolutions'] = [item['resolution'] for item in capabilities['outputs']] if capabilities else []
     return rules
@@ -153,9 +156,10 @@ def check_video_config(kind, protocol, adapter, capabilities):
 
 def validate_video_duration(config, seconds, resolution=None):
     from opendirector.domain.input_rules import input_rules
-    capabilities = input_rules(config).get('videoCapabilities')
+    rules = input_rules(config)
+    capabilities = rules.get('videoCapabilities')
     if capabilities:
         choices = [item for item in capabilities['outputs']
                    if not resolution or resolution in {'auto', '自动'} or item['resolution'] == resolution.upper()]
-        if not any(seconds in item['durations'] for item in choices):
+        if not any(duration_matches(seconds, item['durations'], rules.get('fractionalDuration', False)) for item in choices):
             raise ProjectError('当前模型的分辨率与时长组合不受支持，请重新选择；未提交生成。')

packages/core/src/opendirector/integrations/providers.py

@@ -179,7 +179,9 @@ async def generate(self, config, task, submitted, *, result_ready=None):
                                 files.append(('input_references' if fast else 'input_reference', (meta['name'], data, image_type(data) if source_kind == 'image' else meta['type'])))
                         parameters = task['parameters']
                         form = {'model': config['modelId'], 'prompt': prepared['prompt']}
-                        if 'duration' in parameters: form['seconds'] = format(parameters['duration'], 'g')
+                        if 'duration' in parameters:
+                            seconds = parameters['duration']
+                            form['seconds'] = str(int(seconds)) if float(seconds).is_integer() else str(seconds)
                         if 'size' in parameters: form['size'] = parameters['size']
                         if fast:
                             form.update({

src/backend/rules.ts

@@ -3,6 +3,7 @@ import { nodeByteSize } from "../canvas/uploadPolicy.ts";
 import type { CanvasNode, GenerationSettings } from "../canvas/model.ts";
 import {
   RULE_FACTS,
+  durationMatches,
   aspectRatioChoices,
   effectiveVideoMode,
   generateAudioOptions,
@@ -33,6 +34,7 @@ export interface VideoCapabilities {
 }
 
 export interface InputRules {
+  fractionalDuration?: boolean;
   profile: string;
   allowedKinds: string[];
   maxImages: number;
@@ -369,6 +371,7 @@ export function backendResolutions(
 ): string[] {
   const rules = config?.inputRules;
   return resolutionChoices({
+    fractionalDuration: rules?.fractionalDuration,
     kind: settings.mode,
     videoMode: settings.videoMode,
     aspectRatio: settings.aspectRatio,
@@ -483,7 +486,7 @@ export function serviceVideoSettings(
   const next = { ...settings };
   const caps = config.inputRules.videoCapabilities;
   const durations = backendDurations(config);
-  if (durations.length && !durations.includes(next.durationSeconds))
+  if (durations.length && !durationMatches(next.durationSeconds, durations, config.inputRules.fractionalDuration))
     next.durationSeconds = caps!.defaultDuration;
   if (
     next.videoMode === "continuation" &&
@@ -637,7 +640,7 @@ export function backendBlocker(
   const references = backendReferences(settings, [...incoming, ...pinned], config);
   if (settings.mode === "video") {
     const durations = backendDurations(config);
-    if (durations.length && !durations.includes(settings.durationSeconds))
+    if (durations.length && !durationMatches(settings.durationSeconds, durations, rules.fractionalDuration))
       return tNow("errors.rules.duration");
     if (
       rules.outputResolutions &&

src/canvas/DurationControl.tsx

@@ -9,7 +9,9 @@ export function DurationControl({
   disabled,
   onChange,
   scale,
+  fractional = false,
 }: {
+  fractional?: boolean;
   values: readonly number[];
   value: number;
   disabled: boolean;
@@ -52,14 +54,14 @@ export function DurationControl({
   const hint = (seconds: number) =>
     seconds <= minimum || seconds >= maximum
       ? limit
-      : !values.includes(seconds)
+      : !fractional && !values.includes(seconds)
         ? t("canvasNodes.duration.unsupported", {
             values: values.join(t("canvasNodes.duration.separator")),
           })
         : "";
   const choose = (seconds: number) => {
     if (!Number.isFinite(seconds) || !values.length || disabled) return;
-    const allowed = values.reduce((nearest, candidate) =>
+    const allowed = fractional ? Math.max(minimum, Math.min(maximum, seconds)) : values.reduce((nearest, candidate) =>
       Math.abs(candidate - seconds) < Math.abs(nearest - seconds)
         ? candidate
         : nearest,
@@ -97,7 +99,7 @@ export function DurationControl({
           aria-label={t("canvasNodes.duration.aria")}
           min={min}
           max={max}
-          step={1}
+          step={fractional ? "any" : 1}
           disabled={disabled || values.length <= 1}
           value={Math.max(min, Math.min(max, value))}
           aria-valuetext={t("canvasNodes.duration.value", { value })}
@@ -109,6 +111,7 @@ export function DurationControl({
             if (event.buttons === 1) pointerHint(event);
           }}
           onKeyDown={(event) => {
+            if (fractional) return;
             const index = Math.max(0, values.indexOf(value));
             const next =
               event.key === "Home"
@@ -142,9 +145,10 @@ export function DurationControl({
         <input
           type="number"
           aria-label={t("canvasNodes.duration.exactAria")}
+          title={fractional ? t("generation.duration.fractionalHint") : undefined}
           min={minimum}
           max={maximum}
-          step={1}
+          step={fractional ? "any" : 1}
           aria-description={limit}
           aria-describedby={notice ? noticeId : undefined}
           value={draft}
@@ -154,7 +158,7 @@ export function DurationControl({
             setDraft(next);
             // Persist a complete valid number immediately, so dismissing the
             // popover cannot discard it before a native blur is delivered.
-            if (next.trim() && values.includes(Number(next)))
+            if (next.trim() && Number.isFinite(Number(next)) && (values.includes(Number(next)) || fractional && Number(next) >= minimum && Number(next) <= maximum))
               choose(Number(next));
           }}
           onBlur={commit}

src/canvas/NodeComposer.tsx

@@ -788,6 +788,7 @@ export function NodeComposer({
           placement="above-anchor"
         >
           <DurationControl
+            fractional={!!config?.inputRules.fractionalDuration}
             values={durationValues}
             value={settings.durationSeconds}
             disabled={running}

src/contracts/creation.schema.json

@@ -6568,6 +6568,12 @@
       "InputRules": {
         "additionalProperties": false,
         "properties": {
+          "fractionalDuration": {
+            "default": false,
+            "description": "Accept fractional video seconds within each output duration range.",
+            "title": "Fractionalduration",
+            "type": "boolean"
+          },
           "profile": {
             "description": "Model capability profile.",
             "maxLength": 100,
@@ -14301,7 +14307,7 @@
               }
             ],
             "default": null,
-            "description": "Duration in seconds.",
+            "description": "Duration in seconds; fractional values require model inputRules.fractionalDuration. Generated frames may be aligned by the model.",
             "title": "Durationseconds"
           },
           "aspectRatio": {
@@ -21932,6 +21938,12 @@
       "InputRules": {
         "additionalProperties": false,
         "properties": {
+          "fractionalDuration": {
+            "default": false,
+            "description": "Accept fractional video seconds within each output duration range.",
+            "title": "Fractionalduration",
+            "type": "boolean"
+          },
           "profile": {
             "description": "Model capability profile.",
             "maxLength": 100,
@@ -29961,7 +29973,7 @@
               }
             ],
             "default": null,
-            "description": "Duration in seconds.",
+            "description": "Duration in seconds; fractional values require model inputRules.fractionalDuration. Generated frames may be aligned by the model.",
             "title": "Durationseconds"
           },
           "aspectRatio": {

src/contracts/generated.ts

@@ -665,6 +665,7 @@ export interface ImageParametersInput extends ContractInput {
 }
 
 export interface InputRulesInput extends ContractInput {
+  readonly fractionalDuration?: boolean;
   readonly profile: string;
   readonly voiceLibrary?: VoiceLibraryCapabilitiesInput | null;
   readonly maxPromptCharacters?: number | null;
@@ -2240,6 +2241,7 @@ export interface ImageParameters extends Contract {
 }
 
 export interface InputRules extends Contract {
+  readonly fractionalDuration?: boolean;
   readonly profile: string;
   readonly voiceLibrary?: VoiceLibraryCapabilities | null;
   readonly maxPromptCharacters?: number | null;

src/contracts/ruleFacts.generated.ts

@@ -76,7 +76,14 @@ export function aspectRatioChoices(input: {
   return ["auto"];
 }
 
+export function durationMatches(seconds: number, durations: readonly number[], fractional = false): boolean {
+  return fractional
+    ? durations.length > 0 && seconds >= Math.min(...durations) && seconds <= Math.max(...durations)
+    : durations.includes(seconds);
+}
+
 export function resolutionChoices(input: {
+  readonly fractionalDuration?: boolean;
   readonly kind: string;
   readonly videoMode?: string | null;
   readonly aspectRatio?: string | null;
@@ -120,7 +127,7 @@ export function resolutionChoices(input: {
       ? choices.filter((label) =>
           outputs.some(
             (item) =>
-              item.resolution === label && item.durations.includes(duration),
+              item.resolution === label && durationMatches(duration, item.durations, input.fractionalDuration),
           ),
         )
       : choices;

src/i18n/messages/en-US/generation.ts

@@ -1,5 +1,6 @@
 /** Composer, generation state, model choice and generation rules. */
 export const generation = {
+  "generation.duration.fractionalHint": "Fractional seconds are supported. H3 may align the output to its supported frame count.",
   "generation.bigmodel.watermarkWaiver":
     "BigModel requires a signed waiver to remove watermarks. Complete it in Personal Center → Security Management → Watermark Removal Management; otherwise outputs may still contain watermarks.",
   "generation.model.unconfigured":