Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions docs/architecture/unified-project.md
Original file line number Diff line number Diff line change
Expand Up @@ -148,4 +148,6 @@ UI、Agent 和后台完成写入共享可恢复变更通道。通知提示对象

模型接入使用能力契约和协议适配器,参数依次贯穿表单、生成配置、实际执行和历史依据。无法支持的参数明确限制,不静默忽略或降级。

生成结果保留服务端返回的原始文件,不自动裁剪、丢帧、重编码或替换音轨。请求时长与帧数属于执行输入,输出媒体事实依据实际文件记录;模型对齐产生的差异如实保留。片段范围通过时间线显式编辑,转码和音轨调整产生独立派生素材,保留原始结果及来源关系。首次下载、失败恢复及各操作入口遵循同一规则。

浏览器与后台的执行职责必须明确。Agent 可以修改时间线,不代表后台已经能够导出视频;只有存在可调用渲染服务并验证真实输出,才可声称支持关闭浏览器后的自主导出。浏览器合成也可以作为明确需要客户端参与的能力保留。
2 changes: 2 additions & 0 deletions docs/guides/user-guide.md
Original file line number Diff line number Diff line change
Expand Up @@ -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.
Expand Down
2 changes: 2 additions & 0 deletions docs/guides/user-guide.zh-CN.md
Original file line number Diff line number Diff line change
Expand Up @@ -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。
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -29,10 +29,12 @@
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
from opendirector.integrations.fasth3 import video_parameters
from opendirector.integrations.vllm_omni import FPS, video_frame_parameters
from opendirector.integrations.model_adapters import capabilities_of
from opendirector.integrations.providers import GenerationProviders
from opendirector.media.filenames import with_extension
Expand Down Expand Up @@ -195,9 +197,6 @@ def _parameters(self, spec, config, node, records):
seconds = intent.get('durationSeconds')
if rules['profile'] == 'fasth3-omni':
params = video_parameters(node, records, duration_seconds=seconds)
# FastH3's wire contract sends duration; its derived frame
# estimate is not an explicitly submitted execution parameter.
params.pop('num_frames', None)
params['seed'] = new_seed()
elif rules['profile'] == 'standard-video':
params = {}
Expand All @@ -212,7 +211,7 @@ 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])
Expand All @@ -226,9 +225,8 @@ def _parameters(self, spec, config, node, records):
params['ratio'] = '16:9'
else:
apply_video_resolution(params, node, records, config, output)
if rules['profile'] == 'vllm-omni-h3':
from opendirector.integrations.vllm_omni import FPS, h3_frames
params.update(fps=FPS, num_frames=h3_frames(params['duration']))
if rules['profile'] in {'vllm-omni-h3', 'fasth3-omni'} and 'duration' in params:
params.update(video_frame_parameters(params['duration'], params.get('fps', FPS)))
if spec.kind in {'image', 'video'} and config['protocol'] == 'volcengine':
params['watermark'] = not self.settings.get()['removeWatermark']
if spec.kind == 'image' and input_rules(config)['profile'] == 'moss-image':
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -9,7 +9,6 @@
from opendirector.domain.generation import GenerationError, GenerationFailure
from opendirector.ports.gateway import gateway
from opendirector.integrations.http_settings import MODEL_JOB_CONTEXT
from opendirector.integrations.fasth3 import finalize_video_edit
from opendirector.integrations.generation_errors import transport_failure

def _self_hosted_video(task):
Expand Down Expand Up @@ -152,7 +151,6 @@ async def result_ready(suffix, url):
suffix, data = await self.provider.generate(config, task, submitted, result_ready=result_ready)
else:
suffix, data = await self.provider.generate(config, task, submitted)
suffix, data = await asyncio.to_thread(finalize_video_edit, self.store, task, suffix, data)
elif cached:
suffix, data = cached
# Preserve complete bytes before catalog/link writes. Cancellation
Expand Down
2 changes: 1 addition & 1 deletion packages/core/src/opendirector/cli.py
Original file line number Diff line number Diff line change
Expand Up @@ -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. Native H3 accepts 4-15 seconds and aligns frames upward to 17k+5 at 24 fps (15 seconds becomes 362 frames).")],
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,
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -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
Expand All @@ -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.')

Expand Down Expand Up @@ -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。')
1 change: 1 addition & 0 deletions packages/core/src/opendirector/domain/creation/models.py
Original file line number Diff line number Diff line change
Expand Up @@ -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.')
Expand Down
17 changes: 15 additions & 2 deletions packages/core/src/opendirector/domain/rule_facts.py
Original file line number Diff line number Diff line change
Expand Up @@ -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)
Comment thread
FayeSpica marked this conversation as resolved.
Resolved
return seconds in durations


def resolution_choices(spec):
"""Sizes offered before submit. Callers report whether an image reference is connected."""
kind = spec['kind']
Expand All @@ -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]


Expand Down Expand Up @@ -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;
Expand Down Expand Up @@ -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;
Expand Down
10 changes: 8 additions & 2 deletions packages/core/src/opendirector/domain/video_capabilities.py
Original file line number Diff line number Diff line change
Expand Up @@ -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
Expand Down Expand Up @@ -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 rules['profile'] in {'vllm-omni-h3', 'fasth3-omni'}:
rules['fractionalDuration'] = True
Comment thread
FayeSpica marked this conversation as resolved.
Resolved
rules['videoCapabilities'] = capabilities
rules['outputResolutions'] = [item['resolution'] for item in capabilities['outputs']] if capabilities else []
return rules
Expand Down Expand Up @@ -153,9 +156,12 @@ 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)
if config.get('protocol') == 'vllm_omni' and not rules.get('fractionalDuration', False) and not float(seconds).is_integer():
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。')
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。')
Loading
Loading