Video · text-to-image · image-to-image · text-to-video · video-to-video · reference-to-video · reference-video-to-video

Bernini

ByteDance

Official inference source is vendored in-tree and loaded from the self-contained Diffusers checkpoint.

Integrated独立环境源码已固定bernini

构建运行命令

选择仓库中已记录的 variant，然后复制对应的准备、检查或运行命令。

worldfoundry.pipeline

Variant

任务text-to-image

环境worldfoundry-bernini-cu124

设备CUDA 12.4

worldfoundry-eval evaluate \
  --mode model \
  --model-id bernini \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/bernini \
  --metric artifact_count \
  --json

兼容性与版本

由 Manifest 生成

集成状态: integrated
Runner 证据: pending gpu validation
环境: worldfoundry-bernini-cu124
Python: 3.11
CUDA: CUDA 12.4
PyTorch: torch==2.5.1+cu124
源码 revision: 82d573e9a553dccb1405dde4cb0f26466f9df6c5
Checkpoint revision: b988d6af97714041e0d21a31d4faff5ac42bac09

Runtime profile: bernini
Pipeline binding: bernini
Runner: worldfoundry.pipeline
Pipeline target: worldfoundry.pipelines.bernini.pipeline_bernini:BerniniWorldFoundryPipeline
Backend stage: official_in_tree_runtime
Runtime 状态: checkpoint_assets_partial_not_runnable_until_staged
Driver 状态: compatible
环境类型: 独立环境

安装环境

环境解析器会读取已记录 profile，并选择此处显示的统一或独立环境。

bash scripts/setup/model_env_install.sh --model bernini

依赖版本约束 19

torch==2.5.1+cu124
torchvision==0.20.1+cu124
diffusers==0.35.2
accelerate==0.34.2
transformers==4.57.3
huggingface-hub
flash-attn==2.8.3
safetensors
einops
numpy
Pillow
tqdm
ftfy
decord
imageio
imageio-ffmpeg
scipy
loguru
veomni @ git+https://github.com/ByteDance-Seed/VeOmni.git@v0.1.10

Conda 依赖 5

python=3.11
pip
ffmpeg
ninja
cuda-nvcc

Checkpoint 与资产

分配算力前先做本地检查；gated、private 与 license 字段直接来自 checkpoint manifest。

worldfoundry-eval zoo model-download --model-id bernini --check-local --json

ByteDance/Bernini-Diffusers

Revision: —
License: apache-2.0
Gated: —
Private: —

ByteDance/Bernini-Diffusers

Revision: b988d6af97714041e0d21a31d4faff5ac42bac09
License: —
Gated: —
Private: —

运行与输出

生成的命令通过共享 evaluation 边界运行，并持久保存结果 manifest 与 artifact。

worldfoundry-eval evaluate \
  --mode model \
  --model-id bernini \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/bernini \
  --metric artifact_count \
  --json

输入契约

字段	已记录契约
`prompt`	Required
`image`	Required
`images`	Required
`video`	Required
`task_type`	t2i, i2i, t2v, v2v, r2v, rv2v

Artifact 契约

Artifact 类型	文件名 / 路径
`generated_image`	`—`
`generated_video`	`—`
`generated_video`	`bernini.mp4`

证据与来源

Catalog 集成、原生 demo parity 与 runner parity 是三条独立记录。

集成状态: integrated
Runner 证据: pending gpu validation
原生 Demo 证据: 未记录
验证 Imports: torch, torchvision, diffusers, transformers, huggingface_hub, accelerate, flash_attn, veomni, decord, imageio, loguru, worldfoundry.pipelines.bernini, worldfoundry.synthesis.visual_generation.bernini

配方溯源

worldfoundry/data/models/catalog/video/bernini.yamlProjecthttps://bernini-ai.github.io/Paperhttps://arxiv.org/abs/2605.22344GitHub82d573e9a553dccb1405dde4cb0f26466f9df6c5Hugging Facehttps://huggingface.co/ByteDance/Bernini-Diffusers

构建运行命令

输入契约

Artifact 契约

相关配方