strands-diffusers 0.1.1__tar.gz → 0.3.0__tar.gz

strands_diffusers-0.3.0/.github/workflows/docs.yml

@@ -0,0 +1,52 @@
+name: Docs
+
+on:
+  push:
+    branches: [main]
+    paths:
+      - 'docs/**'
+      - 'mkdocs.yml'
+      - '.github/workflows/docs.yml'
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  pages: write
+  id-token: write
+
+# Allow one concurrent deployment; don't cancel an in-progress run.
+concurrency:
+  group: pages
+  cancel-in-progress: false
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - name: Install docs deps
+        run: |
+          python -m pip install --upgrade pip
+          pip install "mkdocs==1.6.1" "mkdocs-material==9.7.6"
+      - name: Build (strict)
+        run: mkdocs build --strict
+      - name: Upload Pages artifact
+        uses: actions/upload-pages-artifact@v3
+        with:
+          path: site
+
+  deploy:
+    needs: build
+    runs-on: ubuntu-latest
+    environment:
+      name: github-pages
+      url: ${{ steps.deployment.outputs.page_url }}
+    steps:
+      - name: Deploy to GitHub Pages
+        id: deployment
+        uses: actions/deploy-pages@v4

strands_diffusers-0.3.0/.gitignore

@@ -0,0 +1,26 @@
+__pycache__/
+*.pyc
+*.egg-info/
+build/
+dist/
+.venv/
+.pytest_cache/
+.ruff_cache/
+.coverage
+*.bak
+*.json.tmp
+strands_diffusers/_version.py
+system_prompt.prompt
+site/
+
+# Generated media is ignored everywhere EXCEPT the committed docs gallery.
+*.mp4
+*.png
+*.jpg
+*.wav
+*.gif
+/assets/
+
+# Docs gallery assets are real, committed outputs — always track them.
+!docs/assets/
+!docs/assets/**

strands_diffusers-0.3.0/PKG-INFO

@@ -0,0 +1,238 @@
+Metadata-Version: 2.4
+Name: strands-diffusers
+Version: 0.3.0
+Summary: The universal entrypoint to HuggingFace diffusers for Strands agents — 100% pipeline & modality coverage, zero hardcoding. Special focus on Physical-AI world-foundation models (Cosmos) with robot action outputs.
+Author-email: Cagatay Cali <cagataycali@icloud.com>
+License: MIT
+Project-URL: Homepage, https://github.com/cagataycali/strands-diffusers
+Project-URL: Repository, https://github.com/cagataycali/strands-diffusers
+Project-URL: Issues, https://github.com/cagataycali/strands-diffusers/issues
+Keywords: strands,diffusers,huggingface,ai,agents,diffusion,video,image,vla,wfm,world-foundation-model,cosmos,robotics,physical-ai
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+Requires-Dist: strands-agents
+Requires-Dist: diffusers>=0.30
+Requires-Dist: transformers>=4.40
+Requires-Dist: torch
+Requires-Dist: pillow
+Requires-Dist: numpy
+Requires-Dist: accelerate
+Requires-Dist: matplotlib
+Provides-Extra: video
+Requires-Dist: imageio[ffmpeg]; extra == "video"
+Requires-Dist: opencv-python; extra == "video"
+Requires-Dist: av; extra == "video"
+Provides-Extra: audio
+Requires-Dist: soundfile; extra == "audio"
+Requires-Dist: librosa; extra == "audio"
+Provides-Extra: cosmos
+Requires-Dist: cosmos_guardrail; extra == "cosmos"
+Provides-Extra: dev
+Requires-Dist: pytest>=7.0; extra == "dev"
+Requires-Dist: black; extra == "dev"
+Requires-Dist: ruff; extra == "dev"
+Provides-Extra: all
+Requires-Dist: strands-diffusers[audio,dev,video]; extra == "all"
+
+# strands-diffusers
+
+<p align="center">
+  <img src="docs/assets/anim/banner.svg" alt="strands-diffusers — one tool, 300+ diffusion pipelines, every modality" width="100%"/>
+</p>
+
+
+**The universal entrypoint to HuggingFace `diffusers` for Strands agents.**
+One tool — `use_diffusers` — wraps the whole library with zero hardcoding:
+discover and run any of its 300+ pipelines across every modality. It's a *visual*
+library, so here's what it actually produces — every asset below is **real
+model output**, not a placeholder:
+
+<table>
+  <tr>
+    <td align="center" width="25%">
+      <b>text → image</b><br/>
+      <img src="docs/assets/text_to_image.png" width="200"/><br/>
+      <sub>any of 108 image pipelines</sub>
+    </td>
+    <td align="center" width="25%">
+      <b>text → video</b><br/>
+      <img src="docs/assets/text_to_video.gif" width="200"/><br/>
+      <sub>LTX · Wan · CogVideoX · Hunyuan</sub>
+    </td>
+    <td align="center" width="25%">
+      <b>robot actions</b> 🤖<br/>
+      <img src="docs/assets/cosmos_world.gif" width="200"/><br/>
+      <sub>Cosmos WFM: world video + actions</sub>
+    </td>
+    <td align="center" width="25%">
+      <b>text → audio</b><br/>
+      <img src="docs/assets/text_to_audio.png" width="200"/><br/>
+      <sub>StableAudio · AudioLDM2</sub>
+    </td>
+  </tr>
+</table>
+
+```
+text / image / video / robot-state  IN
+image / video / audio / actions / 3d  OUT
+```
+
+The registry is built at runtime from `diffusers._import_structure`, so new
+pipelines are supported automatically with no code change. Same philosophy as
+`use_aws`, `use_lerobot`, and `use_transformers`: **discover, don't hardcode.**
+
+<table>
+  <tr>
+    <td align="center" width="50%">
+      <b>3D mesh</b><br/>
+      <img src="docs/assets/mesh_render.png" width="200"/><br/>
+      <sub>ShapE - verts/faces to .ply</sub>
+    </td>
+    <td align="center" width="50%">
+      <b>audio</b> (<a href="docs/assets/text_to_audio.wav">hear the .wav</a>)<br/>
+      <img src="docs/assets/text_to_audio.png" width="300"/><br/>
+      <sub>StableAudio - waveform to .wav</sub>
+    </td>
+  </tr>
+</table>
+
+## 100% coverage, zero hardcoding
+
+<p align="center">
+  <img src="docs/assets/modality_coverage.png" width="640"/>
+</p>
+
+Every pipeline, model, and scheduler diffusers ships is reachable through one
+tool. When diffusers adds a new pipeline, `use_diffusers` exposes it immediately.
+
+## Physical-AI: world-foundation models with action outputs
+
+<p align="center">
+  <img src="docs/assets/cosmos_world.gif" width="360" alt="Cosmos world rollout"/>
+</p>
+
+<table>
+  <tr>
+    <td align="center"><img src="docs/assets/rollout_policy_1.gif" width="220"/><br/><sub>"Put the pot to the left of the purple item."</sub></td>
+    <td align="center"><img src="docs/assets/rollout_policy_2.gif" width="220"/><br/><sub>"Pick up the cloth and place it in the bowl."</sub></td>
+    <td align="center"><img src="docs/assets/rollout_policy_4.gif" width="220"/><br/><sub>"Open the drawer and place the spoon inside."</sub></td>
+  </tr>
+</table>
+
+Same robot, same first observation — **different task prompt → different imagined
+world and different predicted actions.** Five real rollouts + all three Cosmos
+action modes in the [WFM gallery](https://cagataycali.github.io/strands-diffusers/wfm/).
+
+
+This is the headline. A Cosmos action-policy rollout predicts both a future world
+**video** and the **robot action chunk** that produces it. One
+`use_diffusers(action="run", ...)` returns a `.mp4` world video, a `.json` action
+chunk (normalized `[-1, 1]`, shape `[num_chunks, T, action_dim]`), and optional
+`.wav` sound — and you can *see* the motion:
+
+<table>
+  <tr>
+    <td align="center"><b>time-series</b> (every dim, gripper highlighted)<br/><img src="docs/assets/cosmos_action_timeseries.png" width="380"/></td>
+    <td align="center"><b>end-effector path</b> (dims 0–2)<br/><img src="docs/assets/cosmos_action_trajectory.png" width="300"/></td>
+  </tr>
+</table>
+
+Verified end-to-end on NVIDIA Thor (`nvidia/Cosmos3-Nano`, bf16/cuda): one call
+produced a world video `(17, 480, 640, 3)` and an action chunk `(1, 16, 10)`. See
+[`examples/cosmos_action_policy.py`](examples/cosmos_action_policy.py).
+
+## Install
+
+```bash
+pip install -e .
+pip install -e ".[video,audio]"   # mp4 export, wav I/O
+```
+
+## Quick start
+
+```python
+from strands import Agent
+from strands_diffusers import use_diffusers
+
+agent = Agent(tools=[use_diffusers])
+agent("Generate an image of a robot arm in a kitchen")
+agent("Run a Cosmos action-policy rollout on robot.mp4 and give me the actions")
+```
+
+Direct:
+
+```python
+use_diffusers(action="run", pipeline="StableDiffusionPipeline",
+              model="stabilityai/stable-diffusion-2-1",
+              parameters={"prompt": "a robot arm in a kitchen"})
+# -> {"artifacts": ["/tmp/strands_diffusers/image_*.png"]}
+```
+
+## Two layers
+
+`run` loads a pipeline via `from_pretrained` and calls it; inputs are coerced
+(path / URL / base64 to PIL / video), outputs auto-saved and returned by path.
+
+`call` resolves and calls any diffusers class, function, or method (schedulers,
+VAEs, `CosmosActionCondition`, utils). `cached:key` references resolve to live
+objects; `"**"` unpacks a cached mapping into kwargs.
+
+```python
+use_diffusers(action="call", target="CosmosActionCondition",
+              parameters={"mode": "policy", "video": "robot.mp4"}, cache_key="cond")
+use_diffusers(action="run", pipeline="Cosmos3OmniPipeline", model="nvidia/Cosmos3-Nano",
+              parameters={"prompt": "...", "action": "cached:cond"},
+              dtype="bfloat16", device="cuda")
+```
+
+## Discovery
+
+| action | returns |
+|---|---|
+| `pipelines` / `models` / `schedulers` | classes + derived modality |
+| `tasks` / `modalities` / `wfm` | task maps / modality groups / world-foundation models |
+| `pipeline_info` / `inspect` | signature + docs |
+| `visualize` | action chunk to plots + animation |
+| `cache` / `clear_cache` | manage loaded pipelines |
+
+## Architecture
+
+```
+core/registry.py  zero-hardcode taxonomy from diffusers._import_structure
+core/engine.py    load/cache pipelines, auto device+dtype
+core/io.py        coerce inputs; serialize video/image/audio/action/mesh
+core/viz.py       render robot action chunks to plots + animation
+tools/use_diffusers.py  the single @tool: run + call + discovery
+```
+
+## Testing
+
+```bash
+pip install -e ".[video,audio,dev]"
+pytest tests/ -q          # unit tests, no GPU, no downloads
+python examples/smoke.py  # E2E gate on tiny fixtures
+```
+
+Every visual in this README and the [docs](https://cagataycali.github.io/strands-diffusers/)
+is produced by real `use_diffusers` calls — regenerate them with:
+
+```bash
+python examples/generate_docs_assets.py
+```
+
+## Docs
+
+📖 **[cagataycali.github.io/strands-diffusers](https://cagataycali.github.io/strands-diffusers/)**
+— quickstart, full gallery (images / video / audio / actions / 3D), the
+world-foundation-model story, discovery, and the two-layer design.
+
+MIT

strands_diffusers-0.3.0/README.md

@@ -0,0 +1,193 @@
+# strands-diffusers
+
+<p align="center">
+  <img src="docs/assets/anim/banner.svg" alt="strands-diffusers — one tool, 300+ diffusion pipelines, every modality" width="100%"/>
+</p>
+
+
+**The universal entrypoint to HuggingFace `diffusers` for Strands agents.**
+One tool — `use_diffusers` — wraps the whole library with zero hardcoding:
+discover and run any of its 300+ pipelines across every modality. It's a *visual*
+library, so here's what it actually produces — every asset below is **real
+model output**, not a placeholder:
+
+<table>
+  <tr>
+    <td align="center" width="25%">
+      <b>text → image</b><br/>
+      <img src="docs/assets/text_to_image.png" width="200"/><br/>
+      <sub>any of 108 image pipelines</sub>
+    </td>
+    <td align="center" width="25%">
+      <b>text → video</b><br/>
+      <img src="docs/assets/text_to_video.gif" width="200"/><br/>
+      <sub>LTX · Wan · CogVideoX · Hunyuan</sub>
+    </td>
+    <td align="center" width="25%">
+      <b>robot actions</b> 🤖<br/>
+      <img src="docs/assets/cosmos_world.gif" width="200"/><br/>
+      <sub>Cosmos WFM: world video + actions</sub>
+    </td>
+    <td align="center" width="25%">
+      <b>text → audio</b><br/>
+      <img src="docs/assets/text_to_audio.png" width="200"/><br/>
+      <sub>StableAudio · AudioLDM2</sub>
+    </td>
+  </tr>
+</table>
+
+```
+text / image / video / robot-state  IN
+image / video / audio / actions / 3d  OUT
+```
+
+The registry is built at runtime from `diffusers._import_structure`, so new
+pipelines are supported automatically with no code change. Same philosophy as
+`use_aws`, `use_lerobot`, and `use_transformers`: **discover, don't hardcode.**
+
+<table>
+  <tr>
+    <td align="center" width="50%">
+      <b>3D mesh</b><br/>
+      <img src="docs/assets/mesh_render.png" width="200"/><br/>
+      <sub>ShapE - verts/faces to .ply</sub>
+    </td>
+    <td align="center" width="50%">
+      <b>audio</b> (<a href="docs/assets/text_to_audio.wav">hear the .wav</a>)<br/>
+      <img src="docs/assets/text_to_audio.png" width="300"/><br/>
+      <sub>StableAudio - waveform to .wav</sub>
+    </td>
+  </tr>
+</table>
+
+## 100% coverage, zero hardcoding
+
+<p align="center">
+  <img src="docs/assets/modality_coverage.png" width="640"/>
+</p>
+
+Every pipeline, model, and scheduler diffusers ships is reachable through one
+tool. When diffusers adds a new pipeline, `use_diffusers` exposes it immediately.
+
+## Physical-AI: world-foundation models with action outputs
+
+<p align="center">
+  <img src="docs/assets/cosmos_world.gif" width="360" alt="Cosmos world rollout"/>
+</p>
+
+<table>
+  <tr>
+    <td align="center"><img src="docs/assets/rollout_policy_1.gif" width="220"/><br/><sub>"Put the pot to the left of the purple item."</sub></td>
+    <td align="center"><img src="docs/assets/rollout_policy_2.gif" width="220"/><br/><sub>"Pick up the cloth and place it in the bowl."</sub></td>
+    <td align="center"><img src="docs/assets/rollout_policy_4.gif" width="220"/><br/><sub>"Open the drawer and place the spoon inside."</sub></td>
+  </tr>
+</table>
+
+Same robot, same first observation — **different task prompt → different imagined
+world and different predicted actions.** Five real rollouts + all three Cosmos
+action modes in the [WFM gallery](https://cagataycali.github.io/strands-diffusers/wfm/).
+
+
+This is the headline. A Cosmos action-policy rollout predicts both a future world
+**video** and the **robot action chunk** that produces it. One
+`use_diffusers(action="run", ...)` returns a `.mp4` world video, a `.json` action
+chunk (normalized `[-1, 1]`, shape `[num_chunks, T, action_dim]`), and optional
+`.wav` sound — and you can *see* the motion:
+
+<table>
+  <tr>
+    <td align="center"><b>time-series</b> (every dim, gripper highlighted)<br/><img src="docs/assets/cosmos_action_timeseries.png" width="380"/></td>
+    <td align="center"><b>end-effector path</b> (dims 0–2)<br/><img src="docs/assets/cosmos_action_trajectory.png" width="300"/></td>
+  </tr>
+</table>
+
+Verified end-to-end on NVIDIA Thor (`nvidia/Cosmos3-Nano`, bf16/cuda): one call
+produced a world video `(17, 480, 640, 3)` and an action chunk `(1, 16, 10)`. See
+[`examples/cosmos_action_policy.py`](examples/cosmos_action_policy.py).
+
+## Install
+
+```bash
+pip install -e .
+pip install -e ".[video,audio]"   # mp4 export, wav I/O
+```
+
+## Quick start
+
+```python
+from strands import Agent
+from strands_diffusers import use_diffusers
+
+agent = Agent(tools=[use_diffusers])
+agent("Generate an image of a robot arm in a kitchen")
+agent("Run a Cosmos action-policy rollout on robot.mp4 and give me the actions")
+```
+
+Direct:
+
+```python
+use_diffusers(action="run", pipeline="StableDiffusionPipeline",
+              model="stabilityai/stable-diffusion-2-1",
+              parameters={"prompt": "a robot arm in a kitchen"})
+# -> {"artifacts": ["/tmp/strands_diffusers/image_*.png"]}
+```
+
+## Two layers
+
+`run` loads a pipeline via `from_pretrained` and calls it; inputs are coerced
+(path / URL / base64 to PIL / video), outputs auto-saved and returned by path.
+
+`call` resolves and calls any diffusers class, function, or method (schedulers,
+VAEs, `CosmosActionCondition`, utils). `cached:key` references resolve to live
+objects; `"**"` unpacks a cached mapping into kwargs.
+
+```python
+use_diffusers(action="call", target="CosmosActionCondition",
+              parameters={"mode": "policy", "video": "robot.mp4"}, cache_key="cond")
+use_diffusers(action="run", pipeline="Cosmos3OmniPipeline", model="nvidia/Cosmos3-Nano",
+              parameters={"prompt": "...", "action": "cached:cond"},
+              dtype="bfloat16", device="cuda")
+```
+
+## Discovery
+
+| action | returns |
+|---|---|
+| `pipelines` / `models` / `schedulers` | classes + derived modality |
+| `tasks` / `modalities` / `wfm` | task maps / modality groups / world-foundation models |
+| `pipeline_info` / `inspect` | signature + docs |
+| `visualize` | action chunk to plots + animation |
+| `cache` / `clear_cache` | manage loaded pipelines |
+
+## Architecture
+
+```
+core/registry.py  zero-hardcode taxonomy from diffusers._import_structure
+core/engine.py    load/cache pipelines, auto device+dtype
+core/io.py        coerce inputs; serialize video/image/audio/action/mesh
+core/viz.py       render robot action chunks to plots + animation
+tools/use_diffusers.py  the single @tool: run + call + discovery
+```
+
+## Testing
+
+```bash
+pip install -e ".[video,audio,dev]"
+pytest tests/ -q          # unit tests, no GPU, no downloads
+python examples/smoke.py  # E2E gate on tiny fixtures
+```
+
+Every visual in this README and the [docs](https://cagataycali.github.io/strands-diffusers/)
+is produced by real `use_diffusers` calls — regenerate them with:
+
+```bash
+python examples/generate_docs_assets.py
+```
+
+## Docs
+
+📖 **[cagataycali.github.io/strands-diffusers](https://cagataycali.github.io/strands-diffusers/)**
+— quickstart, full gallery (images / video / audio / actions / 3D), the
+world-foundation-model story, discovery, and the two-layer design.
+
+MIT

strands_diffusers-0.3.0/docs/assets/anim/banner.svg

@@ -0,0 +1,140 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 1280 360" width="1280" height="360" fill="none" role="img" aria-label="strands-diffusers: one tool, 300+ pipelines, every modality">
+  <defs>
+    <linearGradient id="g" x1="0" y1="0" x2="1" y2="1">
+      <stop offset="0" stop-color="#8B5CFF"/>
+      <stop offset="0.5" stop-color="#5C7CFF"/>
+      <stop offset="1" stop-color="#22D3EE"/>
+    </linearGradient>
+    <linearGradient id="gh" x1="0" y1="0" x2="1" y2="0">
+      <stop offset="0" stop-color="#8B5CFF"/>
+      <stop offset="0.5" stop-color="#5C7CFF"/>
+      <stop offset="1" stop-color="#22D3EE"/>
+    </linearGradient>
+  </defs>
+  <style>
+    .lbl{font:600 13px -apple-system,BlinkMacSystemFont,system-ui,sans-serif;fill:#86868b}
+    .mono{font:600 12px ui-monospace,SFMono-Regular,Menlo,monospace;fill:#86868b}
+    .cap{font:700 14px -apple-system,system-ui,sans-serif}
+    .btitle{font:800 34px -apple-system,system-ui,sans-serif;fill:url(#gh)}
+    .bcap{font:800 22px -apple-system,system-ui,sans-serif}
+    .bsub{font:600 13px -apple-system,system-ui,sans-serif;fill:#86868b}
+    .bmono{font:600 15px ui-monospace,Menlo,monospace;fill:#a1a1a6}
+    .blbl{font:600 16px -apple-system,system-ui,sans-serif;fill:#86868b}
+    .bpulse{transform-origin:640px 180px;animation:bp 3.4s ease-in-out infinite}
+    @keyframes bp{0%,100%{opacity:.9}50%{opacity:1;transform:scale(1.03)}}
+    .bnoise{animation:bn 6s ease-in-out infinite}
+    @keyframes bn{0%,15%{opacity:.9}55%{opacity:.25}100%{opacity:.9}}
+    .bcurve{stroke-dasharray:340;stroke-dashoffset:340;animation:bc 6s ease-in-out infinite}
+    @keyframes bc{0%,12%{stroke-dashoffset:340}55%,100%{stroke-dashoffset:0}}
+    @media (prefers-reduced-motion: reduce){
+      * { animation: none !important; }
+      .ron { opacity: 1 !important; }
+      .roff { opacity: .25 !important; }
+    }
+  </style>
+  <rect x="0" y="0" width="1280" height="360" fill="none"/>
+  <g class="bnoise" fill="#86868b">
+      <circle cx="245" cy="66" r="1.7" opacity="0.29"/>
+      <circle cx="276" cy="222" r="2.8" opacity="0.23"/>
+      <circle cx="175" cy="67" r="1.6" opacity="0.40"/>
+      <circle cx="48" cy="108" r="2.4" opacity="0.42"/>
+      <circle cx="111" cy="201" r="2.7" opacity="0.20"/>
+      <circle cx="298" cy="228" r="1.8" opacity="0.26"/>
+      <circle cx="346" cy="141" r="1.4" opacity="0.24"/>
+      <circle cx="311" cy="205" r="2.7" opacity="0.49"/>
+      <circle cx="212" cy="294" r="1.9" opacity="0.42"/>
+      <circle cx="305" cy="208" r="2.8" opacity="0.43"/>
+      <circle cx="265" cy="71" r="1.6" opacity="0.32"/>
+      <circle cx="66" cy="116" r="1.4" opacity="0.31"/>
+      <circle cx="243" cy="148" r="1.9" opacity="0.28"/>
+      <circle cx="125" cy="285" r="2.4" opacity="0.44"/>
+      <circle cx="95" cy="235" r="1.5" opacity="0.35"/>
+      <circle cx="357" cy="214" r="2.2" opacity="0.47"/>
+      <circle cx="310" cy="246" r="1.6" opacity="0.21"/>
+      <circle cx="141" cy="124" r="1.6" opacity="0.58"/>
+      <circle cx="320" cy="136" r="2.4" opacity="0.36"/>
+      <circle cx="333" cy="170" r="1.7" opacity="0.30"/>
+      <circle cx="220" cy="123" r="2.3" opacity="0.56"/>
+      <circle cx="168" cy="113" r="3.0" opacity="0.40"/>
+      <circle cx="69" cy="71" r="1.4" opacity="0.45"/>
+      <circle cx="293" cy="161" r="1.3" opacity="0.35"/>
+      <circle cx="359" cy="187" r="2.9" opacity="0.54"/>
+      <circle cx="44" cy="233" r="2.4" opacity="0.41"/>
+      <circle cx="125" cy="214" r="1.4" opacity="0.37"/>
+      <circle cx="185" cy="289" r="2.8" opacity="0.31"/>
+      <circle cx="200" cy="103" r="2.8" opacity="0.55"/>
+      <circle cx="136" cy="213" r="2.3" opacity="0.26"/>
+      <circle cx="284" cy="189" r="2.6" opacity="0.41"/>
+      <circle cx="40" cy="138" r="1.2" opacity="0.57"/>
+      <circle cx="321" cy="260" r="1.8" opacity="0.22"/>
+      <circle cx="321" cy="287" r="1.4" opacity="0.39"/>
+      <circle cx="62" cy="243" r="2.6" opacity="0.25"/>
+      <circle cx="192" cy="192" r="1.7" opacity="0.55"/>
+      <circle cx="175" cy="111" r="2.2" opacity="0.49"/>
+      <circle cx="104" cy="135" r="3.0" opacity="0.46"/>
+      <circle cx="180" cy="184" r="1.4" opacity="0.29"/>
+      <circle cx="148" cy="201" r="1.6" opacity="0.29"/>
+      <circle cx="63" cy="211" r="1.6" opacity="0.56"/>
+      <circle cx="315" cy="77" r="1.6" opacity="0.47"/>
+      <circle cx="109" cy="92" r="2.9" opacity="0.43"/>
+      <circle cx="191" cy="248" r="2.7" opacity="0.28"/>
+      <circle cx="71" cy="163" r="2.0" opacity="0.39"/>
+      <circle cx="273" cy="222" r="3.0" opacity="0.24"/>
+      <circle cx="169" cy="141" r="2.8" opacity="0.30"/>
+      <circle cx="101" cy="168" r="2.0" opacity="0.31"/>
+      <circle cx="120" cy="282" r="2.0" opacity="0.54"/>
+      <circle cx="216" cy="72" r="3.0" opacity="0.53"/>
+      <circle cx="350" cy="282" r="2.7" opacity="0.27"/>
+      <circle cx="195" cy="111" r="1.9" opacity="0.22"/>
+      <circle cx="161" cy="296" r="1.7" opacity="0.51"/>
+      <circle cx="186" cy="162" r="2.9" opacity="0.60"/>
+      <circle cx="218" cy="232" r="1.5" opacity="0.32"/>
+      <circle cx="350" cy="199" r="2.2" opacity="0.50"/>
+      <circle cx="58" cy="200" r="2.1" opacity="0.54"/>
+      <circle cx="90" cy="291" r="1.3" opacity="0.27"/>
+      <circle cx="230" cy="222" r="1.6" opacity="0.25"/>
+      <circle cx="325" cy="119" r="2.3" opacity="0.45"/>
+      <circle cx="174" cy="200" r="2.1" opacity="0.57"/>
+      <circle cx="105" cy="232" r="1.6" opacity="0.36"/>
+      <circle cx="255" cy="132" r="1.8" opacity="0.50"/>
+      <circle cx="63" cy="170" r="3.0" opacity="0.60"/>
+      <circle cx="63" cy="111" r="1.7" opacity="0.57"/>
+      <circle cx="322" cy="271" r="1.9" opacity="0.26"/>
+      <circle cx="307" cy="229" r="2.3" opacity="0.59"/>
+      <circle cx="249" cy="62" r="2.7" opacity="0.32"/>
+      <circle cx="252" cy="285" r="1.4" opacity="0.25"/>
+      <circle cx="74" cy="193" r="1.7" opacity="0.44"/>
+      <circle cx="270" cy="109" r="2.3" opacity="0.31"/>
+      <circle cx="196" cy="277" r="2.7" opacity="0.24"/>
+      <circle cx="176" cy="126" r="1.2" opacity="0.51"/>
+      <circle cx="244" cy="123" r="2.5" opacity="0.42"/>
+      <circle cx="177" cy="62" r="1.3" opacity="0.55"/>
+      <circle cx="329" cy="191" r="2.7" opacity="0.43"/>
+      <circle cx="87" cy="91" r="1.8" opacity="0.56"/>
+      <circle cx="295" cy="267" r="2.8" opacity="0.28"/>
+      <circle cx="120" cy="85" r="2.6" opacity="0.55"/>
+      <circle cx="170" cy="209" r="1.5" opacity="0.57"/>
+      <circle cx="317" cy="294" r="2.7" opacity="0.55"/>
+      <circle cx="48" cy="237" r="1.8" opacity="0.57"/>
+      <circle cx="297" cy="267" r="2.7" opacity="0.31"/>
+      <circle cx="292" cy="86" r="2.8" opacity="0.54"/>
+      <circle cx="111" cy="256" r="2.0" opacity="0.32"/>
+      <circle cx="295" cy="115" r="1.2" opacity="0.28"/>
+      <circle cx="145" cy="267" r="2.9" opacity="0.31"/>
+      <circle cx="245" cy="156" r="3.0" opacity="0.41"/>
+      <circle cx="341" cy="88" r="2.9" opacity="0.27"/>
+      <circle cx="348" cy="124" r="1.4" opacity="0.37"/>
+  </g>
+  <path class="bcurve" d="M60 280 C 160 120, 300 180, 360 150"
+        stroke="url(#gh)" stroke-width="5" fill="none" stroke-linecap="round"/>
+  <path d="M380 120 C 470 120, 500 180, 560 180" stroke="#3a3a3e" stroke-width="1.5" fill="none" opacity="0.45"/><circle r="3" fill="url(#gh)" opacity="1.0"><animateMotion dur="2.6s" begin="0.7285450606527043s" repeatCount="indefinite" path="M380 120 C 470 120, 500 180, 560 180" calcMode="spline" keyTimes="0;1" keySplines="0.4 0 0.2 1"/></circle><text class="bmono" x="372" y="124" text-anchor="end">text</text><path d="M380 165 C 470 165, 500 180, 560 180" stroke="#3a3a3e" stroke-width="1.5" fill="none" opacity="0.45"/><circle r="3" fill="url(#gh)" opacity="1.0"><animateMotion dur="2.6s" begin="0.31367731419499123s" repeatCount="indefinite" path="M380 165 C 470 165, 500 180, 560 180" calcMode="spline" keyTimes="0;1" keySplines="0.4 0 0.2 1"/></circle><text class="bmono" x="372" y="169" text-anchor="end">image</text><path d="M380 210 C 470 210, 500 180, 560 180" stroke="#3a3a3e" stroke-width="1.5" fill="none" opacity="0.45"/><circle r="3" fill="url(#gh)" opacity="1.0"><animateMotion dur="2.6s" begin="0.6062088533061433s" repeatCount="indefinite" path="M380 210 C 470 210, 500 180, 560 180" calcMode="spline" keyTimes="0;1" keySplines="0.4 0 0.2 1"/></circle><text class="bmono" x="372" y="214" text-anchor="end">video</text><path d="M380 255 C 470 255, 500 180, 560 180" stroke="#3a3a3e" stroke-width="1.5" fill="none" opacity="0.45"/><circle r="3" fill="url(#gh)" opacity="1.0"><animateMotion dur="2.6s" begin="0.5114230596694781s" repeatCount="indefinite" path="M380 255 C 470 255, 500 180, 560 180" calcMode="spline" keyTimes="0;1" keySplines="0.4 0 0.2 1"/></circle><text class="bmono" x="372" y="259" text-anchor="end">robot state</text>
+  <path d="M720 180 C 790 180, 800 80, 900 80" stroke="#3a3a3e" stroke-width="1.5" fill="none" opacity="0.45"/><circle r="3" fill="url(#gh)" opacity="1.0"><animateMotion dur="2.6s" begin="1.0s" repeatCount="indefinite" path="M720 180 C 790 180, 800 80, 900 80" calcMode="spline" keyTimes="0;1" keySplines="0.4 0 0.2 1"/></circle><text class="bmono" x="912" y="84" text-anchor="start">image</text><path d="M720 180 C 790 180, 800 128, 900 128" stroke="#3a3a3e" stroke-width="1.5" fill="none" opacity="0.45"/><circle r="3" fill="url(#gh)" opacity="1.0"><animateMotion dur="2.6s" begin="1.25s" repeatCount="indefinite" path="M720 180 C 790 180, 800 128, 900 128" calcMode="spline" keyTimes="0;1" keySplines="0.4 0 0.2 1"/></circle><text class="bmono" x="912" y="132" text-anchor="start">video</text><path d="M720 180 C 790 180, 800 176, 900 176" stroke="#3a3a3e" stroke-width="1.5" fill="none" opacity="0.45"/><circle r="3" fill="url(#gh)" opacity="1.0"><animateMotion dur="2.6s" begin="1.5s" repeatCount="indefinite" path="M720 180 C 790 180, 800 176, 900 176" calcMode="spline" keyTimes="0;1" keySplines="0.4 0 0.2 1"/></circle><text class="bmono" x="912" y="180" text-anchor="start">audio</text><path d="M720 180 C 790 180, 800 224, 900 224" stroke="#3a3a3e" stroke-width="1.5" fill="none" opacity="0.45"/><circle r="3" fill="url(#gh)" opacity="1.0"><animateMotion dur="2.6s" begin="1.75s" repeatCount="indefinite" path="M720 180 C 790 180, 800 224, 900 224" calcMode="spline" keyTimes="0;1" keySplines="0.4 0 0.2 1"/></circle><text class="bmono" x="912" y="228" text-anchor="start">actions</text><path d="M720 180 C 790 180, 800 272, 900 272" stroke="#3a3a3e" stroke-width="1.5" fill="none" opacity="0.45"/><circle r="3" fill="url(#gh)" opacity="1.0"><animateMotion dur="2.6s" begin="2.0s" repeatCount="indefinite" path="M720 180 C 790 180, 800 272, 900 272" calcMode="spline" keyTimes="0;1" keySplines="0.4 0 0.2 1"/></circle><text class="bmono" x="912" y="276" text-anchor="start">3D</text>
+  <g class="bpulse">
+    <rect x="560" y="148" width="160" height="64" rx="20"
+          fill="rgba(124,92,255,0.12)" stroke="url(#gh)" stroke-width="2.2"/>
+  </g>
+  <text class="bcap" x="640" y="178" text-anchor="middle" fill="url(#gh)">use_diffusers</text>
+  <text class="bsub" x="640" y="200" text-anchor="middle">one @tool</text>
+  <text class="btitle" x="640" y="56" text-anchor="middle">strands-diffusers</text>
+  <text class="blbl" x="640" y="330" text-anchor="middle">300+ diffusion pipelines · every modality · even robot world models</text>
+</svg>