Preserve an image’s structure while changing its style, subject, or finish with ControlNet-driven generation.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install controlled-generation
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install controlled-generation using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Controlled Generation (ControlNet)?
Controlled generation (ControlNet) is an Openclaw Skills workflow for generating a new image that matches the source composition while changing the look.
It extracts a structural guide such as Canny edges, pose, depth, or segmentation from the source image, then conditions image synthesis on that guide. The result is a same composition, new look pipeline for restyling, sketch-to-render, interior redesign, and repeatable asset variation.
Use it when layout fidelity matters more than a freeform prompt. If you need subject identity consistency rather than composition consistency, use character-consistency instead; for composition-preserving variations, Controlled Generation is the right Openclaw Skills tool.
Controlled Generation (ControlNet) Use Cases
- Restyle a sketch, render, or photo while keeping the original silhouette or layout
- Turn floor plans, interiors, or room references into new design renders
- Generate pose-driven people scenes from a source photo or reference pose
- Create repeatable product, object, or game-asset variations with the same composition
- Keep scene regions stable while changing materials, colors, lighting, or style
- Reuse one guide image across a batch for consistent composition across multiple prompts
How Controlled Generation (ControlNet) Works
- Collect the source image, decide what structure to preserve, define the desired change, and choose how tightly the composition should be held.
- Inspect the live schemas for the preprocessing and inference calls so every field name, model ID, and range is schema-valid.
- Run the matching ControlNet preprocessor on the source image to produce a guide image: Canny for outlines, OpenPose for people, depth for perspective, or segmentation for region-aware restyles.
- Generate the final image by calling image inference with a compatible base model and the matching ControlNet, passing the guide image plus weight, startStep, endStep, and control mode.
- Review the result, then tune thresholds, weight, or the step window to trade off structure versus creativity.
- For batches, reuse the same guide image and vary only the prompt so every output keeps one composition.
Controlled Generation (ControlNet) Setup
Controlled Generation is a workflow skill, not a standalone package. To use it, make sure your Openclaw Skills runtime can access runware-run and runware-models, then confirm the live AIRs before sending requests. The skill also references references/examples.md for schema-verified recipes.
# Example request configuration for the two-call pipeline
SOURCE_IMAGE="<url-or-base64>"
PREPROCESSOR_TYPE="canny"
BASE_MODEL="runware:101@1"
CONTROLNET_MODEL="runware:25@1"
CONTROL_MODE="balanced"
- Verify that the preprocessor and matching ControlNet are live before hardcoding anything.
- Prefer FLUX.1 [dev] for the quality tier and SDXL for a faster, cheaper tier.
- For Canny workflows, prepare a clean, high-contrast source image with clear outlines and low noise.
- Load
references/examples.mdto start from known-good control map recipes.
Controlled Generation (ControlNet) Data Schema & Taxonomy
Controlled Generation uses transient request and response artifacts rather than persistent local files by default. The core data model is a two-call pipeline: preprocess the source image into a guide image, then feed that guide into image inference.
| Artifact | Role | Key metadata |
|---|---|---|
| Source image | Structural input used to derive the guide | inputImage, URL or base64, required |
| Control map / guide image | Structure-preserving guide produced by preprocessing | preProcessorType, Canny thresholds, output becomes guideImage |
| Inference request | Final synthesis payload | model, prompt, controlNet[], controlMode |
| ControlNet entry | Binds the guide to the base model | matching ControlNet model, guideImage, weight, startStep, endStep |
| Output image | Returned result | synchronous image URL(s), one or many variants |
Metadata taxonomy
- Structure type:
canny,depth,openpose,seg,softedge - Base family: FLUX.1 [dev] or SDXL
- Control strength:
weight - Guidance window:
startStep/endStep - Control priority:
balanced,controlnet,prompt - Variant grouping: reuse one
guideImageacross multiple prompts
Reference file
references/examples.mdcontains three schema-verified recipes: Canny restyle, OpenPose generation, and depth-guided interior restyle.
Controlled Generation (ControlNet) Advanced Features
- Live schema discovery with runware-run, so model IDs and field ranges stay current instead of hardcoded
- Multiple control types: Canny, depth, OpenPose, segmentation, and soft edges
- Paired base plus ControlNet selection lets you switch between FLUX quality and SDXL speed or cost
- Adjustable control window with
startStepandendStepfor precise structure-versus-creativity tuning - Reusable guide images for consistent multi-image sets and batch variations
- Canny threshold tuning for tighter or looser edge locking
- Prompt-first or control-first balancing via
balanced,controlnet, andpromptmodes - Optional LoRA layering to lock style while the guide preserves composition
SKILL.md
Loading
Related Openclaw Skills
2D Game Assets
Openclaw Skills for 2D game assets generates cohesive sprite sets, icons, stickers, and transparent cutouts that all share one visual style.
Image to 3D Asset
Openclaw Skills converts photos or prompts into production-ready GLB 3D assets with controllable topology, materials, and polygon budgets.
LLM Agent
Build OpenAI-compatible LLM agents on Runware that can reason, call your functions, and answer with grounded tool results.
Logos and Vectors
Openclaw Skills creates real SVG logos or converts raster images into editable vector art that stays crisp at any size.
Composite Scene
Composite Scene merges multiple real images into one coherent AI-generated scene by reconciling placement, scale, lighting, and perspective.
Cinematic Video
Generate short, shot-directed videos that feel filmed rather than synthesized, using precise cinematic prompts and live model-aware workflows.
Featured*
