LIBRARY IMAGE
Image details
Pinch or scroll to zoom. Drag whenever the image exceeds the view.
NO WORKER CONNECTED
Start the generator on this computer
Once its local worker is running, this page will connect to it automatically. You can also sign in to use a linked worker elsewhere.
YOUR WORK
Library
Loading your scaffold groups…
Your library is empty
Generated images will appear here automatically, grouped by the scaffold they share.
CREATE
New generation
A separate local GPU worker creates images while the browser stays responsive.
Run this job on
Worker
Auto chooses an online worker with this model and the shortest queue.
Model
Choose a local checkpoint
Prompting guide for this model
LoRA adapters
Trained details
Reference images
0 / 4
Guide the result
Use an image from your library
More steps can add detail, but take longer.
Batch generation
Creates each image one after another with this prompt and model.
1 image · generated sequentially
Learning your hardware — complete a generation to see a personal estimate.
1024 × 1024
Face pass steps
Uses a 1024px face canvas, then softly blends it back into the image.
Face redraw strength
Higher values rebuild more of the face; lower values preserve the original more closely.
Used only for the face crop. The main prompt still controls the composition, body, outfit, and scene.
EXPERIMENTAL
Creates the whole image at a model-friendly size, then softly blends low-strength detail tiles for a sharper large final image.
Tile detail steps
Tile redraw strength
Higher values add sharper new detail, but can alter the scaffold more. Start around 0.20–0.25.
Tile overlap
More overlap produces softer joins, but can require more tiles and take longer.
Reads the scaffold once and gives each tile a local prompt for the objects actually visible there. Uses Microsoft Florence-2 Base, downloaded on first use.
Enable high-res finish to calculate its tile estimate.
Advanced settings
Latest image
Images are saved automatically.
Your canvas is ready
Write a prompt, choose your settings, and generate.
Creating your images
Preparing the model…
SAVED COMPOSITION
Finish from scaffold
Composition, shape, model, prompt, sampler, and CFG stay fixed so only safe finishing choices can change.
LOCKED SOURCE
The saved scaffold and finishing passes will appear here as they are assembled.
Run this job on
The saved scaffold stays on its source worker.
Finish method
Choose a standard pass or a larger, blended tile finish.
Final resolution
Changes the size of the final conservative refinement pass.
Base image steps
Controls the full-image finish generated from the saved scaffold before optional tile and face passes.
Full-image pass
LoRA adapters
Apply compatible trained details during the scaffold finishing pass.
Tile detail steps
Tile redraw strength
Higher values add more detail and sharpness, but may change more of the saved composition.
Tile overlap
More overlap gives softer joins, but takes longer.
Builds local tile prompts from the saved composition instead of repeating the complete image prompt everywhere.
Face pass steps
![Saved composition for manual face marking]()
Face redraw strength
Used only for the face crop; the saved scaffold preserves the rest of the image.
Use this when a profile, tilted, or obscured face is missed. Drag a box tightly around the face.
Variations
Each result keeps the same composition but gets fresh finishing detail.
1 variation · generated sequentially