Master the WaveSpeed Desktop Playground: Batch Processing, LoRA, and Mask Editor

WaveSpeed Desktop’s Playground is where you interact with 1,000+ AI models. But it’s more than a simple prompt box — it includes batch processing, LoRA support, an interactive mask editor, multi-tab workflows, and template management.

This guide covers the advanced features that make the Playground a production-ready tool, not just a demo interface.

---

Multi-Tab Playground

Run multiple models simultaneously in separate tabs.

How It Works

• Open a new tab for each model you want to use
• Each tab maintains its own parameters, prompt, and configuration
• Tabs run independently — start a FLUX generation in one tab while editing parameters in another
• Switch between tabs without losing state

Use Cases

• Compare models — Run the same prompt through FLUX, Qwen Image, and Seedream in parallel and compare outputs
• Pipeline workflow — Generate an image in Tab 1, then use it as input for an editing model in Tab 2
• Different tasks — Image generation in one tab, video creation in another, audio processing in a third

---

Batch Processing

Run the same prediction 2 to 16 times with automatically randomized seeds. Every other parameter stays the same — only the seed changes between runs.

Why Batch Processing Matters

AI image generation is non-deterministic. The same prompt with different seeds produces different outputs. Batch processing lets you:

• Generate variations — Get 8 different interpretations of your prompt in one click
• Find the best output — Pick the best result from multiple generations instead of rerolling one at a time
• Save time — Queue all runs at once instead of clicking “Generate” repeatedly

How to Use It

1. Configure your model parameters and prompt
2. Set the batch count (2–16)
3. Click Run
4. All results appear in your output panel and are saved to My Assets

Each batch run uses a unique random seed while keeping your prompt, model settings, and other parameters identical.

---

LoRA Support

Apply LoRA (Low-Rank Adaptation) weights to supported models for fine-tuned outputs.

What LoRA Does

LoRA weights modify a base model’s behavior without retraining the entire model. Common uses:

• Style LoRAs — Apply specific artistic styles (anime, photorealistic, oil painting)
• Subject LoRAs — Generate consistent characters or objects
• Concept LoRAs — Add learned concepts (specific poses, compositions, lighting setups)

How to Use LoRA in WaveSpeed Desktop

1. Select a LoRA-compatible model (FLUX Dev LoRA, Wan 2.2 LoRA, etc.)
2. Enter your LoRA URL or select from available LoRAs
3. Adjust the LoRA strength (weight)
4. Choose between high-noise and low-noise variants when available
5. Generate as normal

The Playground dynamically renders LoRA parameters based on the model’s schema — if a model supports LoRA, the options appear automatically.

---

Mask Editor

An interactive canvas-based tool for drawing masks directly on images. Essential for inpainting, outpainting, and targeted editing tasks.

Tools

Tool	Function
Brush	Paint mask areas (adjustable size)
Eraser	Remove mask from painted areas
Fill	Fill entire regions or the full canvas

How to Use It

1. Select a model that accepts mask input (inpainting models, image erasers, etc.)
2. Upload your source image
3. Click the mask editor icon to open the canvas
4. Paint over the areas you want the AI to modify
5. Close the editor — your mask is attached to the prediction
6. Run the model

The mask editor works on both desktop (mouse) and mobile (touch) — with touch-optimized controls for finger painting on Android.

Tips

• Use a larger brush for rough area selection, then eraser to clean edges
• Fill is useful for selecting everything except the subject (invert selection)
• Zoom in for precise mask edges on detailed areas
• The mask is saved with your prediction for easy iteration

---

Template Management

Save and reuse your favorite configurations as templates.

What Gets Saved

• Model selection
• All parameter values (prompt, negative prompt, resolution, steps, etc.)
• LoRA configuration
• Batch settings

How to Use Templates

1. Configure your ideal settings for a model
2. Click Save as Template
3. Name your template (e.g., “FLUX Portraits 1024x1024”)
4. Next time, load the template to instantly apply all settings

Templates are stored locally and persist across app updates.

---

Dynamic Form Generation

The Playground automatically generates the correct input form for every model based on its schema. This means:

• No memorizing parameters — Every option has a label, description, and valid range
• Correct input types — Sliders for numbers, dropdowns for options, file pickers for uploads
• Validation — The form prevents invalid inputs before you submit
• Upload components — Image, video, and audio upload fields with preview

When WaveSpeed adds a new model, the Playground immediately supports it — no app update required.

---

History and Asset Management

History

View your last 24 hours of predictions with full details:

• Input parameters used
• Output files
• Cost per prediction
• Timestamps

My Assets

All generated outputs are automatically saved to your local assets folder:

• Auto-save — Every output is saved without manual intervention
• Tags — Organize assets with custom tags
• Search — Find previous outputs by model name, date, or tag
• Quick actions — Re-use any output as input for another model

---

Cost Controls

The Playground includes built-in spending management:

• Real-time cost estimate — See the estimated cost before clicking Run
• Account balance — Check your WaveSpeed balance directly in Settings
• Per-execution tracking — Know exactly what each generation costs

---

Platform Availability

The full Playground is available on:

Platform	Playground	Mask Editor	Batch Processing	LoRA
Windows	Yes	Yes	Yes	Yes
macOS	Yes	Yes	Yes	Yes
Linux	Yes	Yes	Yes	Yes
Android	Yes	Yes (touch)	Yes	Yes

---

Download

Download WaveSpeed Desktop

Available for Windows, macOS, Linux, and Android. Free to download — cloud model usage is pay-per-use through your WaveSpeed account.

---

FAQ

How many models support LoRA? All FLUX variants, Wan 2.2 variants, Qwen Image LoRA models, and many others. If a model supports LoRA, the option appears automatically in the form.

Can I run batch processing on mobile? Yes — batch processing (2–16 runs) works on Android.

Is there a limit to saved templates? No hard limit. Templates are lightweight JSON stored locally.

Do I need an API key for the Playground? Yes — the Playground runs cloud models that require a WaveSpeed account. The 12 free tools do not require an API key.

How do I get an API key? Sign up at wavespeed.ai and generate a key at wavespeed.ai/accesskey.