Introduction
Leonardo Phoenix 2.0 is a comprehensive AI image generation platform with a lot of features — and a lot of questions from users trying to understand what it can and cannot do. This FAQ addresses the most common questions about Phoenix 2.0’s capabilities, organized by topic.
This is not a marketing document. Where Leonardo excels, we say so. Where it falls short or where the answer is “it depends,” we say that too.
Model Fine-Tuning (LoRA Training)
What is LoRA training and why does it matter?
LoRA (Low-Rank Adaptation) is a technique for customizing an AI model by training a small adapter that modifies the base model’s behavior. Instead of retraining the entire model (which requires massive compute resources), LoRA trains a lightweight layer that biases the model toward your specific style, subject, or concept.
In practical terms: you upload reference images of your art style, a specific character, or a product, and Leonardo creates a custom adapter that makes the model generate images that match your reference material.
How many images do I need to train a LoRA?
- Minimum: 10 images
- Recommended: 15–25 images
- Optimal for style LoRAs: 20–30 images with diverse examples of the target style
- Optimal for character LoRAs: 10–15 images showing the character from multiple angles
Quality matters more than quantity. 15 high-quality, consistent reference images will produce better results than 50 inconsistent ones.
How long does training take?
Training typically completes in 10–20 minutes depending on dataset size and server load. This is significantly faster than training a LoRA locally on consumer hardware (30–120 minutes) or using cloud GPU services.
How many LoRAs can I train per month?
| Plan | LoRA Trainings per Month |
|---|---|
| Free | 0 |
| Apprentice | 1 |
| Artisan | 5 |
| Maestro | 20 |
Can I combine multiple LoRAs?
Yes. Phoenix 2.0 supports combinable LoRAs with adjustable weighting. You can apply a style LoRA (e.g., your studio’s art direction) and a subject LoRA (e.g., a specific character) simultaneously, controlling the influence of each with weight sliders.
Example configuration:
- Studio style LoRA: weight 0.8
- Character LoRA: weight 0.6
- Result: your character rendered in your studio’s art style
Can I use my LoRA with the Consistent Character Engine at the same time?
Yes. The Consistent Character Engine and LoRA system are designed to work together. The Character Engine maintains character identity while the LoRA maintains visual style. This combination is one of Leonardo’s key differentiators.
Can I download my trained LoRA?
No. Trained LoRAs remain on Leonardo’s platform and can only be used through Leonardo’s generation interface or API. They cannot be exported for use with Stable Diffusion or other platforms.
What happens to my LoRA if I cancel my subscription?
Trained LoRAs are retained on Leonardo’s servers for a period after cancellation. If you resubscribe, they should still be available. However, Leonardo does not guarantee indefinite storage of LoRAs for inactive accounts. It is advisable to maintain your subscription if you rely on trained models.
Consistent Character Engine
How is the Character Engine different from LoRA training?
| Aspect | LoRA Training | Consistent Character Engine |
|---|---|---|
| Purpose | Style and subject customization | Character identity preservation |
| Training required | Yes (10–20 minutes) | No (real-time) |
| Reference images needed | 10–30 | 2–5 |
| What it preserves | Visual style, subject appearance | Facial identity, body proportions |
| Token cost | Standard generation cost | ~25% additional tokens |
| Setup time | 15–30 minutes | 5–10 minutes |
How consistent is the Character Engine, really?
Based on practical testing:
- Facial identity: 85–90% consistency (measured by facial recognition similarity)
- Clothing and accessories: 80–85% consistency
- Body proportions: 80–85% consistency
- Extreme angles: 70–80% consistency (full profile and extreme close-ups show more drift)
This is sufficient for most professional use cases (comics, game art, marketing) but not perfect. Expect to do occasional manual corrections for critical shots.
Can I use the Character Engine with AI-generated reference images?
Yes. You can generate a character using Leonardo, select the best results, and use those as reference images for the Character Engine. Many users follow this workflow:
- Generate initial character concepts with standard prompts
- Select and refine the best result
- Register the refined image as a Character Engine reference
- Generate consistent variations from that point forward
Is the Character Engine available on the Free plan?
No. The Consistent Character Engine requires a paid plan (Apprentice or higher).
API Access
Which plans include API access?
| Plan | API Access |
|---|---|
| Free | No |
| Apprentice | No |
| Artisan | Yes |
| Maestro | Yes |
What can I do with the API?
Leonardo’s REST API supports:
- Image generation: Generate images with all Phoenix 2.0 parameters, including LoRA and Character Engine
- LoRA training: Initiate and monitor training jobs programmatically
- Image editing: Inpainting, outpainting, and image-to-image transformation
- Motion generation: Create animated clips from static images
- Batch processing: Queue multiple generation jobs
- Webhook callbacks: Receive notifications when async operations complete
- Model management: List, select, and configure generation models
Is there a separate API pricing?
No. API usage draws from your plan’s monthly token allocation. There is no separate API subscription or per-call fee beyond your existing plan tokens. The same token costs apply whether you generate through the web interface or the API.
What are the rate limits?
Rate limits depend on your plan:
| Plan | Requests per Minute | Concurrent Jobs |
|---|---|---|
| Artisan | 30 | 5 |
| Maestro | 60 | 10 |
For most use cases, these limits are generous. High-volume production pipelines may need to implement queuing logic.
Is there an SDK?
Leonardo provides official SDK support for:
- Python: Full-featured SDK with async support
- JavaScript/TypeScript: Node.js SDK
- REST: Direct HTTP API for any language
Community SDKs exist for additional languages, though they are not officially supported.
Commercial Rights
Can I use Leonardo-generated images commercially?
Yes, on all paid plans. Leonardo’s terms of service grant commercial usage rights for images generated by paid subscribers. You can use generated images for:
- Client work and freelance projects
- Product packaging and advertising
- Published books, comics, and editorial content
- Social media and marketing campaigns
- Game assets and software products
- Merchandise and physical products
Does Leonardo offer copyright indemnity?
No. Unlike Adobe Firefly, Leonardo does not provide copyright indemnity — a legal guarantee that they will defend you against intellectual property claims related to generated content. This means:
- The risk of IP claims is borne by the user
- As of early 2026, no major successful IP claims have been made against AI-generated images
- The legal landscape is still evolving
- Risk-averse enterprises may prefer Adobe Firefly’s indemnity guarantee
Do I own the copyright to images I generate?
Copyright ownership of AI-generated images is an evolving legal area. In many jurisdictions (including the US, as of early 2026), purely AI-generated images may not be eligible for copyright protection. However:
- Images that involve significant human creative input (prompting, editing, composition decisions) may qualify for copyright in some jurisdictions
- Images refined through inpainting and manual editing have a stronger copyright claim
- Consult legal counsel for your specific jurisdiction and use case
Can I use generated images on the Free plan commercially?
No. Commercial usage rights are reserved for paid plans. Free plan generations are for personal, non-commercial use only.
Motion Generation
What is motion generation?
Motion generation transforms a static image into a short animated clip (3–5 seconds). It adds subtle movement — camera pans, atmospheric effects, character breathing, environmental motion — to bring still images to life.
Is this the same as video generation (like Runway or Kling)?
No. Leonardo’s motion generation is more limited:
- Duration: 3–5 seconds (vs. 10–60+ seconds for dedicated video tools)
- Complexity: Subtle motion and camera movement (vs. full narrative video)
- Control: Limited motion direction and intensity controls (vs. detailed video editing)
- Purpose: Bringing concept art to life for presentations and social media (vs. video production)
What does motion generation cost?
Motion generation costs approximately 40–60 tokens per clip, depending on resolution and motion complexity. This is 2–4x the cost of a standard image generation.
Is motion generation available on all plans?
| Plan | Motion Generation |
|---|---|
| Free | No |
| Apprentice | Limited (reduced quality/length) |
| Artisan | Full access |
| Maestro | Full access, priority processing |
Technical Specifications
What is the maximum resolution?
Phoenix 2.0 generates natively at up to 2048×2048 pixels. Higher resolutions can be achieved through upscaling tools, but native generation above 2048×2048 is not currently supported.
What output formats are supported?
- PNG: Lossless, transparent background support
- JPEG: Compressed, smaller file size
- WebP: Modern compressed format
What aspect ratios are supported?
Phoenix 2.0 supports a wide range of aspect ratios:
- 1:1 (square)
- 4:3 and 3:4 (standard)
- 16:9 and 9:16 (widescreen/vertical)
- 3:2 and 2:3 (photo standard)
- Custom ratios within supported resolution limits
How fast is generation?
Standard generation times:
| Resolution | Standard | With Character Engine |
|---|---|---|
| 512×512 | 2–4 seconds | 3–5 seconds |
| 1024×1024 | 4–8 seconds | 6–10 seconds |
| 2048×2048 | 8–15 seconds | 12–20 seconds |
Priority queue (Artisan and Maestro plans) reduces these times, especially during peak usage periods.
Does Leonardo support negative prompts?
Yes. You can specify negative prompts to exclude unwanted elements from generated images. Example: “blurry, low quality, extra fingers, distorted face” as a negative prompt helps avoid common generation artifacts.
Can I control the seed for reproducibility?
Yes. Leonardo allows you to specify a generation seed, which ensures that the same prompt with the same seed produces the same result. This is useful for reproducible workflows and systematic variation testing.
Account and Platform
Can multiple team members share an account?
Leonardo’s terms of service restrict account sharing. Each user should have their own account and subscription. For team use, each member needs their own plan — there is currently no team or organization plan with shared billing.
Is there an enterprise plan?
Leonardo offers custom enterprise arrangements for organizations with large-scale needs. Contact Leonardo’s sales team for enterprise pricing, which may include volume discounts, dedicated support, and custom token allocations.
What happens to my generations if I cancel?
Generated images are retained in your gallery for a period after cancellation. Leonardo does not guarantee indefinite storage for inactive accounts. Download any important images before canceling your subscription.
Does Leonardo have a mobile app?
As of early 2026, Leonardo is primarily a web-based platform. There is no dedicated mobile app, but the web interface is responsive and functional on mobile browsers for basic generation tasks. The full canvas editing experience is best on desktop.
Conclusion
Leonardo Phoenix 2.0 is a capable platform with genuine strengths in custom training and character consistency. Its limitations — no copyright indemnity, no offline capability, token-based pricing that requires management — are real and worth understanding before committing.
The best way to evaluate whether Leonardo fits your workflow is to start with the Free plan, test the generation quality and prompt handling, and upgrade to Apprentice if you want to explore the character consistency and LoRA features that define the platform.