Image models & modes
A detailed reference for choosing the right model, engine, or mode for your image workflows in Spaces.
This article covers the configuration details behind the image nodes. For an overview of what each node does and how to use it, see Image Nodes.
In this article
- Image Generator models
- How to choose a model
- Model feature tags
- Model providers
- Image Upscaler: Creative mode
- Image Upscaler: Precision mode
- Creative vs Precision comparison
- Variation modes
Image Generator models
The Image Generator supports dozens of AI models from leading providers. Each model has different strengths, capabilities, and credit costs. You can switch models any time from the model selector on the node card, or search for a specific model in Spotlight (for example, search "Flux Ultra" to add an Image Generator pre-configured with that model).
How to choose a model
| You need | Look for |
|---|---|
| Photorealism | Models labeled Photo or Realistic. Trained on photographic data. |
| Illustration and graphic art | Flux and Ideogram models. Strong for stylized visuals. |
| Fast iteration | Models with shorter generation times (shown in Spotlight tooltips). |
| Lower cost | Check the credits tag in the model selector. Costs range from 0 to 20+ credits. |
| Reference image support | Not all models support references. Check the feature tags before connecting a Reference port. |
| High resolution | Some models support up to 4K output, others are limited to 1K or 2K. |
| Text in images | Ideogram models. The best choice when your image needs readable text. |
Model feature tags
When browsing models in the selector or in Spotlight tooltips, you will see feature tags that indicate what each model supports.
| Tag | What it means |
|---|---|
| Reference Images | Can use uploaded images to guide generation. The tag shows the max number of references accepted. |
| Negative Prompt | Supports describing what to avoid in the output. |
| Smart Prompt | AI-enhanced prompt optimization that rewrites your prompt for better results. |
| Resolution | Available output resolutions (1K, 2K, 4K). |
| Style Transfer | Can transfer visual style from reference images. |
| Color Palette | Supports color scheme guidance. |
| Inpainting | Can edit specific regions of an image. |
| Credits | Cost per generation at default settings. |
Hover over any model in the selector to see a detailed tooltip with its features, credit cost, and average generation time.
Model providers
| Provider | Known for |
|---|---|
| Flux | High quality, versatile image generation across many styles |
| Google (Imagen) | Fast generation with good prompt adherence |
| OpenAI (DALL-E) | Strong prompt understanding and creative interpretation |
| Stability AI | Stable Diffusion family. Wide style range and community ecosystem |
| Ideogram | Excellent text rendering in images. Best choice when your image needs readable text |
| Freepik (Mystic) | Custom models optimized for Freepik's creative workflows |
Image Upscaler: Creative mode (Magnific)
Creative mode uses Magnific to upscale images with AI-generated detail enhancement. It can reimagine textures, add fine details, and produce results that go beyond simple interpolation. Max scale: 4x. Max output: 4096px.
Models
| Model | Best for |
|---|---|
| Magnific | Larger outputs with rich visual detail. |
| Classic | Quick processing with style presets. |
Engines
| Engine | Best for |
|---|---|
| Illusio | Illustrations, cartoon art, stylized content. Adds artistic flair. |
| Sharpy | Detailed, sharp images. Maximum clarity on edges and fine details. |
| Sparkle | Photographs. Natural enhancement that respects the original character. |
| Automatic | General use. The system analyzes the image and selects the best engine. |
Presets
| Preset | Behavior |
|---|---|
| Subtle | Minimal AI interpretation. Stays very close to the original. |
| Vivid | Balanced. Noticeable improvement while respecting the original content. |
| Wild | Maximum AI creativity. May add significant new details and textures. |
| Custom | Set each slider manually for full control. |
Sliders
All sliders range from -10 to +10.
| Slider | What it controls |
|---|---|
| Creativity | How much the AI can reimagine details. Negative = conservative, positive = imaginative. |
| Resemblance | How closely the output matches the original. Negative = more freedom, positive = more faithful. |
| HDR | Dynamic range enhancement. Higher values produce more dramatic tonal range. |
| Fractality | Controls the level of fractal-like detail added during upscaling. |
Optimized For
Select the content category that best describes your image to guide the AI's enhancement strategy. 13 categories available: Standard Ultra, Portrait Soft, Portrait Hard, Landscape, Photography, Sci-Fi, 3D Renders, Video Games, Films, Illustrations, Art, Horror, Anime.
Image Upscaler: Precision mode
Precision mode is powered by Clarity. It upscales images faithfully, enhancing resolution and sharpness without adding creative interpretation. Best for photos, products, and any image where fidelity matters. Max scale: 16x. Max output: 16384px.
Models
| Model | Best for |
|---|---|
| Ultra Sublime | Product photography, archival fidelity. Preserves maximum detail. |
| Ultra Photo | Portraits, editorial. Natural skin and texture refinement. |
| Ultra Denoiser | Low-light, compressed, or noisy images. Removes noise while upscaling. |
| Ultra | General purpose upscaling. Fastest option. |
Presets
| Preset | Description |
|---|---|
| Balanced | General-purpose upscaling suitable for most images. |
| Portraits | Optimized for faces and skin. Preserves natural texture. |
| Grainy Analog | Preserves film grain aesthetic for analog-style photography. |
Sliders
All sliders range from 0 to 100.
| Slider | What it controls |
|---|---|
| Sharpness | Edge sharpness enhancement. |
| Grain | Film grain control. Increase to add grain, decrease to reduce it. |
| Ultra Detail | Fine detail enhancement level. |
Creative vs Precision comparison
| Feature | Creative (Magnific) | Precision (Clarity) |
|---|---|---|
| Max scale | 4x | 16x |
| Max output | 4096px | 16384px at 16x |
| AI creativity | High. Reimagines details. | Low. Faithful reproduction. |
| Best for | Art, illustrations, creative content | Photos, products, print |
| Prompt support | Yes | No |
| Speed | Moderate | Fast |
| Content optimization | 13 categories | Preset-based |
| Slider range | -10 to +10 | 0 to 100 |
Variation modes
Each variation mode takes a single source image and produces a grid of alternatives along a specific creative axis.
Angles
Generates the same subject from different camera angles. The AI maintains the subject's identity, lighting, and setting while changing the perspective. You choose the grid size and select specific camera angles (front, side, three-quarter, overhead, low angle, etc.).
Best for product photography, character turnarounds, and architectural visualization.
Demographics
Varies the demographic representation of people in the image while maintaining the same pose, outfit, lighting, and setting. You can filter by ethnicity and gender representation.
Best for inclusive marketing materials, diverse advertising campaigns, and stock photo creation.
Expressions
Creates different facial expressions for the same person: happy, sad, surprised, angry, contemplative, and more.
Best for character design, game development (NPC expression libraries), animation reference, and sticker creation. A well-lit, front-facing portrait produces the most consistent results.
Age
Shows the same person at different life stages, from childhood through old age. The AI preserves recognizable features while aging or de-aging the subject.
Best for creative storytelling, aging simulations, and character development across a timeline. A 4x1 or 1x4 grid reads naturally as a progression.
Storyboard
Creates a sequence of scenes that tell a visual story. The source image sets the visual style, and your prompt drives the narrative arc. Write the prompt as a sequence of events and the AI maps them to panels in order.
Best for pre-production, pitch decks, visual storytelling, and video planning. More panels means more granular narrative beats. Start with 3x3 for a solid arc.
Custom
Free-form variations based on any text prompt you provide. This is the most open-ended mode. The AI interprets your instructions and applies them to the source image.
Best for creative exploration, A/B testing visual concepts, batch style transfer, and "what if" experiments. Be specific for predictable results, or be vague for serendipity. Custom mode can replicate what other modes do, but specialized modes are more consistent at their specific task.
Common settings
All variation modes share these settings: Aspect Ratio and Resolution control the output dimensions. Modes that accept specific selections (like Angles letting you choose camera positions, or Demographics letting you select ethnicities and genders) show those options in the node inspector when selected.
Storyboard and Custom modes require a text prompt. Other modes work with the image alone.
Can't find an answer to your question?
Our support team is here to help you with any questions or issues.
Submit a request