Image Generation
Generate and edit images in Studio with reference images and up to 4K output, plus one-click upscaling and background removal.
The Studio image view turns prompts into images and edits reference images you provide. Choose a model, set your aspect ratio and quality, optionally attach reference images, and generate a batch.
You can also generate images inline in the Chat app with the /image command (see Images & 3D Models). Studio gives you a full choice of models, reference-image editing, multi-image composition, and higher resolutions.
Choosing a model#
Studio offers several image models, each tuned for different strengths — some prioritize speed for quick drafts, others maximize detail, text legibility, or multi-image composition. The model picker describes what each is best at so you can match it to the task.
| Model | Best for | Reference images | Quality |
|---|---|---|---|
| Nano Banana | Fast, reliable single-image edits and everyday generations | 1 | 1K |
| Nano Banana Pro | Multi-image composition and accurate in-image text | Up to 6 | 1K–4K |
| Nano Banana 2 | Google's fast model — sharp results with multi-image edits | Up to 6 | 1K–4K |
| GPT Image 2 | Prompt accuracy and legible text | Multiple | 1K–4K |
| Flux Schnell | Fastest drafts, great for quick iterations | 1 | 1K–4K |
| Flux 2 | High detail with up to 4 reference images | Up to 4 | 1K–4K |
| Grok Imagine | Quick, playful creative shots | Up to 3 | 1K–2K |
| Ideogram 4.0 | Posters, logos, and crisp typography | — (text-to-image only) | 2K |
The lineup evolves over time, so the picker in the app is always the source of truth — new models appear there before they appear here.
Compare models#
Open the model menu and turn on Compare models to run one prompt on up to 4 models at once. Each model generates into the same batch so you can judge the results side by side and keep the one that fits.
Settings#
- Prompt — Up to 2,000 characters. Be specific about subject, style, lighting, composition, and color.
- Aspect ratio — Square (1:1), landscape (16:9, 4:3, 3:2, 21:9), or portrait (9:16, 3:4, 2:3).
- Quality — 1K, 2K, or 4K (the available range can depend on the model you pick).
- Reference images — Attach one or more images to edit a starting image, transfer style, or compose multiple inputs. How many you can attach depends on the model.
Some models expose a few extra options of their own (for example a quality level or a fast-vs-quality toggle); these appear in the panel when you select that model.
Prompting tips#
- Be specific about the subject, style, and mood
- Mention lighting — "golden hour", "studio lighting", "dramatic shadows"
- Specify a style — "photorealistic", "illustration", "watercolor", "pixel art"
- Include composition details — "close-up", "wide angle", "aerial view"
- Name your colors — "warm earth tones", "cool blues", "vibrant neon"
Reference-image editing#
Attach one or more images to guide a generation. Depending on the model you can:
- Edit a single image (change a color, remove an object, restyle it)
- Compose several images into one scene
- Transfer style from a reference onto a new subject
Upscale and remove background#
Hover any finished image to run one-click tools on it:
- Upscale — Enlarge the image 2× or 4× with an AI upscaler that adds detail as it scales, great for turning a draft into a print- or hero-ready image.
- Remove background — Cut the subject out and get a transparent PNG, ready to composite into a design or scene.
Both tools produce a new image in your session — the original is untouched — and the result is saved to your Asset Library like any other generation.
Using your images#
Finished images are saved to your Asset Library and can be:
- Downloaded as PNG, WebP, or JPEG
- Sent to 3D as the base for an image-to-3D generation
- Used as a video start frame for image-to-video
- Referenced in Chat or dropped into a Canvas design
Some models stream a progressive preview as they render, so you can watch the image resolve in real time; others return the finished result in one step.