AI Image Generator

Every picture in this AI image generator has a price in credits, and the exact number appears on the Generate button before you start. Most images cost between 1 and 6 credits, depending on the model and the resolution or quality you pick. If the model cannot produce an image, the credits return to your balance automatically, so a failed attempt costs nothing.

Z-Image is the free-plan model. It costs a fixed 3 credits per picture, and a free account can use it with no daily limit, paid from welcome credits and the daily check-in. The check-in adds 15 credits a day, enough for five Z-Image pictures. Guests can try the AI image generator without an account: pressing Generate starts a free guest session, and a new guest's welcome credits pay for a Z-Image after a short queue.

Every other model in the AI image generator needs purchased credits. Buying a pack removes the queue and unlocks all 20 models; the packs are listed on the pricing page. Paid pictures are often cheaper than you might expect: Nano Banana, Nano Banana 2.1 at 1K, Seedream 5.0 Flash, Grok Imagine Image 2.0 and Imagen 4 Fast each cost 1 credit per image.

The AI image generator works in two modes. With a prompt alone, the model creates a new picture from scratch. Add one or more reference images and most models switch to editing: they keep a face, a product or a layout from your upload and change what you describe, such as the background, the outfit, the season or the art style. Models that accept references carry an Edit label on their card, together with the number of images they take.

The limits differ a lot between models. The GPT Image models accept up to 16 reference images, Seedream 5.0 Lite and Seedream 4.5 take up to 14, the Nano Banana 2 family takes up to 10, and Wan 2.7 Image Pro takes up to 9. Flux Kontext and Ideogram V3 work from a single image. Z-Image and Imagen 4 are text only, so they take no uploads and suit pure creation from a prompt. If your source is a single photo, any model with an Edit label will do; the high limits matter only when you mix several sources in one picture.

References can be JPG, PNG or WebP files up to 10 MB, and you can upload them, drag them in or paste them. Several references let the AI image generator combine sources: put a product into a new scene, dress a character in a garment from another photo, or merge a logo into a mockup. To cut a subject out before you use it as a reference, run it through the free background remover first.

Resolution in the AI image generator comes as named steps rather than pixel sizes. The Nano Banana 2 family and the GPT Image 2 family offer 1K, 2K and 4K. Seedream 5.0 Lite renders 2K, 3K or 4K for the same 2 credits, and Seedream 4.5 renders 2K or 4K for 2 credits as well. Flux 2 Pro, Qwen Image 3.0 Pro and Wan 2.7 Image Pro stop at 2K. Other models trade a quality or speed setting instead: GPT Image 1.5 has Medium and High, Ideogram V3 has Turbo, Balanced and Quality, and Imagen 4 has Fast, Standard and Ultra.

Aspect ratio decides where the picture can go. Use 1:1 for profile images and feeds, 4:5 or 3:4 for portrait posts, 9:16 for stories and phone wallpapers, and 16:9 or 21:9 for banners and slides. Nano Banana 2.1 and Nano Banana 2 add extreme strips of 1:4, 4:1, 1:8 and 8:1, and GPT Image 2 adds 2:1 and 1:2. Several models also offer Auto, which follows the shape of your reference image. Pick the ratio before you generate rather than cropping later, because cutting a finished square down to a wide banner throws away most of the picture and the detail you paid for.

A sensible workflow is to draft at 1K or on a fast model and keep the higher setting for the final version. If a picture comes out right but too small for print, enlarge it with the free image upscaler instead of paying the AI image generator for a full new render.

Nano Banana Pro from Google is the default model of the AI image generator and carries the Pro badge. It costs 5 credits at 1K or 2K and 6 credits at 4K, takes up to 8 reference images, and accepts prompts up to 20,000 characters. Nano Banana 2.1, marked New, costs 1 to 3 credits from 1K to 4K, and Nano Banana 2 costs 2 to 5. The original Nano Banana, marked Fast, costs 1 credit per image with up to 10 references and no resolution choice, which makes it a quick model for drafts and edits.

Imagen 4, also from Google, is a text-only model in three versions: Fast for 1 credit, Standard for 2 and Ultra for 3. Choose it when you want a clean new picture from a prompt and have nothing to upload. Imagen 4 accepts prompts up to 5,000 characters in five ratios: 1:1, 16:9, 9:16, 3:4 and 4:3. Fast is handy for testing a composition in the AI image generator, and Ultra is the step to take once the prompt is settled.

OpenAI's GPT Image line is the best option in this AI image generator when an edit needs many inputs, because every model in it accepts up to 16 reference images. GPT Image 2.5 Sunburst and GPT Image 2.5 Flare, both marked New, and GPT Image 2 each cost 2, 3 or 4 credits at 1K, 2K or 4K. GPT Image 1.5 costs 1 credit at Medium quality and 6 at High, in square, 2:3 or 3:2 formats.

ByteDance's Seedream family covers every budget. Seedream 5.0 Pro, marked Hot, costs 2 credits at 1K and 4 at 2K, and the price rises slightly when you add more than one reference image. Seedream 5.0 Lite costs 2 credits at any of its 2K, 3K or 4K sizes, Seedream 5.0 Flash, marked Fast, costs 1 credit, and Seedream 4.5 costs 2 credits at 2K or 4K. Lite and 4.5 accept up to 14 references, which helps when a whole set of product photos must stay consistent.

Black Forest Labs offers Flux 2 Pro for 2 credits at 1K or 2K with up to 8 references, and Flux Kontext in Pro (2 credits) and Max (3 credits) for single-image edits. Kontext works in English, so prompts in other languages are translated to English before they reach it. Qwen Image 3.0 Pro from Alibaba costs 2 or 3 credits with up to 3 references, and Wan 2.7 Image Pro costs 3 credits with up to 9.

Ideogram V3 trades speed for detail across three rendering speeds: Turbo for 1 credit, Balanced for 2 and Quality for 3, from text or one reference image. Grok Imagine Image 2.0 from xAI costs 1 credit and takes up to 5 references. Z-Image from Alibaba, the free-plan model, rounds out the list at 3 credits with a 1,000-character prompt limit.

A strong prompt for an AI image generator describes a picture, not a topic. Start with the subject, then the setting, then the style, the light and the camera. Compare a coffee shop with a small corner coffee shop at dawn, steam rising from a ceramic cup on a wooden counter, soft window light, shot on 35mm film, shallow depth of field. The second prompt gives the model decisions it would otherwise make at random. Name colors, materials and textures too, because words like brushed steel, linen or matte paper give the AI image generator something concrete to render.

For edits, describe the change and say what must stay the same: keep the face and the pose, replace the background with a sunny beach, match the lighting. When you combine several references, say which image supplies what, for example the product from the first image and the room from the second. Keep prompts within the model's limit: Z-Image accepts 1,000 characters, Seedream 4.5 and Seedream 5.0 Lite about 3,000, and the Nano Banana and GPT Image models 20,000.

Test on a cheap model, change one thing at a time, and switch to a premium model once the idea works. Every result stays in your history, and the Use as reference button sends a finished picture straight back into the reference slot, so you can refine it step by step in the AI image generator instead of starting over from text. When the picture is final, you can bring it to life in the AI video generator as the start frame of a clip, so the same AI image generator session can feed a whole campaign of stills and short videos.