ZeroTwo can generate and edit pictures, handy for UI assets, banners, illustrations, sprite sheets and placeholders made alongside your other work. It runs on the web, the desktop app and the mobile apps, and it is not tied to one model: the catalog includes GPT Image, Gemini and Nano Banana, FLUX, Qwen Image, Seedream, Grok Imagine, Ideogram, Recraft and others.
In a chat, choose Create image from the + menu, or just describe the picture in the composer. Attach a reference image if you want it transformed or used as a guide.
Ask in the task prompt and the agent calls its image tools for you. Reference images can come from attachments or from your Library.
Open Images from the overflow menu in the rail for a dedicated page with a composer, a gallery of styles to try on your own image, and a grid of what you have made.
Review and edit generated images
Click a generated image in a chat to open it in the Canvas tab. A toggle at the top switches between Single image and All images; the grid can be scoped to the whole thread or just the current run. A floating bar offers Comment, Remove and Resize, and Multi-select in the grid lets you gather several images so one message can give feedback on all of them. Say what to change and what to keep.
Generate or edit an image
Describe the result in plain language, and attach a reference when you are extending existing art. Each request spends credits from your plan, and the cost follows what the model's provider charges. Generation stops when your credits run out. Upscaling, images on Pages and auto-mask use credits too. To choose which model draws and which edits, open Settings > Personalization > Models and set the image generation model and the editing model.
Write effective image prompts
One to three sentences usually suffice if they cover purpose, subject, setting and style, and any constraints or exclusions. Concrete visuals beat vague praise, and fixed requirements are worth repeating. You can also state an orientation, such as square for an avatar, portrait for a poster or landscape for a banner.
A minimal editorial illustration of a person planning a project at a desk, soft window light, muted colors, no text.
Refine the result
Start from the core idea and change one thing per round so the rest stays stable. You can also open the image, add a Comment where the fix belongs, and state what must be preserved.
Replace only the mug with a potted plant; keep everything else identical.
Use multiple reference images
Attach more than one image and label them by order, saying what each contributes, such as content from one and style from another. Words like foreground or left help placement. How many references work well depends on the model you picked.
Image 1 is the product; apply the line work and palette of image 2, keeping the product centered.
Add text to an image
Keep text short, put it in quotes with the casing, style and position you want, and say whether other text is allowed. Text rendering varies between models, so proofread the result.
Include only the title "SPRING WORKSHOP", in bold white sans-serif, centered at the top.
Create infographics and dense layouts
Models can draft posters, diagrams and timelines. Give a clear hierarchy and short labels, then proofread every word and finish critical typography in a design tool.
Additional considerations
- Get permission before depicting a real person.
- Request original designs, not imitations of a brand or artist.
- Follow your organization's rules.