Generate photorealistic images or refine existing visuals with precise prompt-based edits.
Generate from a prompt or edit one existing image with prompt-based instructions.
Describe the image to generate or the exact edit to apply. Maximum 5000 characters.
Each generated variant is billed separately.
Sign in to spend tokens
Prompt is required
Your generated images will appear here
MAI-Image-2.5 generates photorealistic images from text and edits existing visuals with precise prompt-based instructions — from 1 token per image on Prizmad. Pick a mode, write a prompt or upload an image to refine, generate.
MAI-Image-2.5 is Microsoft AI's image model, built around two jobs: generating photorealistic images from a text prompt, and refining existing images through prompt-based edits. The realism focus makes it a natural fit for e-commerce work — product shots with believable lighting, crisp detail, and clean composition rather than an obviously synthetic look.
The edit mode is what sets the workflow apart for store owners. Upload one image — a supplier packshot, a phone photo of your product, a creative that's almost right — and describe the change in plain language: "turn this packshot into a clean ecommerce hero image with soft studio lighting, subtle shadow, and a white background." Prompts can run up to 5,000 characters, so you can be as specific as the edit demands. Output covers eight aspect ratio options (including Auto, which follows the source or prompt), three file formats (JPEG, PNG, WebP), and 1–4 variants per run, with each generated variant billed separately.
On Prizmad, MAI-Image-2.5 uses the same token balance as , video, , captions, and music. No Microsoft account or API key is involved — you pay in the same tokens as every other tool, and every image you generate or edit is yours to use commercially.
Turn rough supplier photos into clean e-commerce hero images with studio lighting, subtle shadow, and a white background — a core dropshipping workflow.
Realistic campaign visuals — serum on polished stone, earbuds on reflective black — that read as photography, not AI art.
Consistent, crisp listing images across your catalog by running the same edit prompt on every SKU's source photo.
Take a proven ad image and prompt targeted changes — new background, new lighting, seasonal restage — instead of rebuilding from scratch.
Re-run the same concept at 9:16, 1:1, and 16:9 to cover Stories, feed, and banner placements with matching visuals.
Text-to-image output tuned for realism — believable lighting, crisp product detail, and clean campaign-ready composition.
Upload one JPG, PNG, or WebP image and describe the exact change — relight it, restage it, or turn a raw packshot into a polished hero shot.
Up to 5,000 characters of prompt, enough to specify lighting, surface, palette, shadow behavior, and composition in one pass.
Auto, 1:1, 4:3, 3:4, 16:9, 9:16, 3:2, and 2:3 — Auto keeps the natural format while the fixed ratios target specific ad placements.
Export as PNG, JPEG, or WebP depending on where the image is headed — ad platform, product page, or email.
Generate 1–4 variations of the same prompt or edit in one job; each generated variant is billed separately.
Pick a mode: Generate from prompt for a new image, or Edit existing image to refine a visual you already have.
In edit mode, upload one image — JPG, PNG, or WebP. Supplier packshots, phone photos, and near-final creatives all work as starting points.
Write the prompt. For generation, describe subject, surface, lighting, and composition. For edits, state the exact change you want — the field accepts up to 5,000 characters, so be specific.
Set the aspect ratio. Auto follows the source or prompt; fixed options like 9:16, 1:1, and 16:9 target specific ad placements.
Choose the output format (PNG, JPEG, or WebP) and the number of images (1–4). Each generated variant is billed separately.
Click Generate. Results arrive in roughly 1–3 minutes and are saved to your asset library, ready for ads, product pages, or Prizmad video projects.
MAI-Image-2.5 generations start at 1 token per image, in both generate and edit modes. When you request multiple variants in one run, each generated variant is billed separately.
MAI-Image-2.5 is unlocked on the same token balance you use for AI avatars, video, voiceover, and music — no Microsoft account, no API key, no separate billing.
If your tokens run out mid-campaign, buy a one-off top-up directly from the top-up page. Every image you generate or edit stays yours with full commercial rights.
It runs on your Prizmad token balance — no separate Microsoft account or API key is needed. Generations start at 1 token per image.
From 1 token per image, in both text-to-image and edit modes. If you generate multiple variants in one run (up to 4), each variant is billed separately.
Yes — that's its second mode. Upload one JPG, PNG, or WebP image and describe the edit in the prompt: change the background, relight the scene, clean up a packshot into a hero image. The model applies the instruction while keeping the product recognizable.
Up to 5,000 characters. That's enough to specify lighting, surface, palette, shadow, and composition in detail — longer, more specific prompts generally produce more controllable edits.
Yes. You keep full commercial rights to every image generated or edited with MAI-Image-2.5 on Prizmad — paid ads, product listings, email, and print included.
Aspect ratios: Auto, 1:1, 4:3, 3:4, 16:9, 9:16, 3:2, and 2:3. Output formats: PNG (default), JPEG, and WebP.
Edit mode takes exactly one source image per run, but you can request up to 4 output variants of that edit in a single job — each generated variant is billed separately.
MAI-Image-2.5 is the pick for photorealism and for editing existing images — packshot cleanups, relights, and restages. KREA 2 is the pick for art direction: style references, creativity levels, and expressive campaign looks. Both start at 1 token and live on the same token balance, so run the same brief through both and keep the winner.
Published 2026-07-16 · Last updated 2026-07-16