Chrome ExtensionMAI-Image-2.6 Prompt Generator
Generate optimized prompts for Microsoft AI's MAI-Image-2.6 — released August 10, 2026 and #2 on the Arena text-to-image leaderboard at launch, with +91 Elo on text rendering over 2.5, multi-image references, web grounding and output up to 1.5K. Built for photorealism, accurate in-image text, portraits, product shots and surgical edits, with Generate and Edit modes and anti-slop rules baked in.
Be specific: describe who/what is in the image with details
Quick Palettes
Generated Prompt
Fill in the form and click "Generate" to create an optimized MAI-Image-2.6 prompt.
Tip: Be as specific as possible with your subject description. Instead of "a dog," try "a golden retriever puppy with a red bandana."
MAI-Image-2.6 Tips
- • Structure: Scene → Subject → Key Details → Composition/Camera → Lighting → Style → Constraints
- • Anti-slop: swap "stunning/masterpiece/8K" for concrete facts ("overcast daylight", "50mm feel", "brushed aluminum")
- • In-image text is a MAI-Image-2.6 strength (+91 Elo over 2.5) — quote the literal text and specify font, color, size, and placement
- • For edits, say "Change only X, keep everything else identical" and list what to preserve — MAI-Image edits surgically
- • MAI-Image-2.6 accepts several reference images — say which image supplies which element ("subject from image 1, style from image 2")
- • For people, ask it to preserve facial identity across pose/expression/viewpoint changes
- • No negative-prompt field — fold exclusions into plain words ("no watermark, no duplicate text")
- • 2.6 supports dynamic aspect ratios up to 1.5K; pick Flash for fast, cheap drafts and the full model for final fidelity
MAI-Image-2.6 Prompt Templates
Copy-ready image briefs built around what this model actually rewards. Swap the [BRACKETED] parts for your own subject and settings, then paste straight into the tool.
Surgical Object Replacement
EditingMAI-Image-2.6's signature — localised edits that leave the rest untouched.
In the supplied image, replace [OBJECT] with [NEW OBJECT]. Match the new object's scale, perspective, lighting direction and shadow to the existing scene exactly. Leave every other pixel of the image — background, other objects, colour grade, grain — completely unchanged.
In-Image Text Update
EditingChanging text on a sign or label without re-rendering the image.
In the supplied image, change the text on [WHERE, e.g. the storefront sign] from its current wording to read exactly "[NEW TEXT]". Match the existing font style, weight, colour, perspective and any surface texture or wear. Leave the rest of the image untouched.
Identity-Preserving Repose
PeopleFacial identity preservation across pose and viewpoint changes.
Using the supplied portrait as the identity reference, generate the same person [NEW POSE / VIEWPOINT, e.g. turned three-quarters to camera right, looking slightly upward] with [NEW EXPRESSION]. Preserve their exact facial structure, skin tone, hair and distinguishing features. New setting: [SETTING]. Lighting: [LIGHTING].
Artifact Cleanup
EditingRemoving blur, noise or unwanted elements without touching the subject.
In the supplied image, remove [WHAT TO REMOVE, e.g. the motion blur on the subject / the power lines across the sky / the sensor dust spots] and reconstruct the area cleanly to match its surroundings. Preserve the subject's detail, the original lighting and the existing colour grade exactly. Do not sharpen or alter anything outside the corrected region.
Photoreal Product Scene
ProductText-to-image generation with MAI's photorealism.
A [PRODUCT] on [SURFACE] in [SETTING]. [MATERIAL AND FINISH DETAIL]. [CAMERA ANGLE] at [HEIGHT]. Lit by [LIGHTING SETUP] with [SHADOW BEHAVIOUR]. Photorealistic rendering, accurate material response, true-to-life colour, sharp focus on [FOCAL POINT]. Output 1024x1024.
Background Replacement
EditingSwapping the environment while keeping the subject pixel-accurate.
Replace the background of the supplied image with [NEW BACKGROUND]. Keep the subject exactly as it is — same pose, same edges, same detail. Relight only the subject's rim and edges so they sit believably in the new environment, matching [LIGHT DIRECTION AND COLOUR TEMPERATURE]. Clean cut-out with no halo or fringing.
MAI-Image Update Log
MAI-Image-2.6 & 2.6-Flash
● LatestAugust 10, 2026- Announced August 10, 2026; public preview in Microsoft Foundry from September 4, 2026, and available to try in the MAI Playground.
- +79 Elo overall and +91 Elo on text rendering versus MAI-Image-2.5; launched at #2 on the Arena text-to-image leaderboard.
- Generation and editing with multiple reference images, web grounding (uses current web content and documents to inform visuals), dynamic aspect ratios and resolutions up to 1.5K.
- Better portraits, 3D imagery and commercial work such as product, branding and cinematic shots.
- MAI-Image-2.6-Flash targets production volume at less than half the price of the full model. Foundry preview pricing: 2.6 at $5 / $8 / $38 and Flash at $1.75 / $2.50 / $19 per 1M text-input / image-input / image-output tokens.
MAI-Image-2.5 & 2.5-Flash
June 2026- Diffusion model with a flow-matching objective for text-to-image and image-to-image editing (model version 2026-06-02).
- Ranked #2 for image editing and #3 for text-to-image on Arena at launch; PNG output up to 1024×1024.
- Known for surgical localized edits, facial-identity preservation and in-image text; wired into PowerPoint, OneDrive and Bing Image Creator / Copilot.
How to Use the Prompt Generator
Pick Mode & Describe Your Vision
Choose Generate or Edit and a model variant (full or Flash). Pick a use case (product photo, portrait, slide graphic, logo, retouch, etc.), then describe your scene and subject, set style, composition and lighting, and add any text that must appear in the image. For edits, fill in the “Preserve” field so face, layout, and brand elements stay locked.
Generate Your Prompt
Click Generate. The tool orders your inputs the way MAI-Image prefers, swaps vague praise for concrete visual facts, quotes in-image text with font and placement directives, and adds a “Change only X, keep everything else identical” instruction plus a preserve list when you're editing.
Copy & Create
Paste into Microsoft Foundry or the MAI Playground (MAI-Image-2.6), or any MAI-Image-2.5 surface such as OpenRouter or Copilot / Bing Image Creator. For iterative edits, reuse the prompt as a base and make one small change at a time, repeating the preserve list each round to prevent drift.
Frequently Asked Questions
What is MAI-Image-2.6?
MAI-Image-2.6 is Microsoft AI's latest in-house image model, announced on August 10, 2026 alongside a cheaper MAI-Image-2.6-Flash. It handles text-to-image generation and natural-language editing, and launched at #2 on the Arena text-to-image leaderboard. It entered public preview in Microsoft Foundry on September 4, 2026 and can be tried in the MAI Playground.
What's new in MAI-Image-2.6?
Compared with MAI-Image-2.5, Microsoft reports +79 Elo overall and +91 Elo on text rendering. It adds generation and editing with multiple reference images, web grounding (it can use current web content and documents to inform an image), dynamic aspect ratios and resolutions up to 1.5K (2.5 topped out at 1024×1024). Microsoft also highlights better portraits, 3D imagery, and product, branding and cinematic shots. The Flash variant is described as production-ready quality at less than half the price.
What was MAI-Image-2.5?
MAI-Image-2.5 is Microsoft AI's in-house image model (model version 2026-06-02), built by the Microsoft AI Superintelligence team. It is a diffusion model trained with a flow-matching objective that handles both text-to-image generation and precise image-to-image editing. At launch it ranked #2 for image editing and #3 for text-to-image on the Arena (LMArena) leaderboards — ahead of Imagen Ultra 4.0, FLUX-2-max, and Seedream-4.5, and behind OpenAI's gpt-image-2 and Google's nano-banana-2. A cheaper, faster MAI-Image-2.5-Flash variant launched alongside it.
How does this prompt generator work?
You fill in structured inputs — scene, subject, style, composition, lighting, intended use, mode, and constraints. Our AI assembles them into MAI-Image's preferred order (Scene → Subject → Details → Composition/Camera → Lighting → Style → Constraints), strips out vague "slop" words in favor of concrete visual facts, formats in-image text correctly, and adds a preserve list when you're editing — so the prompt drops straight into Foundry, the MAI Playground, or other MAI-Image surfaces.
What is MAI-Image-2.6 best at?
Carried over from 2.5 and improved in 2.6, three things stand out: (1) surgical localized editing — replacing an object, updating in-image text, removing motion blur, or inpainting without changing the rest of the image; (2) facial-identity preservation — keeping a person recognizable across pose, expression, and viewpoint changes; and (3) photorealism with accurate in-image text rendering. That makes it strong for branding, product shots, portraits, presentations, and retouching existing images.
What is the difference between Generate and Edit mode?
Generate creates a brand-new image from a text description. Edit modifies an existing image you upload — write the instruction as "Change only X, keep everything else identical," then list what to preserve (face, layout, lighting, brand text). MAI-Image is specifically tuned for these targeted, surgical edits, and 2.6 can also take several reference images, so being explicit about what stays the same gives the cleanest results.
How good is text rendering in MAI-Image-2.6?
In-image text is one of its core strengths, and Microsoft reports a +91 Elo gain on text rendering in 2.6 over 2.5. To get the best results, put the exact text in double quotes, specify the font style, weight, color, and size, and state where it should sit in the frame. Keep the text short and spell unusual brand names out letter-by-letter. This is why it is wired into PowerPoint — it can render readable headlines and labels directly inside a generated graphic.
Where can I use the prompts, and how do I access MAI-Image-2.6?
MAI-Image-2.6 and 2.6-Flash are in public preview in Microsoft Foundry (since September 4, 2026) and in the MAI Playground, and Microsoft says they are rolling out to other Microsoft products. MAI-Image-2.5 remains available through Microsoft Foundry / Azure AI Foundry (REST endpoints /mai/v1/images/generations and /edits, model id "MAI-Image-2.5"), the MAI Playground, OpenRouter (model "microsoft/mai-image-2.5"), and fal.ai, plus product integrations in PowerPoint (live), OneDrive (rolling out), and Bing Image Creator / Copilot. The prompts this tool produces also work well on other modern image models thanks to their concrete, structured language.
How much does MAI-Image-2.6 cost?
Pricing is token-based, not per image. At Foundry preview pricing, MAI-Image-2.6 is $5 per 1M text-input tokens, $8 per 1M image-input tokens and $38 per 1M image-output tokens; MAI-Image-2.6-Flash is $1.75 / $2.50 / $19. For comparison, MAI-Image-2.5 is $5 per 1M text-input tokens, $8 per 1M image-input tokens, and $47 per 1M image-output tokens. The cheaper MAI-Image-2.5-Flash is $1.75 / $1.75 / $19.50 per 1M tokens. Use Flash for fast, low-cost drafts and the full model for final fidelity. (This prompt generator itself is free.)
Does MAI-Image-2.6 support negative prompts and what resolution does it output?
There is no separate negative-prompt parameter — phrase exclusions in plain language inside the prompt ("no watermark, no duplicate text, no extra fingers"), which our generator does for you. MAI-Image-2.6 supports dynamic aspect ratios and resolutions up to 1.5K; MAI-Image-2.5 outputs PNG up to 1024×1024.
Is this tool free to use?
Yes. You get 1 free prompt generation per day with no signup required. For unlimited access, sign up for a Promptslove membership which includes all AI tools and 29,000+ premium prompts.
Want Unlimited AI Prompt Generation?
Get unlimited access to all AI tools, 29,000+ premium prompts, courses, and resources designed to maximize your creative output.