Changing a photo from flat daylight to warm evening light sounds like a small edit. In practice, the light has to agree with the shadows, reflections, background and subject. If only the color changes, the result often looks like a filter rather than a believable photograph.
Microsoft's October 27, 2026 announcement of MAI Image 2.7 makes AI photo relighting a timely skill to revisit. The company describes its new image-editing preview as supporting scene restyling, relighting, restoration, and text or symbol edits. This guide focuses on a practical application: changing the lighting of an existing image without accidentally redesigning its subject.
What is new in MAI Image 2.7?
Microsoft's launch announcement positions MAI Image 2.7 as an image-editing model and highlights stronger control over scene and style changes. Seeking Alpha's October 27 reporting independently confirms the announcement and reports availability on MAI Playground and evaluation platforms.
Microsoft also reports improvements over version 2.6 across the editing categories it cites. Those are vendor-reported evaluation results, not proof that every individual photo will improve. The workflow below is an editorial recipe, not a hands-on benchmark of MAI Image 2.7. Preview access, limits and terms can change; check the service you use before uploading client material.
This is not an announcement that MAI Image 2.7 is available inside AI Photo Generator. You can apply the same briefing principles to an image-to-image editor you already use, but supported controls and results will vary.
Start by separating lighting from identity
A good relighting brief has two lists: what may change and what must remain fixed. For a portrait, facial proportions, expression, hairstyle and clothing usually belong on the fixed list. For a product, protect its silhouette, material, color, label and visible components.
Choose a source photo with a clearly visible subject, enough detail for the intended output, and lighting that has not erased important features. A very dark or clipped area may require invented detail to make it visible; generated reconstruction is not evidence of what was originally there.
Save the original separately. Work from an image you own or have permission to edit, and get appropriate consent before uploading someone else's portrait. For confidential products or client assets, review the editor's data-handling terms first.
A reusable AI photo relighting prompt
Structure the instruction around five decisions: the lighting change, its direction, its softness, the environment, and the protected details.
Relight this existing photo with soft, warm late-afternoon light coming from camera-left. Keep the same subject, pose, camera angle, framing and background layout. Preserve facial features, clothing details and original material colors. Adjust cast shadows and reflections to match the new light direction. Keep skin texture natural and avoid blown highlights. Change lighting only; do not add objects or redesign the scene.
Replace the portrait-specific details with product constraints when needed. Directions such as camera-left are useful, but verify that the result follows them: a written instruction is not a locked editing control.
Three useful lighting briefs
1. A cleaner studio product photo
Relight the supplied product photo as a neutral studio shot with one large soft light above and to camera-left, plus gentle fill from the right. Preserve the exact product shape, label wording, logo, surface finish and base color. Keep the original viewpoint and background. Create a believable soft contact shadow beneath the product. Do not add accessories or alter packaging.
Review the label at full size. A more attractive reflection does not compensate for a changed ingredient list, missing button or invented brand mark. If exact packaging cannot be preserved, use a conventional editor to retain the original product and adjust its surroundings instead.
2. A warm portrait without orange skin
Give this portrait gentle golden-hour illumination from camera-right, with soft fill on the opposite side. Preserve the person's facial structure, expression, skin texture, hairstyle and clothing. Keep skin tones believable rather than uniformly orange. Retain the existing composition and background objects. Do not smooth the face or change age or body shape.
Compare the eyes, teeth, hairline and outline of the face with the source. Reject an image that beautifies or changes the person when the brief was only to change the light.
3. A moody interior with readable detail
Relight this room for a calm evening mood using the visible table lamp as the main warm light source. Preserve the furniture, room geometry, window positions and surface materials. Keep subtle ambient light so dark areas remain readable. Match shadows and reflections to the lamp's position. Do not add lighting fixtures or change the decor.
For property listings, do not let the edit misrepresent fixtures, views, condition or room features. Use clear disclosure where appropriate and follow the listing platform's rules.
Use a one-change-at-a-time workflow
- Set one target. Choose studio softness, golden-hour warmth or evening atmosphere. Avoid combining a new background, outfit, pose and lighting setup in the first pass.
- Generate a small comparison set. Keep the source image and core brief unchanged. Where available, record model version and edit settings so you know what produced each candidate.
- Compare with the source. Review both the full composition and close crops of protected details. A side-by-side view is enough; an overlay in a conventional editor can make shape changes easier to spot.
- Correct the largest mismatch. Ask for one correction, such as softer shadows or a less warm skin tone. If the subject has drifted, restart from the original rather than repeatedly editing the flawed result.
- Export and inspect again. Review the actual delivery file, not just the preview. Check the intended crop, dimensions, compression and any readable text.
A five-point realism check before publishing
- Direction: Do the illuminated side of the subject and the cast shadows agree with the requested light source?
- Contact: Do feet, furniture and products still feel grounded rather than floating?
- Reflections: Do glass, metal and glossy surfaces respond plausibly to the new lighting?
- Continuity: Does the subject belong in the background, without bright cutout edges or incompatible color casts?
- Fidelity: Are identity, product details, room features and important text unchanged?
Treat fidelity as a pass-or-fail requirement, not something to average against visual appeal. An attractive image with an incorrect product is still unusable.
When should you use a conventional photo editor instead?
If the task is only exposure, white balance or a subtle tonal correction, start with ordinary adjustment tools. They offer a more controlled route when preserving exact pixels and documentary accuracy matters. Generative relighting is more appropriate for a creative reinterpretation in which you can inspect and approve the changed scene.
Ready to build your brief? Start with one image in AI Photo Generator, select an available image-to-image workflow suited to your task, and specify both the lighting change and the details to preserve. The useful lesson from this week's editing-model news is not to chase a leaderboard position: it is to make the requested change precise enough that you can tell whether the result actually succeeds.