THE SHORT VERSION Keep the creative brief. Replace the model adapter. xAI has put a date on the old Quality slug and changed what an omitted quality setting means. | • | On September 2, xAI said grok-imagine-image-quality will retire on November 2, 2026. |
| • | After that date, calls to the old slug will be served by grok-imagine-image-2.0 with quality set to low; xAI says the request and response shape will not change. |
| • | Since August 28, omitting quality on Image 2.0 uses Auto. Auto currently routes image generation to low and image editing to medium. |
| • | For repeatable work, pin the model and quality explicitly, then rerun one anchor brief before moving production traffic. |
ONE · WHAT CHANGED, AND WHEN xAI’s September 2 release note gives the old Quality endpoint a firm retirement date: November 2, 2026. The slug is grok-imagine-image-quality. After the deadline, requests sent there will be served by grok-imagine-image-2.0 with quality set to low. xAI says the request and response shape stays the same and the per-image price is lower. The quiet part arrived first. On August 28, xAI changed the default for grok-imagine-image-2.0. If quality is omitted, the request now uses Auto instead of medium. Auto currently sends a new generation to low and an edit to medium. The same update raised image editing to five source images and added 21:9 and 5:2 output ratios. That makes this a configuration story with a visual consequence. A prompt library can be perfectly intact while its output shifts because the model route or quality setting underneath it changed. TWO · YOUR PROMPT DID NOT FAIL; YOUR ADAPTER BECAME IMPLICIT A useful image prompt describes the job: the subject, frame, light, material, exact text, exclusions, and the details that must survive an edit. Model names, quality levels, aspect-ratio parameters, and reference-image order are not the job. They are the adapter that delivers the job to a particular system. When the adapter is implicit, a provider can move a default and your creative brief appears to become less reliable overnight. The repair is not to stuff more adjectives into the prompt. Write the adapter beside the brief and version both. For Grok, record grok-imagine-image-2.0 plus low or medium. Use low when speed and cost are the constraint; use medium when preserving small text, product edges, and controlled edits matters more. Do not call one setting universally better. Compare it against the visible job. If low passes the same acceptance sheet, keep the cheaper route. If it loses label spacing, glass edges, typography, or edit containment, the quality saving is not a saving. THREE · MAKE THE MIGRATION IN TEN MINUTES First, search code, automations, saved presets, and runbooks for grok-imagine-image-quality. Replace it with grok-imagine-image-2.0 before November 2 rather than relying on the automatic fallback. Second, add an explicit quality value. Do not leave it blank in a repeatable production route. Third, choose one anchor brief with difficult material, small but important text, negative space, and a targeted edit. Render it on the old route if you still can, then on Image 2.0 at the quality you intend to use. Fourth, score identity, composition, text accuracy, material, edit containment, and delivery size. Fifth, save the output beside the model, quality, date, aspect ratio, and reference order. That record is the difference between a real regression check and a memory of which thumbnail felt nicer. FOUR · UPDATE ALL THREE IMAGE ROUTES, NOT ONLY GROK This week is a good moment to clean the whole model card. OpenAI retired the official DALL·E GPT on August 30 and directs image creation and editing to ChatGPT Images; user-created GPTs with image generation are not affected. Keep a ChatGPT adapter that states the deliverable, ratio, reference role, and the one thing an edit may change. Google’s current image guide calls Gemini 3.1 Flash Image, or Nano Banana 2, its versatile general-purpose route. It lists the Lite model for speed and scale and Gemini 3 Pro Image for the most complex asset production. Keep those routing choices outside the creative brief so a model change does not rewrite the object you are trying to make. For Grok, use Image 2.0 and pin quality. Across all three, keep one invariant brief and one short adapter. The brief is owned craft. The adapter is rented infrastructure. PROMPT OF THE DAY Use one anchor brief to catch configuration drift. Keep the creative paragraph unchanged; record the model and quality on a separate run sheet. Create a 4:5 launch image for a fictional skincare product. Show one unbranded amber frosted-glass mist bottle with a matte cream cylindrical cap, fully visible in the right third at eye level. Leave the left forty-five percent as uninterrupted warm-lilac negative space. Use soft window light from the upper left, visible amber translucency at the glass edge, and one restrained oval contact shadow. The label area must stay blank and free of letters. MUST HOLD: one bottle, complete cap, full base, unchanged camera height, clean left field, no cropped edges. Exclude hands, flowers, water, extra packaging, logos, badges, and decorative marks. For the edit check, use this exact instruction: Change only the background; preserve bottle geometry, scale, glass color, cap shape, label area, camera, and shadow. The load-bearing line: “Change only the background; preserve bottle geometry, scale, glass color, cap shape, label area, camera, and shadow.” It makes edit containment visible. Run the unchanged brief at the quality you plan to ship, then inspect the protected bottle rather than judging the overall mood. START HERE Open the xAI release note, find every use of the old Quality slug, and move one anchor brief to grok-imagine-image-2.0 with an explicit quality value. Save the old and new outputs side by side with their settings. Do this before changing any creative language. CHANGE ONLY: QUALITY ••• I am turning this into a one-page image-model migration card: the invariant brief, the ChatGPT, Gemini, and Grok adapters, and a six-point acceptance sheet. Want the finished card? Reply with send me the migration card and I will share it when it is ready. A QUESTION FOR YOU Which part of an image workflow do you still leave implicit: model, quality, ratio, reference order, or edit boundary? Reply with the one field you want on the migration card. I will use the answers to decide what deserves the largest space. If this was useful, forward it to someone whose prompt stayed the same while the output quietly moved. Until next time, Luxe Prompting Luxe Prompting IMAGE PROMPTING AND IMAGE MODEL CRAFT Luxe Prompting · 1428 Bryn Mawr St, Scranton, PA 18504, United States Email service provider: beehiiv | 
