Grok Imagine 2.0 region editing

Grok Imagine Image 2.0, shipped 7–8 August 2026, is the stills row you reach for when the job is "change this part" or "combine these files," not "roll another poster and hope the bottle survives."

xAI's 2.0 stills model landed second on both text-to-image and editing arenas, with region-level editing, multi-image compositing, templates, and Quality Mode as the default inside Grok (kingy.ai's Grok Image 2 comparison; OrcaRouter on Quality Mode; InVideo's GPT Image roundup, for the 21 April GPT Image 2 date). GPT Image 2 remains first on those arenas. Nano Banana 2 remains the fast Google stills line (26 February 2026). This page is not a three-way quality ranking. It is the 2.0 capability that changed the brief: edit a region, composite a stack, stop regenerating the whole frame to move a shadow.

On Versely, the 2.0 stills are Grok Imagine Image 2.0 (text-to-image, 1K or 2K, 2 credits, discounted from 5) and its edit sibling, Grok Imagine Image 2.0 Edit (5 credits, up to three source images plus a text instruction), which the 2.0 page lists as a variant. The older stills rows are Grok Imagine Image Quality (Quality tier, up to 2K, 4 credits) and Grok Imagine (faster ideation stills, 4K listed, 2 credits). Quality Mode is a Grok-app default in the August reporting; in a multi-model studio you pick the row when you mean it.

What 7–8 August actually shipped

Treat the August date as a product boundary. Before it, "Grok Imagine" in most briefs meant a loosely-styled still or a video cousin. After it, the stills story is 2.0: arena rank, region edit, compositing, templates, Quality Mode.

Arena rank is a preference score, not a packing spec. #2 on text-to-image and on editing means people in those arenas preferred 2.0 to most of the field, and not to GPT Image 2. Use that as a prior when the job is "make me a still I'll iterate on," not as a reason to abandon a pack shot that already works on Nano Banana 2.

Templates, in that reporting, are starting points — a layout, a genre, a composition you are not inventing from a blank prompt. They are not a brand kit. A template will not remember your hex. If the still has to match a kit, the kit still has to be injected, and the product still has to be a reference.

Video is a different family. Grok Imagine Video is a published video row (durations 6 / 10 / 15 / 20 / 30s, 720p, native audio). Grok Imagine Extend continues a Grok clip. Neither is region editing. Do not ask the video row to recrop a label. Do not ask the stills Quality row for fifteen seconds of motion.

Grok the chat brain is a third object. If you pin a Grok planner in an agent host, you have not selected Imagine 2.0. Name the stills slug when you want pixels.

Region edit is not a reroll

A reroll samples a new image. A region edit takes the pixels you already approved and changes the part you named.

That is the job 2.0 is for. The bottle is right; the condensation is wrong. The room is right; the ugly exit sign has to go. The face is right; the necklace is from last season. Regenerating the whole frame to fix those notes is how you lose the bottle, the room, and the face. Region edit is how you keep them.

Write the instruction like an art director, not like a prompt poet:

• Name the region. "The label on the carton," "the sky through the window," "the necklace only."
• Name the change. Replace, remove, recolour, relight that region.
• Name the freeze. "Everything else unchanged" is not fluff. It is the contract.

If you cannot point at the region on the still, you do not have a region edit. You have a new generate. New generates belong on a text-to-image row — GPT Image 2 when obedience to a long prompt is the job, Nano Banana 2 when speed is the job, Grok Imagine (the 2-credit card) when you want a weird first take. Grok Imagine Image 2.0 Edit is what you open after a take is almost the one.

Region edit is also how you stop paying video prices for stills problems. The Lite-to-Flash video loop is for motion notes. A label on a pack is a stills note. Fix it here, then animate. Sending a wrong label into image-to-video and hoping Omni Flash will typeset it is the expensive cousin of a reroll.

Limits are real even when the arena rank is pretty. Fine type, tiny logos, and six conflicting notes in one turn will smear. One region, one change, then look. If the region is the entire frame, you are generating.

Composite several stills

Multi-image compositing is the other 2.0 job: several files in, one frame out. On Versely the Edit row takes up to three source images and a text instruction.

The honest use is production-shaped. A signed pack shot, a signed hand, a signed room — composite them instead of asking a single prompt to invent a hand that has never held this pack in this kitchen. The stack is the brief. The model is the glue.

This is not "style transfer as a vibe." If you wanted the look of image A on the subject of image B, say that, and still expect to inspect the label. Compositing a real SKU with a generated kitchen is how ecommerce actually ships. Compositing two generated faces into one spokesperson is how you invent a person. Know which one you are doing. Article 50 labelling for authentic-looking people is a deployer problem when you publish in the EU; a composite that resembles a real person is not exempt because Grok mixed the files.

Practical stack:

• Hero product still you would print
• Context still (room, outdoor, table) you would accept empty
• Optional hand or wardrobe if the format is UGC-shaped
• Instruction that says which file is the product and which is the plate

Do not dump twelve near-duplicates and hope. 2.0's compositing is not Seedance's twelve-reference video stack. It is a stills composite. Three good files beat twelve muddy ones.

Templates can sit in front of a composite when the layout is the thing you keep reusing — a shelf, a split before/after, a simple pack-on-table. The template is the grid. The composite is the SKU. The kit is the type. Keep those three nouns separate in the brief.

Quality Mode is a catalog row

Inside Grok, Quality Mode is the default in the August write-ups. Inside a studio catalog, defaults do not travel. You pick a slug.

Grok Imagine Image Quality is the Quality tier: text-to-image, quality-optimised, up to 2K, 4 credits. Use it when the still is the one you will edit, composite, or send to a client.

Grok Imagine is the faster, cheaper stills card: 2 credits, 4K listed, loosely-styled ideation. Use it when you want unexpected takes and you expect to throw most of them away. Do not run a ten-image ideation grid on the Quality row "because 2.0 is better." That is how you spend Quality credits on sketches.

2K on the Quality card is not a reason to skip it for social. Most feeds will not show 4K. 2K is enough to inspect a label and still cheap next to a video second. If the deliverable is a print poster, check whether 2K is the ceiling you can live with, or whether a different family owns print.

Versely prices these rows in credits: 2 for Grok Imagine Image 2.0, 4 for Quality, 5 for 2.0 Edit. xAI's consumer prices are not Versely rates.

The AI video generator is the wrong launcher for this page's job. Stills first. Motion later, on a video slug, once the region you cared about is the region you approved.

When the next note is "fix this part" rather than "roll another poster," sign up and run Grok Imagine Image 2.0 Edit on the file you already almost like.

FAQ

Is Grok Imagine Image 2.0 better than GPT Image 2 because it is newer?

No. GPT Image 2 launched 21 April 2026 and sits first on the arenas. 2.0 launched 7–8 August and sits second, with region edit and compositing as the reason to pick it. Newer is a date. The job is the picker.

Which Versely URL is "Grok Imagine 2.0"?

Grok Imagine Image 2.0 is the 2.0 text-to-image row at 2 credits, and Grok Imagine Image 2.0 Edit at 5 credits is its edit sibling. For Quality tier stills, use Grok Imagine Image Quality. For fast ideation stills, use Grok Imagine. For motion, use the video and extend rows, which are not this page.

Can region edit replace a brand kit?

No. Region edit changes pixels in a region. A kit injects hex, type, logo, and tone into later jobs. You still need both: the kit so the next still starts on-brand, the region edit so you do not reroll a good one to move a shadow.