Grok Imagine 2.0 vs GPT Image, Nano Banana & Seedream

Grok Imagine 2.0 is xAI's image model, released in October 2026. I tested it against three other image models on 10 design jobs that all have text in them: posters, a menu, a logo, an invitation, an edit and a two-reference job. Each model got the identical prompt, and I scored the first image each one returned.

The five setups, all at 2K resolution:

• Grok Imagine 2.0, Low quality. Grok Imagine 2.0 has two quality settings, Low and Medium. Low is the faster one.
• Grok Imagine 2.0, Medium quality. Medium is Grok Imagine 2.0's highest setting.
• GPT Image 2.5 Sunburst (OpenAI), High quality. GPT Image 2.5 has five quality settings: Low, Medium, High, Extra High and Max. I tested High, which is the default on FontVibe. Extra High and Max were not tested.
• Nano Banana 2.1 (Google). Nano Banana 2.1 has no quality setting.
• Seedream 5.0 Pro (ByteDance). Seedream 5.0 Pro has no quality setting.

The results

Scores are out of 10 and averaged over the first image of each of the 10 jobs. Median time covers 18 new images per model (jobs 1 to 8 and 10, two images each) at 2K. Credits are what one image at these settings costs on FontVibe. Every individual score is in the appendix. In a separate blind pass with the model names hidden, GPT Image 2.5 also came first and Seedream 5.0 Pro also came last.

GPT Image 2.5 Sunburst scored highest, 8.90 out of 10. Grok Imagine 2.0 Medium scored 8.84, 0.06 lower. The first four setups are within 0.35 points of each other. Seedream 5.0 Pro scored 8.20, 0.35 below the fourth-place setup.

GPT Image 2.5 was tested at High, the third of its five quality settings. Grok Imagine 2.0 Medium is Grok's highest setting. So this comparison puts GPT Image 2.5's default setting against Grok's top setting, and this test does not show how GPT Image 2.5 Extra High or Max would score.

Text accuracy

Every word the prompts asked for was spelled correctly in all 50 scored images, including the Simplified Chinese poster and the Japanese poster. Four images had smaller text problems:

• Nano Banana 2.1, job 2: the dot in "AFTER DARK CLUB · EST. 2026" became a hyphen.
• Grok Imagine 2.0 Low, job 8: the dot in "OPEN LATE · TAKEAWAY" is missing.
• Nano Banana 2.1, job 8: a chalkboard sign the prompt did not ask for has garbled, unreadable lettering.
• Seedream 5.0 Pro, job 8: a paper sign and a street sign the prompt did not ask for have garbled lettering.

Three images also contain whole lines of text the prompt did not ask for: Grok Imagine 2.0 Low in job 1 ("Let's celebrate!") and job 3 ("LIVE HARD. BURN LOUD. NO APOLOGIES."), and GPT Image 2.5 in job 8 (a "GOOD NOODLES GOOD MOOD" poster on the door).

How I tested

The 10 jobs. Jobs 1 to 3 use the same prompts as my earlier test of seven AI design tools.

1. Kids' birthday invitation (3:4): glossy balloon letters, three lines of details
2. T-shirt chest print (1:1): wildstyle graffiti with drips, isolated on white
3. Gig poster (2:3): molten fire lettering, four lines of details
4. Chalkboard café menu (3:4): 9 items with prices, an accented É and an en dash
5. Chinese movie poster (2:3): a title and two lines in Simplified Chinese
6. Japanese ramen poster (3:4): brush lettering, kana and kanji, full-width brackets
7. Bakery wordmark (1:1): "Crumb & Co." hand lettered, one color
8. Photo of a noodle shop (3:2): a neon sign and a painted window sign in the scene
9. Edit an existing image (16:9): change one headline, keep everything else
10. Two reference images (1:1): lettering from image 1, colors from image 2

The test covers design images with text. Portraits, landscapes and other images without text were not tested. The full prompt for each job is printed in its section below. The rules:

• Every model got the identical prompt, word for word, through FontVibe's DIY mode, which sends the prompt to the model unchanged. No prompt rewriting and no style presets.
• The aspect ratio was set as a parameter, the same for every model.
• Each setup made two images per job. Only the first image is scored and shown. The second image was used only to check spelling consistency.
• All 100 images were generated on October 21, 2026 and are public on FontVibe.

The scoring rubric

The rubric is FontVibe's internal standard for rating generated designs, written in October 2026 before this test. I also used it in the earlier test of seven tools. Each image gets four scores from 1 to 10, and I wrote down the reason for every deduction before giving the number.

The overall score is calculated, not judged: 0.20 × legibility + 0.35 × lettering + 0.25 × information and layout + 0.20 × appeal.

I knew which model made each image while scoring it. To check for bias, I also ran a blind pass where the model names were hidden.

Jobs 9 and 10 start from existing images. All three were taken from FontVibe's public gallery, and none was made by the four models being tested. All three were made by earlier Seedream models (Seedream 5.0 Lite and Seedream 5.0), the same model family as Seedream 5.0 Pro.

Job by job

Job 1: Kids' birthday invitation

Birthday party invitation for my daughter, "Mia's 7th Birthday" in glossy balloon letters, "Saturday, Nov 14 · 2 PM", "Sunny Park Pavilion", "RSVP by Nov 7", bright and playful, portrait card

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Use as Reference

GPT Image 2.5 and Grok Imagine 2.0 Medium tied for the highest score, 9.2. Both made glossy foil balloon letters and grouped the three detail lines in labeled cards. Grok Imagine 2.0 Low lost 1 information-and-layout point for adding a "Let's celebrate!" banner. Nano Banana 2.1 scored lowest, 8.55: its date line is tightly spaced and the three detail lines use three different typefaces. For balloon lettering as a preset, see the bubble letter style.

Job 2: T-shirt chest print

Front chest print for my streetwear brand, "NIGHTSHIFT" in wildstyle graffiti with drips, small "AFTER DARK CLUB · EST. 2026" underneath, a crescent moon graphic, isolated on plain white so I can send it to the printer

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Grok Imagine 2.0 Medium scored highest, 9.0, with arrow-tipped wildstyle letters, drips and a purple 3D edge that still read as NIGHTSHIFT. GPT Image 2.5 made the most interlocked wildstyle of the five, and its legibility score is 7 because the letters take effort to read. Grok Imagine 2.0 Low scored lowest, 8.3: its letters are a brush tag style without the interlocking shapes that define wildstyle. Nano Banana 2.1 turned the "·" into a hyphen. For more on this look, see the graffiti text style.

Job 3: Gig poster

A3 gig poster for my band "BURNING ORCHARD", band name in molten fire lettering as the main focus, "with special guests THE LOW TIDES", "SAT 14 NOV · DOORS 8 PM", "THE CELLAR · 22 MILL STREET", "TICKETS £12 ON THE DOOR", gritty rock atmosphere with burning trees

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Grok Imagine 2.0 Medium scored highest, 9.2, with no deductions on information and layout: band name, support act, date, venue and price read in that order. Grok Imagine 2.0 Low added the line "LIVE HARD. BURN LOUD. NO APOLOGIES." at the bottom, which the prompt did not contain. Seedream 5.0 Pro scored lowest, 8.5, because it placed the support act above the headliner, which gets the billing order wrong. For fire lettering as a preset, see the fire text style.

Job 4: Chalkboard café menu

Chalkboard menu for my coffee shop, title "SUNNY SIDE CAFÉ" in hand-chalked lettering, then this exact list with prices, one item per line: "Espresso 3.20", "Cappuccino 4.10", "Flat White 4.30", "Oat Latte 4.60", "Matcha Latte 4.80", "Cold Brew 4.50", "Croissant 3.40", "Banana Bread 3.80", "Avocado Toast 7.90", and a small footer "Free Wi-Fi · Open 7 AM–4 PM"

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Use as Reference

All five models wrote all 11 required lines correctly, including the É in CAFÉ and the en dash in 7 AM–4 PM. GPT Image 2.5 scored highest, 9.2: its letters have visible chalk grain, and it separated drinks from food with a ruled line. The menus from Grok Imagine 2.0 and Nano Banana 2.1 use cleaner, print-like letters. Seedream 5.0 Pro scored lowest, 7.75: the board is small in the frame and a colon was added before every price. For chalk lettering as a preset, see the chalk text style.

Job 5: Chinese movie poster

Movie poster, title "长夜将尽" in heavy cracked-steel lettering, below it "十一月十五日 全国上映" and the tagline "黎明之前 无路可退", rainy city at night, all text in Simplified Chinese, no other text

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Use as Reference

All five models used Simplified Chinese, wrote all 22 required characters correctly and added no other text. Grok Imagine 2.0 Low scored highest, 9.0. GPT Image 2.5's cracks are the deepest of the five, and its legibility score is 8 because the last character, 尽, takes a second look. Nano Banana 2.1 scored lowest, 8.05: its letters read as rusted metal rather than cracked steel. For steel and chrome lettering, see the metallic text style.

Job 6: Japanese ramen poster

Summer menu poster for a ramen shop, headline "冷やし中華はじめました" in bold brush lettering, sub-line "期間限定 8月31日まで", price tag "980円（税込）", fresh blue and white summer feel, no other text

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Use as Reference

All five models wrote the kana, kanji and full-width brackets correctly. Grok Imagine 2.0 Low scored highest, 9.2, with visible brush strokes and the dish shown under the headline. Nano Banana 2.1's headline is an outlined display face rather than brush lettering. Seedream 5.0 Pro has some of the best brush lettering in this round but shows no food, so the lower half of a menu poster is empty. Both scored 8.3. For brush lettering, see the Japanese brush ink style.

Job 7: Bakery wordmark

Wordmark logo for a small bakery called "Crumb & Co.", hand-lettered with a playful baseline and a tiny wheat sprig on the ampersand, flat vector look, one color, on plain white, no other text

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Use as Reference

GPT Image 2.5 scored highest, 9.2: the baseline bounces and the wheat sprig grows out of the ampersand, as the prompt asked. Nano Banana 2.1 scored 9.0 for a similar result. Both Grok Imagine 2.0 settings scored 7.85, the lowest in this job: the baseline is nearly flat and the letters look like an existing typeface rather than hand lettering.

Job 8: Photo of a noodle shop

Realistic photo of a small noodle shop at dusk, a glowing neon sign above the door that reads "LUNA NOODLE BAR" and a hand-painted window sign "OPEN LATE · TAKEAWAY", wet street reflections, shot on a 35mm lens

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Grok Imagine 2.0 Medium scored highest, 9.2: both signs sit in the scene with the right perspective and nothing else has text. Nano Banana 2.1 and Seedream 5.0 Pro each added extra signs with garbled lettering. Garbled lettering is a hard error under the rubric, so both got a legibility score of 4, and the unrequested signs cost each 1 information-and-layout point. Seedream 5.0 Pro scored lowest, 7.2, and Nano Banana 2.1 scored 7.75. GPT Image 2.5 made the most convincing street but added a "GOOD NOODLES GOOD MOOD" poster. For signage as a preset, see the neon sign text style.

Job 9: Edit an existing image

Change the headline "Thank you" to "Welcome back" and keep everything else exactly as it is

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Use as Reference

All five models changed the headline, kept the hand-lettered style and left the other three lines unchanged, so all five scored 9.0. The rubric does not measure whether the rest of the image stayed the same, so I measured that separately. Each result and the original were resized to 800 × 450 and converted to grayscale, and I took the mean absolute pixel difference (0 to 255) outside a box around the headline. Lower means closer to the original.

The original is 2848 × 1600 pixels. Seedream 5.0 Pro and GPT Image 2.5 returned the same size. Nano Banana 2.1 returned 2752 × 1536 and Grok Imagine 2.0 returned 2816 × 1584, so part of their difference comes from resizing. On its second image, Grok Imagine 2.0 Low's difference was 15.9.

Job 10: Two reference images

Square social post that says "WEEKEND MARKET" with a small line "Every Saturday · 9 AM". Use the lettering style of reference image 1 and the color palette of reference image 2.

Reference image 1 was pastel balloon letters. Reference image 2 was a red, yellow and blue pop art comic burst. The correct result is balloon-shaped letters in red, yellow and blue.

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Use as Reference

Nano Banana 2.1 was the only model whose first image got both parts right, and it scored 9.2. GPT Image 2.5 made balloon letters but colored them with reference 1's pastels. Grok Imagine 2.0 Medium made balloon letters in mostly red and yellow, with some pink and blue from reference 1. Grok Imagine 2.0 Low and Seedream 5.0 Pro copied the polka-dot comic lettering from reference 2 instead of the balloon shapes. Seedream 5.0 Pro scored 7.1, the lowest score in this test.

Grok Imagine 2.0: Low or Medium?

Use as Reference

Grok Imagine 2.0 Medium scored 8.84 overall and Grok Imagine 2.0 Low scored 8.59. Medium scored higher on lettering (8.5 vs 8.1) and on information and layout (8.8 vs 8.6), and neither of its 10 scored images contains text the prompt did not ask for. Low added unrequested text in 2 of 10. Medium's median time was 84 seconds per image against 23 seconds for Low, and on FontVibe a 2K Medium image costs 40 credits against 30 for Low.

For a final image that will be printed or published, Medium scored higher. For trying several ideas quickly, Low's median time was 27% of Medium's.

Grok Imagine 2.0 Medium had the highest legibility score in this test, 9.9, and the highest score in 4 of the 9 jobs that were not tied (invitation, T-shirt, poster, noodle shop). Its lowest score was the bakery logo, 7.85.

To try Grok Imagine 2.0, choose it in the model menu on FontVibe's generator. Hovering over any Grok image in this post shows a Recreate button that runs the same prompt.

Each model, briefly

GPT Image 2.5 Sunburst (High), 8.90. Highest lettering, information-and-layout and appeal scores (information and layout tied with Grok Imagine 2.0 Medium). Highest score in 3 of the 9 jobs that were not tied (invitation, shared with Grok Imagine 2.0 Medium; menu; logo). Lowest score 8.5, on the reference job. Tested at High; Extra High and Max were not tested.

Grok Imagine 2.0 Medium, 8.84. Highest legibility score, 9.9. Highest score in 4 of the 9 jobs that were not tied (invitation, shared with GPT Image 2.5; T-shirt; poster; noodle shop). Lowest score 7.85, on the logo. Median time 84 seconds.

Grok Imagine 2.0 Low, 8.59. Median time 23 seconds, the fastest in this test. Highest score in 2 jobs (Chinese poster, Japanese poster). Added text the prompt did not ask for in 2 of 10 images.

Nano Banana 2.1, 8.55. Costs 25 credits per 2K image on FontVibe, the lowest of the four models. The only model whose first image followed both reference images correctly in job 10. Garbled background text in job 8.

Seedream 5.0 Pro, 8.20. Lowest overall score and lowest information-and-layout score, 7.8. Closest to the original in the edit job, with a pixel difference of 3.2. Garbled background text in job 8.

A blind check

Because I knew which model made each image while scoring, I also ran a blind pass. A private page showed the five images for each job in random order with the model names hidden. Each image got a 1 to 5 score for the lettering, a 1 to 5 score for the layout, and one of three verdicts: ready to use, needs a small fix, or unusable.

The blind pass and the rubric agree on first place (GPT Image 2.5) and last place (Seedream 5.0 Pro). They disagree on the order of the middle three.

Same request, given to FontVibe's designer

FontVibe also has a designer mode, where you describe what you want and it writes the prompt for the model. I ran jobs 3 and 4 through it with Grok Imagine 2.0 Low, using each job's prompt as the request. These images are not part of the scores.

Use as Reference

Use as Reference

Use as Reference

Use as Reference

The designer version of the poster does not contain the "LIVE HARD. BURN LOUD. NO APOLOGIES." line that Grok Imagine 2.0 Low added on its own. Its layout and lettering were not better than Grok's own versions: the poster title changed from spiked, metal-poster fire lettering to a plain condensed typeface, and the menu lost the dot leaders between items and prices.

The designer was given the same complete prompts used in the test, with the lettering, scene and every line of text already specified. Designer mode is built for one-sentence requests, such as "a poster for my band's Saturday show at The Cellar", where it writes the missing content and the design. That use was not tested here.

Limits of this test

• One image per model per job was scored, and one person scored all 50.
• GPT Image 2.5 and Grok Imagine 2.0 Medium are 0.06 apart, and Grok Imagine 2.0 Low and Nano Banana 2.1 are 0.04 apart. Changing one score by 2 points in one job moves a model's average by 0.04 to 0.07, so those two orders depend on single scores.
• GPT Image 2.5 was tested at its third of five quality settings. Grok Imagine 2.0 was tested at both of its settings.
• The edit and reference images came from earlier Seedream models.
• The results describe these model versions on October 21, 2026.

Every image is public on FontVibe and every score has a written reason. If you rerun a prompt and get a very different result, email hello@fontvibe.ai and I'll add it to this post.

Appendix: every score

Legibility / lettering / information and layout / appeal, then the weighted overall. First image of each setup.

The written reason for every score is kept on file and available on request.