{"id":2461,"date":"2026-06-11T09:00:00","date_gmt":"2026-06-11T09:00:00","guid":{"rendered":"https:\/\/wiro.ai\/blog\/?p=2461"},"modified":"2026-09-27T22:12:58","modified_gmt":"2026-09-27T22:12:58","slug":"ernie-image-turbo-6-prompt-tests","status":"publish","type":"post","link":"https:\/\/wiro.ai\/blog\/ernie-image-turbo-6-prompt-tests\/","title":{"rendered":"ERNIE Image Turbo: 6 Prompt Tests"},"content":{"rendered":"<p><strong>ERNIE Image Turbo prompt tests<\/strong> put Baidu&#8217;s fast text-to-image model through six deliberately different jobs: product photography, a rainy street with exact sign text, macro detail, two printed-text cases, and a crowded food scene. The point was not to declare a winner from one attractive image. It was to see where an eight-step model holds onto the request and where a careful brief still matters.<\/p>\n<p>The six images below are the original blog-hosted outputs. They were generated at 848&#215;1264 with one sample each, steps set to 8, guidance scale set to 1.0, prompt enhancement enabled, and seeds 201 through 206. Those settings matter: the model&#8217;s own Wiro documentation lists 8 steps, 1.0 scale, one sample, and prompt enhancement as its defaults. The documentation also says it accepts English, Chinese, and Japanese prompts.<\/p>\n<h2>Table of contents<\/h2>\n<ul>\n<li><a href=\"#setup\">Test setup and what was measured<\/a><\/li>\n<li><a href=\"#tests\">The six ERNIE Image Turbo prompt tests<\/a><\/li>\n<li><a href=\"#speed-cost\">Run time and cost on Wiro<\/a><\/li>\n<li><a href=\"#pick\">When to pick ERNIE Image Turbo<\/a><\/li>\n<\/ul>\n<h2 id=\"setup\">What these ERNIE Image Turbo prompt tests check<\/h2>\n<p>ERNIE Image Turbo is the distilled, faster member of Baidu&#8217;s ERNIE-Image family. Baidu describes the Turbo release as an eight-inference-step model aimed at fast generation, while the full ERNIE-Image release typically uses 50 steps. That makes this a useful test of practical prompt following rather than a maximum-quality, long-running setup.<\/p>\n<p>Each prompt isolates a failure mode. The earbud case asks for material, droplets, directional lighting, and shallow depth of field. The Tokyo scene asks for a readable English phrase inside a busy, wet night scene. The bee tests small natural detail. The street sign and notebook separate short typography from decorative poster text. The market test stresses composition, hands, faces, steam, bowls, and depth at once.<\/p>\n<table>\n<tbody>\n<tr>\n<th>Setting<\/th>\n<th>Value used<\/th>\n<th>Why it matters<\/th>\n<\/tr>\n<tr>\n<td>Model<\/td>\n<td><a href=\"https:\/\/wiro.ai\/models\/baidu\/ernie-image-turbo\">baidu\/ernie-image-turbo<\/a><\/td>\n<td>The model under test.<\/td>\n<\/tr>\n<tr>\n<td>Resolution<\/td>\n<td>848&#215;1264<\/td>\n<td>A vertical frame suited to the test images.<\/td>\n<\/tr>\n<tr>\n<td>Steps<\/td>\n<td>8<\/td>\n<td>The documented Turbo default and the core speed constraint.<\/td>\n<\/tr>\n<tr>\n<td>Guidance scale<\/td>\n<td>1.0<\/td>\n<td>The documented default used consistently across the set.<\/td>\n<\/tr>\n<tr>\n<td>Samples<\/td>\n<td>1 per prompt<\/td>\n<td>These are single-output checks, not a best-of-many selection.<\/td>\n<\/tr>\n<tr>\n<td>Prompt enhancement<\/td>\n<td>Enabled<\/td>\n<td>Wiro documents this option as an LLM refinement of the input prompt.<\/td>\n<\/tr>\n<tr>\n<td>Seeds<\/td>\n<td>201-206<\/td>\n<td>One fixed seed per numbered test.<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h2 id=\"tests\">The six ERNIE Image Turbo prompt tests<\/h2>\n<h3>1. Product photo realism<\/h3>\n<p><strong>Prompt:<\/strong> Commercial product photo of a matte black wireless earbud case on dark slate, tiny water droplets, softbox key light from left, subtle rim light, shallow depth of field, photorealistic, 50mm lens look<\/p>\n<p>The output gives the case a convincing matte surface, with sharply defined droplets on both the product and slate. The lighting does the requested job: the upper-left portion carries the brightest highlight while the lower body falls into shadow. The image reads as a product shot quickly. It does not prove that every commercial brief will land first time, but it shows that a focused, one-object request is a strong fit.<\/p>\n<figure>\n<img loading=\"lazy\" decoding=\"async\" width=\"848\" height=\"1264\" class=\"wp-image-2454\" src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-1.png\" alt=\"ERNIE Image Turbo prompt tests: matte black wireless earbud case with water droplets on slate\" srcset=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-1.png 848w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-1-342x510.webp 342w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-1-604x900.webp 604w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-1-768x1145.webp 768w\" sizes=\"auto, (max-width: 848px) 100vw, 848px\" \/><figcaption>Prompt: Commercial product photo of a matte black wireless earbud case on dark slate, tiny water droplets, softbox key light from left, subtle rim light, shallow depth of field, photorealistic, 50mm lens look<\/figcaption><\/figure>\n<h3>2. Rainy street scene with exact neon text<\/h3>\n<p><strong>Prompt:<\/strong> Candid night street photo in Tokyo after rain, neon shop sign must read BAY NOODLES in clear block letters, wet asphalt reflections, people with umbrellas in the distance, 35mm film grain, high contrast, photorealistic<\/p>\n<p>This is the clearest text-rendering result in the set. The large sign reads BAY NOODLES correctly on two lines, while a smaller OPEN sign remains readable at the right. Rain, umbrellas, pedestrians, and reflections are all present. The compromise is framing: the requested distant crowd is brought forward and the giant sign dominates the composition. For an ad mockup or editorial visual where the required phrase needs to survive, that trade is often acceptable. For a precise location photograph, it is still a synthetic scene rather than a documentary image.<\/p>\n<figure>\n<img loading=\"lazy\" decoding=\"async\" width=\"848\" height=\"1264\" class=\"wp-image-2455\" src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-2.png\" alt=\"Rainy Tokyo street at night with a readable BAY NOODLES neon sign\" srcset=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-2.png 848w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-2-342x510.webp 342w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-2-604x900.webp 604w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-2-768x1145.webp 768w\" sizes=\"auto, (max-width: 848px) 100vw, 848px\" \/><figcaption>Prompt: Candid night street photo in Tokyo after rain, neon shop sign must read BAY NOODLES in clear block letters, wet asphalt reflections, people with umbrellas in the distance, 35mm film grain, high contrast, photorealistic<\/figcaption><\/figure>\n<h3>3. Macro detail<\/h3>\n<p><strong>Prompt:<\/strong> Extreme macro photo of a honey bee landing on a lavender flower, visible pollen grains on fuzzy legs, translucent wing veins, creamy green bokeh background, razor sharp focus on eyes, natural sunlight, 100mm macro lens look<\/p>\n<p>The bee, lavender, wing veins, and pollen are all visible. The output also creates the requested shallow field with a soft green-and-purple background. Look closely and the bee is more idealized than a field photograph: the eye and wing structure look clean, but natural insect anatomy remains a place to check before using an image in science, education, or a product claim. For a lifestyle illustration, the result is persuasive.<\/p>\n<figure>\n<img loading=\"lazy\" decoding=\"async\" width=\"848\" height=\"1264\" class=\"wp-image-2456\" src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-3.png\" alt=\"Macro-style ERNIE Image Turbo output of a honey bee on lavender with pollen and visible wings\" srcset=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-3.png 848w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-3-342x510.webp 342w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-3-604x900.webp 604w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-3-768x1145.webp 768w\" sizes=\"auto, (max-width: 848px) 100vw, 848px\" \/><figcaption>Prompt: Extreme macro photo of a honey bee landing on a lavender flower, visible pollen grains on fuzzy legs, translucent wing veins, creamy green bokeh background, razor sharp focus on eyes, natural sunlight, 100mm macro lens look<\/figcaption><\/figure>\n<h3>4. Two-line text on a real object<\/h3>\n<p><strong>Prompt:<\/strong> Photo of a vintage metal street sign mounted on a red brick wall, crisp bold typography, text must read OAK STREET on the top line and SAN FRANCISCO on the bottom line, shallow depth of field, natural daylight, photorealistic<\/p>\n<p>The sign succeeds on the difficult part: both requested lines read OAK STREET and SAN FRANCISCO. Letter spacing is even enough to pass at normal viewing size, and the worn metal, mounting hardware, brick texture, and daylight create a believable object shot. The output is a good reminder that short text with a simple hierarchy is easier than a dense paragraph. It is appropriate for concept art, local-event graphics, and mock signage, but not for representing a real street location.<\/p>\n<figure>\n<img loading=\"lazy\" decoding=\"async\" width=\"848\" height=\"1264\" class=\"wp-image-2457\" src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-4.png\" alt=\"Vintage metal street sign with readable OAK STREET and SAN FRANCISCO text\" srcset=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-4.png 848w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-4-342x510.webp 342w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-4-604x900.webp 604w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-4-768x1145.webp 768w\" sizes=\"auto, (max-width: 848px) 100vw, 848px\" \/><figcaption>Prompt: Photo of a vintage metal street sign mounted on a red brick wall, crisp bold typography, text must read OAK STREET on the top line and SAN FRANCISCO on the bottom line, shallow depth of field, natural daylight, photorealistic<\/figcaption><\/figure>\n<h3>5. Crowded night market<\/h3>\n<p><strong>Prompt:<\/strong> Wide cinematic photo of a crowded open air night market food stall, steam rising from noodle pots, glowing paper lanterns, hands serving bowls, realistic faces, deep background, 35mm lens look, photorealistic<\/p>\n<p>The output gets the scene&#8217;s central story right: a wooden stall, noodle preparation, strong steam, lanterns, bowls, and a close crowd. Its composition is more symmetrical and portrait-like than the requested wide 35mm photograph. Faces and hands hold together at this display size, but small background signage is too soft to treat as meaningful text. This is the kind of prompt where ERNIE Image Turbo is useful for mood boards and hero art, then less suitable when exact spatial blocking or close inspection of every background person is required.<\/p>\n<figure>\n<img loading=\"lazy\" decoding=\"async\" width=\"848\" height=\"1264\" class=\"wp-image-2458\" src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-5.png\" alt=\"Crowded night market food stall with noodles, steam, lanterns, and diners\" srcset=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-5.png 848w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-5-342x510.webp 342w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-5-604x900.webp 604w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-5-768x1145.webp 768w\" sizes=\"auto, (max-width: 848px) 100vw, 848px\" \/><figcaption>Prompt: Wide cinematic photo of a crowded open air night market food stall, steam rising from noodle pots, glowing paper lanterns, hands serving bowls, realistic faces, deep background, 35mm lens look, photorealistic<\/figcaption><\/figure>\n<h3>6. Clean printed text<\/h3>\n<p><strong>Prompt:<\/strong> Flat lay photo of a minimalist notebook and pen on a warm wooden desk, on the notebook cover printed text must read WEEKLY PLAN in bold sans serif letters, clean spacing, natural window light, subtle shadows, photorealistic<\/p>\n<p>The final image delivers exactly the requested words, WEEKLY PLAN, in clean black type. The notebook, white pen, window light, and warm wood also arrive without visual clutter. This is the most production-ready image in the set because the brief is narrow and the text is short. It is a good fit for a planning mockup, thumbnail, or simple campaign visual. For a final brand asset, replace the generated lettering with approved design text if typography must match a house font.<\/p>\n<figure>\n<img loading=\"lazy\" decoding=\"async\" width=\"848\" height=\"1264\" class=\"wp-image-2459\" src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-6.png\" alt=\"Minimal notebook on a wooden desk with readable WEEKLY PLAN text\" srcset=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-6.png 848w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-6-342x510.webp 342w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-6-604x900.webp 604w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/ernie-image-turbo-test-6-768x1145.webp 768w\" sizes=\"auto, (max-width: 848px) 100vw, 848px\" \/><figcaption>Prompt: Flat lay photo of a minimalist notebook and pen on a warm wooden desk, on the notebook cover printed text must read WEEKLY PLAN in bold sans serif letters, clean spacing, natural window light, subtle shadows, photorealistic<\/figcaption><\/figure>\n<h2 id=\"speed-cost\">Run time and cost on Wiro<\/h2>\n<p>The six original post records retain the prompts and settings, not per-run billing or elapsed time. No cost has been inferred for them. Wiro&#8217;s ERNIE Image Turbo documentation includes one completed-task example with <strong>6.0000 elapsed seconds<\/strong>. That is an example task, not a timing guarantee for these six images. It also shows <strong>$0.003510000000<\/strong> as the total cost on a separate cancelled-task example. Because that value belongs to a cancelled task and does not state an output count, it is not presented as a cost per image.<\/p>\n<p>For a current production estimate, run the intended prompt and check that task&#8217;s own detail record. Resolution, queue time, prompt enhancement, and future service pricing can change the result. The useful fact from this test is simpler: the model is designed around eight inference steps, so it is a sensible option when iteration speed matters more than exhaustive refinement.<\/p>\n<h2 id=\"pick\">When to pick ERNIE Image Turbo<\/h2>\n<p>Pick ERNIE Image Turbo for fast single-image briefs that combine a clear subject, a visual style, and a short piece of text. The earbud, neon sign, street sign, and notebook show why: it kept the main object coherent and rendered the required words correctly. It also handled atmosphere well in the rainy street and market tests.<\/p>\n<p>Use a slower or more controlled workflow when the job depends on exact factual anatomy, verified location details, long blocks of copy, a strict wide composition, or inspection of many small people and objects. The market image shows the boundary. It looks cohesive, but it does not guarantee every distant detail. For a repeatable production job, keep the seed, resolution, steps, scale, and prompt-enhancer choice fixed while changing one instruction at a time.<\/p>\n<h2>Further reading and related tests<\/h2>\n<ul>\n<li><a href=\"https:\/\/ernie.baidu.com\/blog\/posts\/ernie-image\/\" target=\"_blank\" rel=\"noopener\">https:\/\/ernie.baidu.com\/blog\/posts\/ernie-image\/<\/a><\/li>\n<li><a href=\"https:\/\/huggingface.co\/baidu\/ERNIE-Image-Turbo\" target=\"_blank\" rel=\"noopener\">https:\/\/huggingface.co\/baidu\/ERNIE-Image-Turbo<\/a><\/li>\n<li><a href=\"https:\/\/arxiv.org\/html\/2605.25347v1\" target=\"_blank\" rel=\"noopener\">https:\/\/arxiv.org\/html\/2605.25347v1<\/a><\/li>\n<li><a href=\"https:\/\/wiro.ai\/blog\/ernie-image-turbo-5-prompt-tests-for-clean-layouts\/\">ERNIE-Image-Turbo: 5 Prompt Tests for Clean Layouts<\/a><\/li>\n<li><a href=\"https:\/\/wiro.ai\/blog\/ernie-image-review-layout-tests\/\">ERNIE-Image Review: 3 Smart Layout Tests<\/a><\/li>\n<li><a href=\"https:\/\/wiro.ai\/blog\/openai-gpt-image-2-6-real-image-tests\/\">OpenAI GPT Image 2: 6 Real Image Tests<\/a><\/li>\n<\/ul>\n<h2>Try the model<\/h2>\n<p><a href=\"https:\/\/wiro.ai\/models\/baidu\/ernie-image-turbo\">Run baidu\/ernie-image-turbo on Wiro<\/a> and start with a compact brief. Name the subject, set the scene and lighting, put required short text in quotation marks, and treat the first output as a testable draft rather than final proof.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>ERNIE Image Turbo prompt tests put Baidu&#8217;s fast text-to-image model through six deliberately different jobs: product photography, a rainy street with exact&hellip;<\/p>\n","protected":false},"author":4,"featured_media":2460,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[52],"tags":[208,209,81],"class_list":["post-2461","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-model-reviews","tag-baidu","tag-ernie","tag-text-to-image"],"_links":{"self":[{"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/posts\/2461","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/comments?post=2461"}],"version-history":[{"count":5,"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/posts\/2461\/revisions"}],"predecessor-version":[{"id":4304,"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/posts\/2461\/revisions\/4304"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/media\/2460"}],"wp:attachment":[{"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/media?parent=2461"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/categories?post=2461"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/tags?post=2461"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}