Every guide on this site

Prompt shapes for a specific model

• A Midjourney Prompt From an Image: Ten Slots Plus --ar Measures ar and arParam: gcd reduction then a cap of 60, so 1913×977 → 60:31 and 1000×999 → 60:60 rather than 1:1.
• Stable Diffusion Prompt From an Image: Where the Weights Go Measures 12 slots against the 17 weighted terms a comma parser reads; the masterpiece, best quality, prefix is 27 characters (8.0% of the long line); --to explicit round-trips 336 → 336 bytes.
• Flux Prompt From an Image: Sentences, Not a Comma List Measures sentence counts 3 / 5 / 7 at 100 / 195 / 273 characters; seven sentences become seven parser terms, and the last term swallows four sentences and 21 words; contrast and vertical lighting arrive 0 times out of 9.
• ChatGPT Image Prompt From an Image: One Sentence, Six Commas Measures sentence counts 1 / 2 / 3, six commas fixed in the first sentence, and the ten fragments a comma parser returns — fragment 10 swallows the last two sentences.
• ComfyUI Prompt From an Image: One Text Box, Two Parsers Measures 19 bracket spellings through two parsers: 13 agree, 6 do not; the generated line carries 0 bracket characters and reaches the node as 1 chunk; 18 of 65 shipped nodes carry a weight field.
• Nano Banana Prompt From an Image: Twelve Mismatches Measures 12 of 13 dimensions misaligned; 13 of 2,188 computed ratios land on the 14 accepted buckets; 21:9 is always written 7:3; 8 framing phrases against 14 buckets.
• Nano Banana Pro Prompt From an Image: Three Names You Type Measures 12 outputs: model named 0 times, size 0 times, shape 3 times; the hand-written part is 127–128 characters, 26.8%–37.2% of the request; every published 1K frame is 27.75–28.44× the pixels this tool reads; 505 genuinely 21:9 sizes come out as 7:3.
• A Grok Image Prompt From an Image: Seven Fields, One Yours Measures 7 request fields with 1 carrying content; 1,053 size pairs (7.36%) whose true ratio is one of the 15 official values but whose string is not; the --ar tail is the last 10 characters of all nine outputs.
• Good ChatGPT Prompts for Pictures: One Sentence of Your Own Measures 40 photographs × 3 lengths = 120 lines: 90–145 / 172–239 / 203–269 characters; the short line is a literal prefix of the long one 40 times out of 40; the closing mood sentence is one of only 4 strings.
• Prompt ChatGPT to Make an Image: The Line It Cannot Write
• How to Prompt ChatGPT for Images: A Caption, Not an Order
• Trending Photo Prompts for ChatGPT: What 480 Lines Rank
• Image to Prompt Translation: One Line Into Another Shape
• Gemini Image Prompt From an Image: One Field of Eight
• Image to Prompt Across Models: Four Lines, One Measurement

What the tool measures, and where your image goes

• What an Image to Prompt Tool Actually Measures Measures all twelve signals on three real images: meanL 34.51 / 73.95 / 35.95, stdL 46.93 / 80.36 / 13.68, edge 22.50 / 13.02 / 5.14; the closest any value sits to a cut-off is 1.86.
• Twenty-nine cut points decide the words in a prompt Measures all 33 comparisons in describe.js: 29 cut points, 32 phrases, 10 of 32 rows within 1% of a real file, 3 within 0.5%; 18 cuts flipped by editing pixels, 14 of them within 0.005 of the cut.
• Where Your Image Goes: The Tool Sees 192×192 Pixels Measures the 192×192 grid at 36,864 cells / 147,456 bytes: 16.00% of a 640×360 file survives the resize, 10.24% of an 800×450 one; a transparent pixel counts as RGB 0,0,0.
• Image Description Generator: Nine Slots, 95 Words, One Line Measures a closed vocabulary of 95 entries (42 + 18 + 35) against 81 prompts and 1,937 words with zero misses; 65,664 of 207,360 nominal slot combinations are reachable (31.67%).
• What Is a Picture Prompt? A Line of Adjectives Measures 480 lines against 61 common object nouns: 59 appear zero times, and the only two that appear (sky, forest) are inside colour names; two different photographs produced one identical line; the correlation between file bytes and line length is −0.063 to 0.158 across the twelve cells.
• Best Image to Prompt Site or Tool? Six Checks You Can Run
• What Is a Photo Prompt? Nine Numbers and One Line
• How Much of Your Picture Survives an Image to Prompt Line
• AI Prompt Generator: What the Output Line Is Made Of
• Image to Prompt Round Trip: What the Line Brings Back
• Image to Prompt Stability: Same Bytes, Different Exports
• How an Image to Prompt Tool Names a Colour
• Image to Prompt for Video: One Field Out of Eleven
• Image to Prompt as Alt Text: 480 Lines, 0 Fit
• One Image, Four Prompts: Same Words, Different Punctuation

How to get a prompt you can actually use

• Generate a Prompt From an Image: Five Steps, In Order Measures the pipeline in milliseconds over five runs: decode 10.23–13.36, shrink to 192×192 in 2.37–2.96, measure and write in 1.99–4.05, total 16.08–17.98; then 40 files cropped 8 ways, each crop compared against the uncropped reading slot by slot — framing never moved on a centred crop (0 of 40) and always moved on the square one (40 of 40), and the median file still reads identically down to 70% of its area.
• Get a Usable AI Image Prompt From Any Picture — Three Edits Measures 480 rows of output (40 images × 4 shapes × 3 lengths) as a downloadable CSV, string lengths from 90 to 357 characters, and edge at 9.60 full-resolution against 13.02 on the 192 grid for the same photograph.
• Prompt Weights and Negative Prompts: What Each Mark Does Measures bracket depth 1 to 10 in both directions: five round brackets reach 1.61 and ten square ones fall to 0.39; --ar 16:9 parses as a weight of 9; the three sample images generate 15, 16 and 7 terms, all at weight 1.
• The Best AI Prompts for Images: The Part You Write Yourself Measures a corpus of 40 photographs: the subject slot names something 0 times out of 40, 54.4% of the shortest Flux lines is shared by all forty, four colour names give 1,256,640 ordered combinations, and a 1% brightness change moves a word in half the set.
• How Do I Prompt an Image? Five Edges, and Where Words Land Measures 800 edited lines (40 photographs × 4 shapes × 5 edits) through this site’s own weight parser: the same four-word sentence added at the front becomes the first fragment at weight 1 in 160 of 160 lines, for 19 extra characters; the same words at the end land after --ar on Midjourney only (40 of 40); and deleting the palette clause removes 38–62 characters.
• Good Image Prompt Examples: What One Line Pins Down
• How to Write a Picture Prompt: Words Between Two Commas
• Photoshoot Prompts From an Image: Five of Eight You Write
• Image to Prompt Negative: How Much of the Line You Can Name

When the output comes out wrong

• Your Image to Prompt Output Is Wrong — Fix It in This Order Measures monochrome share on three images at 0.0117 / 0.2016 / 0.0222 against the 0.85 line, and one code screenshot where 91.9% of the frame is a single colour bucket while the palette line sits third in the prompt.
• Which Images Does an Image to Prompt Tool Handle Badly? Measures failures in grid cells: navy takes 27,433 of 36,864 (74.42%), 874 cells (2.37%) are all that separate the backlit frame from a different exposure phrase, and the fourth palette name on the screenshot is 5 cells; the three images share 5 of 9, 3 of 9 and 3 of 9 slots in pairs.
• Image to Prompt by Image Type: How Much Is Already Written

The other direction: text to image

• What “Prompt to Image” Really Involves The other direction: checkpoint, sampler, step count and VRAM ceiling. The one thing measured here is this site’s own parser, on one term at rising bracket depth: 1.3 stays 1.3 through the first pair, then 1.43, 1.57 and 1.73 — four pairs put a hand-set weight out of range. Everything else is quoted from elsewhere.
• How to Convert a Prompt to an Image: What the Line Carries Measures 480 lines (40 photographs × 4 shapes × 3 lengths): 64.3%–82.3% of each line is measured off the file, 120 of 480 carry a size and 0 of 480 name anything in the picture, and the median line survives at a 128×128 grid and 7 bits per channel.