Photos That Look Taken
A laugh caught mid-moment, a wet apron, market light from a skylight. People look like people, not like renders.
MiniMax H3 Image Prompting Guide
MiniMax H3 Image makes photos that look like someone really took them. It follows long, detailed prompts closely, sets quoted text exactly and renders up to 16 megapixels. It also edits a photo or combines up to 8 images. Every image on this page was made on PixelDojo with the prompt shown.
This model rewards detail. Your prompt goes to MiniMax H3 Image exactly as you wrote it, so the more you describe, the more you get. A short prompt works, but a full paragraph gets you the picture in your head.
Write it like you are briefing a photographer. Say what kind of image it is, who or what is in it, where each thing sits, and where the light comes from. The model places things where you say.
For text, put every word in quotes. Quoted words come out spelled as typed. Text you only describe, like "a short line under each step", turns into made-up lettering.
16 MP
Sharpest size (4K)
8
Images per edit
1.5
Credits at 2K
A laugh caught mid-moment, a wet apron, market light from a skylight. People look like people, not like renders.
A compass on the top left corner, a cup on the bottom right, three pencils in a set color order. Each one landed where the prompt put it.
Three lines on a two-color poster, each spelled as typed and placed where the prompt said. Quote the words and name the spot.
The apron turned yellow and snow came through the skylight. His face, the fish, the scale and the counter stayed the same.
Call it a frame from a quiet 1990s independent film and you get the grain, the faded color and the tube light.
A square image at 4K is 4096 x 4096. Labels, fabric and fine print hold up when you crop or print large.
Each example shows the exact prompt that produced the result. Copy any prompt with one click.
4K · 3:4 · 4 credits
A candid photograph of a fishmonger in his fifties with a salt-and-pepper beard and a navy rubber apron, laughing as he weighs a silver sea bass on a brass scale. Crushed ice and small handwritten price tags fill the counter in front of him. Cold morning light falls from a market skylight above. Shot on a 50mm lens, natural skin texture, slight motion in his hands.
"Candid photograph" plus one action in progress keeps the person unposed. Name the light source and a lens so it stays a photo.
4K · 16:9 · 4 credits
A photograph of an architect's desk seen from directly above. A large blueprint is unrolled across the center, held flat by a brass compass on its top left corner and a white coffee cup on its bottom right corner. A wooden scale ruler lies diagonally across the middle. To the left of the blueprint sits a closed black notebook with a red elastic band. To the right, three sharpened pencils lie side by side: one yellow, one blue, one green, in that order from top to bottom. Soft daylight from the left, pale oak desk surface.
Give each object its own sentence and its own spot: top left corner, to the right, in that order. Long prompts are fine here.
4K · 2:3 · 4 credits
A risograph-printed poster for an independent bookshop in teal and coral ink on off-white paper. At the top, large bold sans-serif letters read "THE LANTERN ROOM". Under it, smaller text reads "Used Books, Maps and Records". In the center, a flat illustration of a tall bookshelf with a glowing paper lantern hanging in front of it and a sleeping cat on the lowest shelf. At the bottom, one line reads "Open late every Friday, 14 Harbor Lane". Visible grain and slight misregistration between the two inks. No other text.
Quote every line, say where it sits and end with "No other text". Naming the print method and the two inks sets the whole look.
4K · 4:3 · 4 credits
A close-up photograph of a handwritten recipe card lying on a flour-dusted wooden counter. The card is cream colored with a thin red border. In neat blue ballpoint handwriting it reads: "Nonna's Lemon Cake" as the title, then four lines: "3 eggs", "200 g sugar", "2 lemons, zest and juice", "Bake 40 minutes at 170". A halved lemon and a wooden spoon sit beside the card. Warm window light from the right, shallow depth of field.
Handwriting works the same way as print: quote each line. Say the pen and the paper so the writing looks like it belongs there.
4K · 4:3 · 4 credits
A studio product photograph of a matte olive green enamel kettle with a walnut handle on a pale limestone block. A small cream label on its side reads "FIELD & FLAME" in dark green capital letters. A single sprig of dried eucalyptus lies beside it. Soft diffused light from the upper left, a gentle shadow to the right, a plain warm grey backdrop.
Product, material, surface, light, backdrop. Describe the label as an object (small, cream, on its side) and quote what it says.
4K · 3:4 · 4 credits
A clean educational infographic titled "The Water Cycle" in large dark blue letters at the top. Four labeled stages are arranged in a clockwise loop joined by curved arrows: "1 Evaporation" over a blue sea with rising wavy lines, "2 Condensation" beside a white cloud, "3 Precipitation" under a grey cloud with rain, "4 Collection" beside a river running back to the sea. Flat vector style, a palette of blues, soft green and sand, a mountain in the center, generous white space. No other text.
Write out every label and say what sits next to it. Leave nothing as "a short caption", or the model fills it with fake words.
4K · 16:9 · 4 credits
A frame from a quiet 1990s independent film: a woman in her thirties in a mustard cardigan sits alone in a laundromat at night, watching a row of spinning dryers. Fluorescent tube light overhead, green linoleum floor, a red plastic laundry basket at her feet, rain on the front window with a blurred neon sign outside. 35mm film grain, slightly faded color.
Decade and kind of film first, then one person doing one quiet thing. A few props with colors make the frame feel art-directed.
4K · 16:9 · 4 credits
A hand-painted gouache illustration of a lighthouse keeper in a yellow raincoat rowing a small red boat toward a striped lighthouse at dusk. Two seagulls fly above the boat. A black cat sits at the bow facing forward. The sea is deep teal with white brush-stroke waves and the sky fades from peach to violet. Visible brush strokes, flat shapes, the style of a mid-century picture book.
Name the paint and the kind of book. Counts hold up too: two gulls and one cat, each where the prompt put them.
2K · Auto · 1 image · 1.5 credits
Change his navy rubber apron to a bright yellow one and make it snow lightly through the skylight above him. Keep his face, the fish, the scale and the counter exactly the same.
Say the change, then list what must stay the same. Leave the aspect ratio on Auto to keep the shape of your photo.
Your words go to the model as written. Four or five plain sentences beat one line with style tags.
A candid photograph, a frame from a 1990s film, a risograph poster, a gouache illustration. The first line sets the look.
Top left corner, to the right of the book, at the bottom. One sentence per object keeps a busy scene in order.
Quoted words come out spelled right. Described text, like "a caption under it", comes out as fake lettering.
It stops the model from adding extra signs, logos and small print around your words.
Say where it comes from and what kind it is: a skylight above, a window on the right, a tube light overhead.
Stunning, epic, masterpiece and 8k push the picture toward a glossy look. Plain description keeps it real.
A 2K image takes about 25 seconds. When the idea is right, run it again at 4K for the sharpest version.
1. What the image is: the kind of photo, film frame, print or painting.
2. The subject: one person or object, described so you could pick it out, doing one thing.
3. Everything else in the frame: each object with its color and its place.
4. The place and the light: where it is, where the light comes from, the time of day.
5. The words, if any: each in quotes with its spot, then "No other text".
Attach up to 8 images. Picture 1 is the image that gets edited. The others are references the prompt can name as picture 2, picture 3 and so on, for example "give the woman in picture 1 the hat from picture 2".
Say the change in one or two sentences, then say what must stay the same. "Keep his face, the fish and the counter exactly the same" is what protects the rest of the photo.
Leave the aspect ratio on Auto to keep the shape of picture 1. The size you pick sets how many pixels the result has.
| Setting | Values | Notes |
|---|---|---|
| Prompt | string (required) | What to create, or the edit to make to picture 1 when images are attached. Sent to the model as written. |
| Resolution | 2K · 4K | 1.5 credits per image at 2K (4 megapixels), 4 credits at 4K (16 megapixels, the sharpest). |
| Aspect ratio | Auto · 1:1 · 4:3 · 3:4 · 3:2 · 2:3 · 16:9 · 9:16 | Auto keeps the shape of picture 1, else 1:1. At 4K, 1:1 is 4096 x 4096 and 16:9 is 5472 x 3072. |
| Images | up to 8 (optional) | Picture 1 is edited, the rest are references. |
| Number of images | 1-4 | Each image is billed at its resolution. |
| Prompt upsampling | off · on (API) | Off by default. Turn it on to have a short prompt expanded for you. It adds about 90 seconds. |
MiniMax H3 Image is an image generation and editing model from MiniMax. It makes lifelike photos, film stills, illustrations and designs with exact quoted text up to 16 megapixels, and edits with up to 8 input images.
Photos that look like real photographs, and scenes with many exact details. It places objects where the prompt says and keeps people natural and unposed.
Put every word in quotes and say where it goes. Finish with "No other text". Text that is only described, without the exact words, comes out as made-up lettering.
Attach the photo as picture 1, say the change, then say what must stay the same. To use more images, attach them after it and name them in the prompt as picture 2, picture 3 and so on.
It is included in every PixelDojo plan, from $10/month: 1.5 credits per image at 2K and 4 credits at 4K.
Both come from the same model family. MiniMax H3 Image gives the more natural photo and follows small details more closely. P-H3-Image is faster and costs less, with a 1K size and unlimited Relax runs.
Yes. Call minimax-h3-image at /api/v1/models/minimax-h3-image/run with prompt, aspect_ratio, resolution and reference_images. It is also in Canvas, Character Studio and our MCP server.