Realistic Vision Prompts: 12 Copy-Ready Photo Recipes
Twelve photoreal prompts for Realistic Vision V6.0 B1, each with a matching negative prompt and starting settings, plus the creator's recommended sizes and an inpainting step for the last fix.
By Eftal Yurtseven · Nov 17, 2026 · 9 min read

Most photoreal prompts fail in the same boring way. The face is fine, the light is fine, and the whole thing still looks like a render, because the prompt described a subject and forgot to describe a photograph. Good realistic vision prompts talk like a photographer: which lens, which light, how much skin, what's out of focus, and what the frame should never contain.
Here are 12 copy-ready prompts for Realistic Vision V6.0 B1 on each::labs, covering portraits, product, food, street, interiors and editorial, each with a negative prompt and starting settings. For the general overview of the model, our Realistic Vision image generation guide covers it. This post is the prompt drawer.
What Realistic Vision V6.0 B1 Expects From You
Realistic Vision is a Stable Diffusion fine-tune by the independent creator SG161222. Its model card calls V6.0 "New Vision", says it ships in beta versions ahead of a full release, and states the aim plainly: realism and photorealism. B1 is the first beta. That lineage matters. This isn't a chat-style model that reads a paragraph and reasons about intent. It responds to dense, comma-separated descriptors, and most strongly to the vocabulary of real photographs.
So write tags, not stories. The default prompt on Realistic Vision V3 Inpainting shows the pattern: "RAW photo, a portrait photo of ... in casual clothes, natural skin, 8k uhd, high quality, film grain, Fujifilm XT3". Subject in the middle, photographic evidence on both sides. Every recipe below follows that shape.

The Settings the Creator Recommends, Mapped to each::labs
SG161222's model card for V6.0 B1 is short and specific. It recommends the DPM++ SDE Karras sampler with 25 or more steps, or DPM++ 2M SDE with 50 or more. It lists resolutions by framing: 896x896 for a face portrait, 896x896 or 768x1024 for a portrait, 768x1024 or 640x1152 for half body, and 896x896, 768x1024, 640x1152, 1024x768 or 1152x640 for full body. And one line people skip: use a lower resolution if you get a lot of mutations.
On each::labs the request schema gives you width and height (up to 1920 each), num_inference_steps (default 20, up to 100), guidance_scale (default 7.5, up to 10), seed, use_karras_sigmas and a scheduler from DDIM, DPMSolverMultistep, HeunDiscrete, K_EULER_ANCESTRAL (the default), K_EULER and PNDM. There's no SDE sampler in that list, so we start from the closest family: DPMSolverMultistep with use_karras_sigmas on, 30 steps, which clears the card's 25-step bar.
Change two defaults. The size defaults to 512x728, which isn't on the card's list, so set it yourself every time. Guidance at 7.5 works, but we start at 5 to keep skin from going crunchy; that's our habit, not the card's rule. The default negative prompt is exactly the one the creator recommends. Keep it for anything with a person in it:
(deformed iris, deformed pupils, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime), text, cropped, out of frame, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neckBelow, "card negative" means that block, and "base" means DPMSolverMultistep, use_karras_sigmas true, 30 steps, guidance 5.

The Building Blocks of Realistic Vision Prompts
Think in five layers. A capture word opens the prompt: "RAW photo", "candid photo", "editorial photograph". Then framing: "head and shoulders", "half body", "full body", "overhead". Framing and canvas size have to agree, which is exactly why the card ties resolutions to framings.
Next, the camera. Lens lengths carry meaning: 85mm reads as a flattering portrait, 35mm as documentary and street, 24mm as interiors, 100mm macro as product detail. Aperture words and camera bodies ("f/1.8", "Fujifilm XT3") add texture more than accuracy, but they pull the result toward photo and away from illustration.
Then light, the layer most prompts underwrite. Name a source and a quality: "soft window light from the left", "overcast daylight", "hard light with crisp shadows". Last comes surface truth, the details cameras record and renders forget: "visible pores", "flyaway hairs", "scuffed wood", "film grain". Two or three per prompt is plenty. Ten turns skin into sandpaper.
Portrait Prompts You Can Copy
Portraits are why most people pick Realistic Vision, and they're where the card negative earns its length. Watch the size change with the framing.
1. Weathered fisherman
Prompt: RAW photo, head and shoulders portrait of a 60 year old fisherman with a grey beard, weathered skin, visible pores, wool beanie, harbor at dawn behind him, soft side light from the left, 85mm lens, shallow depth of field, film grain, Fujifilm XT3
Negative: card negative
Settings: 896x896, base, seed 422. Reader by the window
Prompt: candid photo of a young woman with freckles reading a paperback in a cafe window seat, oversized knit sweater, soft overcast window light, natural skin texture, flyaway hairs, 50mm lens, f/1.8, muted film colors
Negative: card negative
Settings: 768x1024, base3. Chef at the pass
Prompt: RAW photo, half body portrait of a chef in a white jacket plating a dish at a steel kitchen pass, focused expression, warm tungsten kitchen lights, steam in the background, 35mm lens, documentary style, natural skin, film grain
Negative: card negative
Settings: 640x1152, base, guidance 6
Product and Food Prompts
Objects need a different negative. Anatomy terms are dead weight here, and the real risk is invented text on labels and packaging. Aim the negative at that.
4. Morning mug
Prompt: RAW photo, product shot of a ceramic coffee mug with a thin wisp of steam, on a walnut table, morning sunlight through a window, soft shadows, 50mm lens, f/2.8, shallow depth of field, high detail
Negative: text, logo, watermark, cgi, 3d, render, cartoon, worst quality, low quality, jpeg artifacts, blurry, duplicate, deformed
Settings: 1024x768, base, seed 75. Glass bottle on stone
Prompt: studio product photograph of an unlabeled amber glass bottle on a slab of rough travertine, single hard light from the right, crisp shadow, dark seamless background, 100mm macro lens, sharp focus, reflections on glass
Negative: text, label, logo, watermark, cgi, 3d, render, plastic, worst quality, low quality, blurry, duplicate
Settings: 896x896, base6. Leather boots, lifestyle
Prompt: RAW photo, pair of worn brown leather boots on a wooden porch step, scuffed toes, loose laces, late afternoon side light, dust in the air, 50mm lens, shallow depth of field, film grain
Negative: text, logo, brand name, cgi, 3d, render, cartoon, low quality, blurry, duplicate, deformed
Settings: 1024x768, base7. Ramen from above
Prompt: overhead food photograph of a bowl of ramen with soft boiled egg, sliced pork, scallions and nori, dark ceramic bowl on a slate table, chopsticks on the rim, soft diffused daylight, high detail
Negative: text, logo, watermark, cgi, 3d, render, cartoon, plastic food, worst quality, low quality, blurry, duplicate
Settings: 896x896, base8. Bakery counter
Prompt: RAW photo, sourdough loaf with a crackled crust on a flour dusted wooden counter, bakery interior out of focus behind, warm morning light, 85mm lens, f/2, crumbs, flour particles in the air
Negative: text, logo, signage, cgi, 3d, render, cartoon, worst quality, low quality, blurry, duplicate
Settings: 1024x768, baseStreet, Interior and Editorial Prompts
Wide scenes are where the card's resolution warning bites. 1152x640 lets a street breathe, but more canvas means more chances for doubled limbs in background figures. Keep crowds small, or drop to 1024x768 when duplicates show up.
9. Rainy alley at night
Prompt: RAW photo, narrow city alley at night after rain, wet cobblestones reflecting warm shop lights, a lone person with a black umbrella walking away from camera, light mist, 35mm lens, film grain, cinematic color
Negative: text, letters, signage, logo, cgi, 3d, render, cartoon, worst quality, low quality, blurry, duplicate, deformed
Settings: 1152x640, base10. Market vendor
Prompt: candid street photograph of an elderly fruit vendor arranging oranges at an outdoor market stall, striped awning shade, dappled sunlight, 35mm lens, documentary style, natural skin, film grain
Negative: card negative
Settings: 1024x768, base11. Scandinavian living room
Prompt: interior photograph of a scandinavian living room, light oak floor, linen sofa, wool throw, large window with soft morning light, plants, 24mm lens, eye level, natural colors
Negative: text, logo, watermark, cgi, 3d render, illustration, distorted perspective, worst quality, low quality, blurry, duplicate
Settings: 1152x640, base12. Editorial on a bridge
Prompt: fashion editorial photograph, full body, a woman in a long camel wool coat and white sneakers walking across an old stone bridge, overcast soft light, wind in her hair, 50mm lens, muted color grade, natural skin texture
Negative: card negative
Settings: 640x1152, base
What We Got When We Ran Them
We ran three generations on each::labs with the base settings. First, the fisherman written as "close-up portrait" at 768x1024, seed 42. Skin, beard and knitwear were convincing. The framing wasn't: the tall canvas gave us a waist-up shot with a hand in frame, and "side light" became a backlit sunset. Recipe 1 is the fix we tested next, "head and shoulders" at 896x896, same seed. That one landed as a real portrait with deep wrinkles and pores, though the harbor drifted into a lake with mountains and the light stayed flat.
The mug, at 1024x768 and seed 7, read as a photograph: believable ceramic, warm window light, sunlit wood grain. The weak spot was steam. We wrote "steam rising" and got a thick cloud across the window, which is why recipe 4 asks for "a thin wisp". The lesson from all three runs: the model takes atmosphere words literally and generously, so ask for less than you want.
Mistakes That Make Photos Look Generated
Leaving the size at the default is the big one. 512x728 matches nothing on the card, and a prompt written for a square headshot drifts when the canvas disagrees with it. Pick the size from the framing.
Stacking "masterpiece, best quality, ultra detailed, 8k, hyperrealistic" on every prompt is next. A couple of quality tags are fine; a wall of them tends to push skin toward plastic, the exact look you're escaping. Spend those words on light and lens.
Then there's the copied negative. "Extra limbs" does nothing for a bottle, while "text, label, logo" is what stops gibberish packaging. And don't crank guidance to the maximum of 10 to make the model listen. Move the important tag earlier in the prompt first; order does more than force.
Fix the Last 10 Percent With Inpainting
Sometimes a frame is 90 percent there and one hand or one patch of background lets it down. Regenerating throws away everything that worked. Realistic Vision V3 Inpainting on each::labs takes your image plus a mask (both required) and repaints only the masked area. Its schema offers prompt, negative_prompt, strength (default 0.8, up to 1), steps (default 20, up to 100), num_outputs (up to 5) and a guidance field spelled guaidance_scale in the API (default 7.5, up to 20). For a hand, mask it with a little margin and prompt only what belongs there:
Prompt: RAW photo, relaxed hand resting on a knee, natural skin, five fingers, soft side light, film grain
Negative: card negative
Settings: strength 0.6, steps 30, num_outputs 4Lower strength keeps more original pixels, so the patch matches the surrounding grain and light; raise it when the area is badly wrong. The V6.0 B1 schema also accepts image, mask and strength, so you can stay on the same model for a fix. Once the still is done, hand it to an image-to-video model or chain the steps in an each::labs flow. For more portrait technique, read how to create realistic AI portraits, and browse the text-to-image models or the full model catalog when a brief needs a different look.
Frequently Asked Questions
What makes good realistic vision prompts?
Photographic evidence. Open with a capture word like "RAW photo", state the framing, name a lens and a light source, and add two or three surface details such as pores or grain. Then match the canvas size to the framing. The subject is usually the easy part.
What negative prompt does Realistic Vision V6.0 B1 use?
The creator's card recommends a long one that opens with "(deformed iris, deformed pupils, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime)" and runs through anatomy and quality terms. On each::labs it's already the default negative_prompt, so you only edit it for objects, food or rooms.
Which resolution should I use for Realistic Vision V6?
Go by framing. The card lists 896x896 for face portraits, 768x1024 or 640x1152 for half body, and adds 1024x768 and 1152x640 for full body; we use those landscape sizes for streets and rooms too. If limbs start duplicating, step down.
Can I fix hands or faces without regenerating the whole image?
Yes, with a mask. Realistic Vision V3 Inpainting repaints only the masked area and can return up to five versions at once, so you pick the best fix while the rest of the photo stays untouched.



