블로그/prompts·2026년 11월 21일·8분·eroq 팀 작성
Krea2 prompt guide — clean, fast photoreal for characters and everyday shots
How to prompt Krea2 on eroq: the caption structure, light and lens words, Prompt and Reference strength, two characters in one picture, and 10 example prompts.
Krea2 is eroq's everyday photoreal engine: clean skin, natural light and believable places, rendered in about fifteen seconds a take, with up to four reference pictures to hold a face. The rule that matters most is to name the light and the lens: "soft window light, 85mm" steers Krea2 further than any stack of adjectives.
It is the engine most character work starts on, because it is quick enough to iterate in batches of four and cheap enough to throw takes away. This guide covers the caption structure that works, the vocabulary it reads, its two dials, two-character pictures, Edit, ten example prompts and the usual mistakes.
What Krea2 is good at, and what it is not
Krea2 is at its best on ordinary, well-lit life:
- Portraits and character renders: a face that reads as a real person, with pores and fly-away hair rather than plastic skin.
- Lifestyle scenes: kitchens, cafés, streets, offices, a gym at 6 a.m.
- Simple product shots: an object on a surface in good light.
- Two people in one picture: @mention two saved Characters and both faces hold.
- Speed: four takes come back in about a minute, so you can explore framing before you commit.
Reach for another engine when the job is something else. Dark or intense fiction that a clean engine would soften goes to Eroq Uncensored, which has the same controls. Exact instruction edits and more than four references go to Eroq One. Words on a sign or a label go to an engine that sets lettering, such as Flash Image. Anime goes to Anima.
Prompt anatomy: a caption a photographer would write
Krea2 responds best to the kind of caption that would sit under a published photo, in one paragraph:
- Framing: headshot, half-body, full-length, over-the-shoulder, top-down.
- Who: a person with an age and two or three visible traits (hair, build, expression).
- Wearing: one outfit, specific ("a rust corduroy jacket over a white tee").
- Doing: one action, mid-moment.
- Where: a place with two concrete details.
- Light: one lighting idea.
- Lens and finish: a focal length or a camera, then the grade.
As a template line:
[framing] of [who, age, traits], wearing [outfit], [action], in [place, two details], [light], [lens], [finish]
Keep it between thirty and sixty words. Krea2 is fast precisely because it does not need a novel.
The vocabulary it reads well
| Layer | Words that work |
|---|---|
| Light | soft window light, overcast daylight, open shade, golden hour backlight, practical lamps, on-camera flash, blue hour, neon spill |
| Lens | 24mm environmental portrait, 35mm street, 50mm, 85mm portrait, 100mm macro, shallow depth of field, f/1.8 |
| Framing | headshot, half-body, full-length, low angle, eye level, over-the-shoulder, top-down flat lay |
| Finish | natural color, muted tones, film grain, high-contrast black and white, warm grade |
Light is the strongest lever. "Overcast daylight" flattens and softens; "golden hour backlight" gives a rim on the hair and a warm haze; "on-camera flash" gives the hard, glossy party-photo look. One of them per prompt.
The two dials
- Prompt strength (CFG, 0 to 3, Loose to Strict, default 1): how literally your words are followed. One step up when a detail keeps going missing; one step down when the picture looks harsh or overworked.
- Reference strength (Auto by default, from Hold reference to Push prompt): shown once pictures are attached. Toward Hold reference, Krea2 copies the reference closely, outfit and room included. Toward Push prompt, your words win and the reference is barely read. Start on Auto; slide toward Push prompt when renders copy the reference's clothes or background instead of following your prompt, and back toward Hold reference when the face starts to drift.
There is no negative prompt on Krea2: describe what you want in frame.
Two characters in one picture
This is where Krea2 earns its place. @mention one or two saved Characters, attach nothing else, and Krea2 renders straight from their identity sheets: both faces hold in the same picture, and Reference strength stays on Auto because the sheets already carry the identity. Each person keeps their own outfit from their own sheet unless your prompt dresses them, in which case your wording wins.
Three or more people, or a mix of Characters and other pictures, still work, but faces are then held more loosely. For a crowded cast, build the picture on Eroq One. More on identity sheets: look sheets.
Ten example prompts
1. Café window — the window does the lighting work.
Half-body portrait of a woman in her late twenties with a cropped auburn bob, wearing a rust corduroy jacket over a white tee, laughing at something off-camera, at a café window table with a half-finished flat white and a paperback, soft window light, 85mm, natural color
Settings: 3:4, four takes.
2. First run of the day — backlight separates the subject from the city.
Full-length shot of a man in his thirties in a grey running jacket and black tights, mid-stride on an empty riverside path at sunrise, breath visible in the cold, bridges in the haze behind, golden hour backlight, 50mm, light film grain
Settings: 16:9, four takes.
3. Sunday kitchen — practical lamps make a domestic scene feel lived in.
Medium shot of a grandmother in her seventies in a floral apron rolling pasta dough on a flour-covered wooden table, a pot steaming on the stove behind her, copper pans on the wall, warm practical lamps and late afternoon light, 35mm, warm grade
Settings: 4:3, two takes.
4. Team page headshot — the plain version every brand needs.
Headshot of a man in his forties with a short grey beard, navy crew-neck sweater, relaxed half-smile, against a pale warm-grey wall, overcast daylight from a large window, 85mm, f/2, natural color
Settings: 1:1, four takes, Prompt strength 1.
5. Two friends, two sheets — the duo render at work.
@Mara and @Theo sitting on the edge of a rooftop at dusk, sharing a bag of takeaway noodles, city lights coming on behind them, Mara pointing at something below, blue hour, 35mm, soft film grain
Settings: 16:9, four takes, Characters only, Reference strength on Auto.
6. Rehearsal room — mixed light sells the location.
Half-body shot of a bass player in her thirties with braided hair and a vintage band tee, eyes closed mid-song, in a cramped rehearsal room lined with egg-crate foam and tangled cables, a red practical lamp and a cool fluorescent tube, 35mm, grainy
Settings: 4:3, four takes.
7. The allotment — open shade keeps faces soft outdoors.
A gardener in his sixties in a waxed jacket and muddy boots kneeling between rows of leeks, holding up a bunch of carrots with soil still on them, a wooden shed and a water butt behind, open shade on an overcast morning, 50mm, muted tones
Settings: 3:4, two takes.
8. Ceramic mug — a simple product shot, light named precisely.
Close-up of a handmade speckled ceramic mug of black coffee on a marble counter, a thin curl of steam, a linen napkin folded beside it, soft side light from a window on the left, 100mm macro, shallow depth of field, warm grade
Settings: 1:1, four takes.
9. Bus stop, raining — flash plus rain gives an editorial snap.
Full-length shot of a woman in her twenties in an oversized yellow raincoat and white sneakers waiting at a bus stop in heavy rain, puddles reflecting shop signs, on-camera flash freezing the raindrops, night, 28mm
Settings: 9:16, four takes.
10. Market morning — a travel photo with a clear subject.
A fishmonger in his fifties in a striped apron holding up a whole sea bream at a crowded morning market, crates of ice and lemons, hanging bulbs over the stalls, customers soft in the background, overcast daylight, 35mm street photography, natural color
Settings: 4:3, two takes.
Editing with Krea2
Switch to the Edit tab, put the picture to change in as Picture 1 (plus one subject picture if the change needs one), and write what should change. Krea2 keeps the source in view and re-renders the scene toward your instruction; Format sets the output frame, and the editor repaints your picture at that ratio. The change lands, but small details such as a hair parting or a background pattern can move. When every other pixel must stay put, use Eroq One, whose editor is built for exactly that. The step-by-step is in edit with reference pictures.
Mistakes to avoid
- Adjective stacks. "Beautiful, stunning, ultra-detailed, 8K masterpiece" does nothing a light and a lens would not do better.
- Two lights, two lenses. "Golden hour, studio flash, 24mm and 85mm" makes Krea2 average them. One of each.
- Negations. There is no negative prompt; "no people in the background" invites people. Write "an empty street".
- Outfits copied from the reference. That is Reference strength at work. Slide it toward Push prompt, or describe the new outfit in detail.
- Crowds of Characters. Two hold best; past that, consider Eroq One.
- Text. Signs and labels come out garbled; leave them to an engine that sets lettering.
Settings on eroq
- Price: 10 credits / image at every format; four takes cost 40 credits. A failed render refunds itself.
- Formats: 1:1, 3:4, 4:3, 9:16 and 16:9.
- Takes: one to four per run, about fifteen seconds each once the engine is warm.
- References: up to four pictures, or your Characters and Elements.
- Access: every account, pay-as-you-go included.
- Privacy: eroq runs Krea2 itself and never trains on your prompts. The Private mode switch at the bottom of Settings keeps the render in your browser only, with no Library entry and your prompt shown as stars in the ledger.
FAQ
Is Krea2 free?
You can start free: a new account gets 40 free credits with no card, which is 4 Krea2 images. After that each image costs 10 credits / image, paid from a pack (from $10) or a plan.
What is the difference between Krea2 and Eroq Uncensored?
The controls are the same: references, Prompt and Reference strength, Edit, four takes. Krea2 is the clean everyday engine; Eroq Uncensored renders darker or more intense fiction as written, within the acceptable-use policy. Same price on both.
Can Krea2 put two of my characters in one picture?
Yes. @mention two saved Characters with nothing else attached, and both render from their identity sheets in the same picture, each in their own outfit unless your prompt says otherwise.
Why does my picture copy the reference's background?
Reference strength is holding the reference too tightly. Slide it toward Push prompt and describe the new setting in the prompt.
Can I use Krea2 through the API?
Yes: POST /v1/images/generations with the model eroq-krea2. It answers at once with one job per take, which you poll until the image is ready.
Open the Image tool and pick Krea2: /studio/image.

프라이버시
프라이빗하고, 제한 없이
- →eroq 자체 엔진엔 필터가 없어요. 대신 명문화된 정책 하나가 있어요.
- →아무것도 보관하지 않고, 작업물을 학습에 쓰지도 않아요.
- →프라이빗 모드는 파일을 전혀 남기지 않아요.