Krea 3 prompt guide: composition, style, negative prompt

TL;DR

• Use edit-as-instruction format: describe the change and what to keep, referencing images by number (image 1 is the base, images 2+ add elements)
• Combine up to 6 reference images: base scene in image 1, character in image 2, product in image 3, lighting and detail references in images 4-6
• Krea 3 has no native negative-prompt box: steer results with positive phrasing and an explicit keep list

Krea 3 launched on the Civitai generator as the newest image model from the Krea team, built for fast, reference-driven editing Civitai. You start with your own images, describe the change you want, and keep what matters — all in a single workflow that supports up to six reference images combined into one edit. This guide covers the prompt formula, composition and style keywords, camera angles, negative-prompt behavior, and the multi-image workflow. For a step-by-step walkthrough of the editor itself, see how to edit an image with a sentence in Krea 3.

How Krea 3 reads a prompt: edit-as-instruction

Krea 3 processes prompts as an edit-as-instruction: you state what to change and what to keep. The model reads your uploaded image as image 1, then reads each reference image in order and assigns it a job — a character, a product, or a lighting reference. The syntax is plain English; there is no special token syntax.

The core formula is:

[change description], keep [named elements from reference images]

Example: "Make the mug red, keep the background and the model's pose." When combining references, list each by number and role: "Put the red canoe from image 2 in the water beside the dock in image 1, its bow pointing toward the camera. Match the misty dawn light of image 1."

The formula has three parts:

1. The change — what to transform (color, object, region, lighting)
2. The keep — what must stay the same, named by reference image number
3. The context — camera angle, lighting mood, or aspect-ratio hint from image 1

The change can be as specific or as broad as you need. The keep list references image numbers and named elements. The context anchors the result to the original scene's lighting and angle. Using this pattern every time reduces the chance that the model drifts into unwanted changes — the same discipline the MiniMax H3 prompt guide applies to video.

Composition and style keywords

Krea 3 prefers plain-language description over preset tag names, but concise keywords work for common styles and compositions:

• Composition: centered, off-center, symmetrical, rule-of-thirds, wide-angle, telephoto, panoramic, minimalist
• Style: cinematic, studio-light, golden-hour, soft-box, backlit, neon, retro, watercolor, oil-painting, documentary

Place a style keyword after the change description and before the keep list, or attach it to the element you want styled: "Give the mug a ceramic-glaze finish, keep the kitchen tiles and the morning light." Avoid paragraph-length style descriptors; short, focused keywords that each modify one visual aspect work best. If you need a complex look, run the edit twice — once with a composition keyword, then a style keyword — and compare.

Camera angles and shot types

Krea 3's Change Camera Angle feature in the Edit UI reframes the same scene from a different virtual camera position without regenerating from scratch Krea Edit docs. You can get the same result from the prompt with natural-language angle descriptors:

• Low-angle shot — looking up at the subject; suggests power or epic scale
• High-angle shot — looking down; suggests weakness or vulnerability
• Eye-level shot — the default; neutral and natural
• Three-quarter angle — a straight-on portrait shifted for a more dynamic feel
• Overhead / top-down — product or flatlay scenes viewed from above
• Dutch angle — rotated horizon; tension or surreal mood

Include the angle descriptor in the context part: "Change the camera angle to a low angle, keeping the model's pose and the background lighting exactly the same." Subtle changes produce the most coherent results; large shifts can make the model reinterpret parts of the scene.

Negative prompt support

Neither the Krea documentation nor the Morphic Krea 3 guide mentions a negative-prompt field or a CFG-scale control for Krea 3. The PAA question collected on 2026-10-15 in the people-also-ask tracking repo asks: "Does Krea 2 use negative prompts?" For Krea 2, the answer is that it does not use standard negative prompts natively — it operates at a low CFG scale where traditional negative prompting is barely functional, and relies on descriptive positive language, style LoRAs, and reference images instead.

Since no source confirms a negative-prompt field for Krea 3, state it plainly: Krea 3 does not natively support a traditional negative-prompt box. The model is steered by positive phrasing and the keep-list structure.

To exclude something, describe what you want more of: instead of "remove the background", write "keep the foreground model in sharp focus." Name what to keep with image numbers and the unreferenced parts fall away. For tighter control, use the Change Region or Annotate features to select an area and prompt only that region Krea Edit docs.

Multi-image reference workflow: up to 6 reference images

Krea 3 supports up to six reference images in a single edit Civitai. The images are numbered in the order you add them, and the prompt points at them by number:

Keep each image's job distinct and named in the prompt: "Put the red canoe from image 2 in the water beside the dock in image 1. Keep the trees from image 4 and the dock rail from image 5." This prevents the model from overlapping jobs across references.

Instant vs Medium mode by prompt complexity

• Instant mode — fast, for quick iterations and single-element changes (color, simple swap, minor pose adjustment). Use it when testing multiple versions of the same edit.
• Medium mode — higher fidelity with a creativity dial, for multiple reference images, complex keep lists, and fine surface detail. Set the dial lower for exact prompt matching, higher for stylistic variation.

With up to six references and a complex keep list, start with Medium for the first run, then switch to Instant for rapid iteration. Cost on Civitai: launch week 30/60 Buzz for 1K/2K (33% off) until 22 Oct 2026, then 45/90 Buzz per image Civitai.

Worked example

Setup: a bathroom render (image 1) with a bathtub and no accessories; a bar of oat soap on a white background (image 2); a terracotta towel (image 3).

Prompt: "Put the soap dish from image 2 on the shelf in image 1 and hang the towel from image 3 beside the basin. Keep the wall, the basin, the shelf and the daylight exactly the same."

Expected output: the soap dish appears on the shelf, the towel hangs beside the basin; wall tile, fixture, shelf and daylight are preserved. Mode: Medium (three references + detailed keep list).

For placement comparisons against other models, see Krea 3 vs Midjourney and Krea 2 Turbo.

Common errors and fixes

FAQ

How do I refer to images in a Krea 3 prompt? By number, in the order you added them: image 1, image 2, and so on. Make image 1 the photo you are editing and say what each other image brings, for example: the bag from image 2.

How many images can I use in one Krea 3 edit? Up to 6. The Civitai launch article confirms up to 6 reference images per edit. Two or three is often enough: the scene plus one or two people or products. Give each image one job so the prompt stays clear.

Does Krea 3 use negative prompts? Krea 3 does not have a traditional negative-prompt field. The collected PAA question for Krea 2 answers the same: Krea 2 does not use standard negative prompts natively and relies on positive descriptive language, style LoRAs, and reference images. Steer Krea 3 with positive phrasing and a keep list.

How do I keep faces the same when combining people in Krea 3? Use clear, well-lit photos of each face and add "keep their faces and clothes" to the prompt. Say what the people are doing together so the pose fits the scene. Each person gets its own reference slot (image 2, image 3).

How do I stop Krea 3 from changing the rest of the photo? Write the change description first, then a keep list of named elements with image numbers. Short keep lists work better than "keep everything." Always include a context clause from image 1 (lighting, camera angle) so the model knows which scene frame to preserve.

Can Krea 3 change the text on a poster or post? Yes. Put the new words in capitals exactly as they should read, and say which text stays the same. Check small lines of text in the result before publishing, because the model may not render fine detail perfectly.

What mode should I use for complex multi-reference edits? Start with Medium for the first run when you have multiple references and a detailed keep list — it gives higher fidelity and a creativity dial. Once you have a composition you like, switch to Instant for rapid iteration.

Sources

1. Civitai — Krea 3 is now on Civitai — launch announcement: reference-driven editing, up to 6 reference images, Instant and Medium modes, 1K/2K output, ~1MP recommended input, 30/60 Buzz launch week (33% off) until 22 Oct 2026 then 45/90 Buzz
2. Krea Edit user guide — Change Region, Annotate, Change Camera Angle, Change Lighting and other edit tools; edit-as-instruction structure
3. Morphic — Krea 3 guide — prompt formula by number, combining people and products, region editing, worked examples
4. Krea changelog — platform feature additions and model updates
5. artificialguybr/people-also-ask — PAA tracking; question "Does Krea 2 use negative prompts?" (collected 2026-10-15, file paa_en_krea_3.json)