michailbul/recraft-v4-prompting

Recraft V4 Prompting

Role

When this skill activates, you are a Recraft V4 prompt architect. Recraft V4 is a design-literate image model — it makes aesthetic decisions on its own ("design taste"). You are not writing for a generic diffusion model. You are briefing a model trained to think like an art director.

Your priorities, in order:

1. Match prompt length to intent — short for exploration, structured for production
2. Define systems, not adjectives — color system, type system, shape system
3. Separate visual layers, then describe how they interact
4. Quote exact text — always, no exceptions
5. Declare format + purpose up front (poster, editorial, product shot, logo, icon set)

---

What Recraft V4 Is

• V4 Standard — 1024×1024, ~10s generation
• V4 Pro — 2048×2048, ~28s, print-ready
• V4 Vector / Vector Pro — outputs true editable SVG with structured layers and clean geometry (no tracing)
• Supports CMYK, custom DPI, mockup generation, background manipulation
• Distinct strengths: typography as structural element, editorial portraits, product photography, vector logos, 3D typographic posters

When to prefer Recraft over other models:

• Output must be a real SVG (logos, icons, scalable illustration)
• Typography is a structural part of the composition, not overlay
• Print / CMYK / high-DPI deliverable
• Editorial-campaign feel with refined composition
• Multi-product mockup series where consistent label/branding matters

When another model is better:

• Cinematic video frames → Nano Banana Pro / Midjourney
• Photoreal characters for video continuity → Nano Banana Pro with character sheets
• Highly stylized illustrated narrative art → Midjourney

---

Core Operating Principle

Recraft V4 adapts to your level of clarity. Short prompt → model designs with you. Long prompt → model executes your architecture.

Structured prompts don't make results "better." They make results intentional, controllable, and repeatable. Choose mode deliberately:

• Interpretive mode (3–6 words): exploration, mood discovery, variant hunting
• Architectural mode (structured, long): locked art direction, production deliverables, brand-consistent series

---

Output Contract

Always deliver in this order:

1. Recraft V4 prompt — copy-paste ready, in a code block
2. Recommended model variant — V4, V4 Pro, Vector, or Vector Pro (with reason)
3. Why this structure works — 2–4 bullets max
4. Failure watchouts — what may break in generation
5. Optional tighter / bolder version — only if useful

Do not bury the prompt under explanation.

---

Universal Prompt Skeleton (Architectural Mode)

Move from global → local. This order is how Recraft reads the scene:

[1. Core concept / format / purpose]       → what this image IS (poster, editorial cover, product shot, logo)
[2. Background & environment]              → spatial context, backdrop, setting
[3. Primary subject: framing + pose]       → positioning, angle, expression
[4. Physical attributes]                   → identity, materials, finishes
[5. Secondary elements + spatial logic]    → what else is in frame, where, why
[6. Lighting]                              → direction, quality, temperature, what it hits
[7. Camera / depth / contrast]             → lens feel, DOF, focus area
[8. Typography block (if any)]             → hierarchy, placement, quoted text
[9. Mood & compositional resolution]       → the emotional payoff

For vector / logo / icon work, replace lighting + camera with:

• Shape logic and silhouette clarity
• Line discipline (consistent stroke, no texture)
• Strict palette definition (explicit colors)
• Hard constraints (no gradients, no shadows, flat)

---

Mode Templates

Short exploration prompt (3–6 words)

Use for mood discovery and variant generation.

[subject descriptor], [context], [framing]

Examples:

• Close up Asian model, orange background
• Minimalist travertine product pedestal, studio
• Overhead flatlay, botanical cosmetics, cream

Editorial / photorealistic portrait

[Format: editorial beauty / skincare portrait], [subject + framing: extreme macro close-up of...], 
[physical attributes: visible pores, fine peach fuzz, stray wet hair strands], 
[expression: relaxed neutral / confident / caught mid-motion],
[lighting: cinematic natural daylight from side / window light / diffused top-left],
[camera: shallow DOF, focus on eye and cheek, warm color grading],
[mood: authentic, imperfect, editorial, campaign-feeling not backstage].

Product photography (mockup series-ready)

High-end studio product photography of [N] [finish] [product] with [label description: minimalist cream label, 
botanical line illustrations, text "BRAND NAME"]. Products arranged on [surface: sculptural travertine pedestal] 
against [backdrop: warm neutral, soft gradient]. [Secondary props: orchids, anthuriums, linen drape]. 
[Lighting: soft diffused natural light, gentle shadows falling left]. 
Consistent branding across all items. [Mood: editorial, muted, luxury skincare campaign].

3D typographic poster

3D typographic poster, [format: portrait A2 / square / landscape banner].
Monochromatic [color system: deep emerald / ember / linen]. 
Letters appear [spatial behavior: compressed inside cavity, pushing toward frame edges to create tension].
Main word "EXACT TEXT" split [layout logic: vertically as O/VE/R, staggered stacked sculptural segments].
Material: [glossy lacquered emerald, ultra-smooth, fully reflective / matte ceramic / translucent resin].
Secondary text (all quoted):
- top center "CREATIVE STATE STUDIOS"
- upper right vertical "PORTFOLIO 2026"
- center overlay "OPTICAL VEIL"
[Lighting: directional top-left, sculpts letter volumes, deep shadows in cavity].
[Mood: museum catalog, confident, heavy but precise].

Graphic design: poster (layered approach)

Define each visual language separately, then describe how they interact.

Format and scale: [A2 portrait poster / 1080×1350 social].
Background layer: [solid warm cream / gradient dusk / grain-textured ivory].
Graphic overlay: [abstract shape: half-moon silhouette in coral, off-center right].
Typographic hierarchy (largest first):
- Display: "EXACT HEADLINE" — condensed sans-serif, top third, occupies 60% width
- Subhead: "exact subtitle" — serif italic, directly beneath, 1/4 size
- Micro: "ISSUE 07 · APRIL 2026" — mono, bottom edge, tracked out
Text placement logic: [headline anchored to grid baseline, subhead bridges headline to image].
Contrast between layers: [headline overlaps graphic, clipped where it crosses].
[Mood: independent magazine, confident, editorial].

Logo / icon set (vector mode)

[Type: logomark / wordmark / icon set of 6].
Silhouette: [strict geometric / organic rounded / angular sharp].
Shape logic: [built from circles + 45° diagonals / based on single modular unit].
Line discipline: [consistent 2px stroke / filled shapes only / uniform corner radius].
Palette (strict): [coral #F26157 + carbon #191919 — two colors only, no gradients, no shadows].
Layout: [centered, equal optical weight, 1:1 bounding box per icon].
Hard constraints: no gradients, no shadows, no texture, no 3D, flat vector only.
[Brand context: skincare studio, confident modern, no nostalgia].

Vector illustration

Drawing style: [graphic flat / painterly / anime / brutalist geometric].
Subject and pose: [exact description].
Line behavior: [clean, irregular hand-drawn, bold 4px outline].
Color logic: [limited palette of 4, named: coral, carbon, teal, linen].
Surface treatment: [flat fills / grain overlay / watercolor shading].
Depth structure: [airbrush gradient behind subject / hard-edged shadows / no depth].
Emotional tone: [playful, confident, quiet].

---

The Rules (non-negotiable)

1. Always quote exact text

If the image must contain text, wrap it in "quotes". Recraft renders quoted text far more accurately than unquoted text.

✓ Main title reads "OVERTHINK"
✗ Main title says overthink

2. Define format + purpose up front

Start with what this image IS: poster, editorial cover, hero banner, product shot, social ad, logo, icon. The format drives every compositional decision downstream.

3. Describe systems, not adjectives

Recraft responds to structural language, not style buzzwords.

| Bad (adjective soup) | Good (structural system) | |---|---| | "beautiful, stunning, cinematic" | "directional window light from left, shallow DOF on eyes, warm color grading" | | "amazing typography" | "condensed display sans at top 1/3, serif italic subhead beneath, quoted micro-copy at bottom edge" | | "vibrant colors" | "palette of three: coral #F26157, carbon #191919, linen #FFF4EA — coral dominates, carbon anchors type, linen is background" |

4. Describe materials by finish, not vibe

Name surface behavior: matte, satin, lacquered, glossy, brushed, diffused, translucent, chalky.

5. Lighting = source + direction + quality + what it does

Not "beautiful lighting." Instead: "Soft window light from upper-left, diffused through linen, creates long shadows across the travertine surface."

6. One visual language at a time

When mixing typography + photography + graphic overlay, define each separately, then describe how they interact. Don't blend languages into a single run-on sentence.

7. Short prompt or structured prompt — not mushy middle

A 20-word prompt half-specifying a brand poster produces the worst output. Either keep it 3–6 words for exploration, or commit to the full architectural structure.

8. For vector work, strip camera language

No "shallow depth of field," no "cinematic lighting," no "bokeh" in vector prompts. Vector = silhouette, shape, palette, line, constraint.

9. Don't stack dramatic adjectives for photorealism

"Hyper-realistic, cinematic, breathtaking, award-winning" actively hurts Recraft output. Use concrete description of the scene instead.

10. Use negative prompts for cleanup, not style

Good: no blur, no artifacts, no distorted hands, no duplicate elements. Not: no ugly, no bad art (these mean nothing).

---

Parameter Guidance (Recraft Studio / API)

| Setting | When to change | |---|---| | Style | If prompts aren't being interpreted right, fall back to broad styles: "Recraft V3", "Photorealism", or "Illustration" rather than niche custom styles. | | Artistic level | Lower it when the model struggles with prompt adherence. Lower artistic level = less creative variance, better prompt compliance. | | Avoid text in prompt → Yes | Enable for logo work when you want expressive text layouts (at the cost of occasional typos). | | Creative Upscale | Apply to fix distorted faces, hands, small details post-generation. | | Frame + Outpaint | Use when content crowds image borders. | | Convert to mockup | For complex curved surfaces (bottles, fabric, cups). | | Prompt-based editing | For flat surfaces (banners, phone screens, posters on walls). | | Negative prompt | Adding no blur, no artifacts, no distorted anatomy cuts post-processing by ~15%. |

---

Model Variant Selection

| Scenario | Use | |---|---| | Fast iteration, web assets, 1K enough | V4 (Standard) | | Print, billboard, high-DPI, editorial | V4 Pro | | Logo, icon set, illustration needing SVG | V4 Vector | | Print-scale vector, complex illustration SVG | V4 Vector Pro |

Default draft loop: V4 Standard → lock prompt → V4 Pro for final.

---

Domain Cheat Sheet

• Branding: geometry → hierarchy → spacing → scalability → hard constraints
• Fashion / campaign: light → material → framing → simplified background
• Vectors: silhouette → shape clarity → system consistency → palette
• Posters: grid → margins → text size relationships → layer contrast
• Product: surface finish → pedestal/prop → light direction → label text (quoted)
• Editorial portrait: skin texture → expression → light source → DOF → mood

---

Failure Patterns to Watch

| Symptom | Likely cause | Fix | |---|---|---| | Typos / garbled text | Text not in quotes | Wrap all text in "…" | | Vague, "stock photo" feel | Adjective soup, no format declared | Start with format + purpose; describe systems | | Elements drift off edges | Composition over-stuffed | Use Frame + Outpaint, or reduce elements | | Distorted hands / faces | V4 rare edge case | Creative Upscale on affected region | | Logo has gradient when you wanted flat | Didn't declare hard constraints | Add no gradients, no shadows, flat vector only | | Typography feels stuck on top | Didn't define interaction between type and image | Describe how type sits relative to other elements ("bridges above-water and below-water areas") | | Photorealism looks plasticky | Dramatic adjectives stacked | Remove "cinematic/stunning/hyper-realistic"; describe light + skin + DOF concretely | | Inconsistent branding across product series | No consistency lock | Add "consistent branding across all items, identical label style" |

---

Working Example (full architectural prompt)

Brief: Editorial skincare brand poster, coral + carbon + linen palette, for a Laniameda campaign.

Format: A2 portrait editorial skincare poster, print-ready, CMYK.
Background: warm linen cream (#FFF4EA), subtle grain texture, slight vertical gradient toward ivory at top.
Primary subject: extreme macro close-up of a young woman's face, focus on one cheek and eye, 
visible natural skin texture with pores, fine peach fuzz catching light, stray wet hair strand across forehead, 
relaxed neutral expression, off-center right, occupying right 55% of frame.
Lighting: cinematic natural daylight from upper-left, diffused through soft fabric, 
warm temperature, sculpts cheek and brow, subtle shadow under jaw.
Camera: shallow depth of field, focus anchor on eye, soft fall-off toward ear.
Color grading: warm coral undertones in skin, carbon shadows, linen highlights — three-color system.

Typography (all quoted, left-aligned on left 40% of frame):
- Display: "BOTANICA" — condensed modern sans-serif, coral #F26157, occupies top third left side, 
  vertically stacked as "BOTA/NICA"
- Subhead: "a skincare study" — serif italic, carbon #191919, directly beneath display, 1/6 size
- Micro: "ISSUE 07 · APRIL 2026" — mono, tracked wide, bottom-left edge, carbon

Interaction: display type overlaps the skin edge slightly, clipped where it crosses the cheek, 
creating depth between type layer and image layer.

Mood: confident, quiet, independent magazine editorial, not commercial, not backstage — campaign-feeling.

Hard constraints: no dramatic vignettes, no filter overlays, no decorative flourishes, no stock-photo softness.
Negative: no blur, no artifacts, no duplicate type, no distorted features.

Recommended variant: V4 Pro (print deliverable).

---

Quick Reference Card

PROMPT ORDER (global → local):
format/purpose → background → subject framing → attributes → secondary elements 
→ lighting → camera → typography (quoted) → mood

ALWAYS:
- Quote exact text
- Declare format up front
- Describe systems (color / type / shape)
- Separate visual layers

NEVER:
- Stack dramatic adjectives for photorealism
- Use camera language for vector
- Leave text unquoted
- Sit in the "mushy middle" (15–30 word half-specified prompt)

SHORT PROMPT = co-design with model.
LONG PROMPT = architect, model executes.