Grok Imagine·xAIImage edit

Grok Imagine Image Edit

xAI Grok Imagine ile doğal dil talimatlarıyla görsel düzenleme. 8 referans görsele kadar; çoklu görsel düzenlemede <IMAGE_0>, <IMAGE_1> notasyonu desteklenir.

Çalışma alanında açxai/grok-imagine-image/edit

Parametreler

ParametreTürVarsayılanAralık veya seçenekler
aspect ratio
aspect_ratio

Aspect ratio of the generated image.

selectautoauto, 1:1, 3:4, 4:3, 9:16, 16:9, 2:3, 3:2, 9:19.5, 19.5:9, 9:20, 20:9, 1:2, 2:1
resolution
resolution

Output resolution.

select1k1k
num images
num_images

Number of images to generate.

select11, 2, 3, 4

Girdiler

Prompt'a ek olarak bu modelin aldığı dosyalar.

GirdiKabul ederEn fazla dosya
image_urls
image_urls
image5Zorunlu

Grok Imagine için prompt rehberi

TARGET MODEL: Grok Imagine (engine: Aurora — xAI autoregressive mixture-of-experts, NOT diffusion). It generates left-to-right like an LLM: THE FIRST SENTENCE CARRIES THE MOST WEIGHT.

Uyulması gereken kurallar

  • Flowing natural prose, never tag stacks. Order: subject + main action in the opening 20-30 words, then environment, then lighting, then style, then camera/technical.
  • Delete booster tags entirely: masterpiece, 8K, ultra detailed, ultra-realistic, breathtaking, stunning, hyperdetailed, best quality, award-winning. Delete weighting syntax: (term:1.4), (subject)++, [x], {x} — they do nothing on Aurora.
  • No negations: on an LLM-native model "no X" reintroduces X. Rephrase every exclusion as the desired positive state: "no blemishes" → "clear skin"; "no blur" → "sharp focus"; "no waves" → "calm sea"; "no people" → "empty, still scene". Two allowed idioms only: "no digital perfection, no smoothing" (anti-plastic skin) and "no additional text or labels" (suppress invented text).
  • Length 30-80 words. Cap the environment at 2-3 sensory elements — Aurora muddies busy scenes. Better direction beats more words.
  • In-image text: put the exact copy in quotes after a verb ("a sign reading 'OPEN LATE'"), short and single-line; font/color/effect as separate clauses; name the surface and placement ("neon on brick wall, top banner"). For non-Latin text write the actual glyphs. When the image should carry only that text, append "no additional text or labels".

En iyi sonuç verenler

  • Photorealism: name the instrument, not the adjective. A camera brief beats the word "photorealistic": "Shot on Leica M10, 35mm Summilux, f/1.4, shallow depth of field, natural film grain". Bodies that land: Leica M10, Sony A7R V, Canon EOS R5, Fujifilm X-T4. Focal feel: 24mm establishing / 35mm street / 50mm natural / 85mm portrait. Film stock carries color+grain in one token: "Kodak Portra 400 color tones" (adds warmth against Grok's cool default), "35mm film grain".
  • Name lighting classically and date it: golden hour, Rembrandt, rim light, chiaroscuro, three-point, backlit, volumetric god rays. "Early March morning" beats "morning"; "overcast afternoon in November" beats "cloudy".
  • Name colors ("electric blue and hot pink", never "colorful"). Mood words only when evocative: nostalgic, melancholic, tense, dreamlike — never happy/nice/cool.
  • Reinforce one critical surface by restating it in different words ("chrome bodysuit; reflective chrome") — Aurora's substitute for weighting.
  • Style: at most ONE movement + ONE technique ("cyberpunk aesthetic with impressionist painting technique"). Grok defaults to glossy 3D and cool editorial tones — for flat mediums (vector, pixel art, watercolor, anime) name the medium explicitly and EARLY; anime needs a hard cue ("80s OVA anime, cel-shaded" or "Makoto Shinkai art style").
  • Physical-detail realism: skin pores, fabric weave, condensation, water droplets; add "candid, not posing" if the subject reads staged.
  • If the user stated an aspect ratio, optionally echo it as the sentence tail ("…documentary feel, 16:9").

LUVI'nin bu aile için yazdığı manuelden; çalışma alanındaki yeniden yazıcı ve MCP prompt motoru aynı kuralları uygular.

Bu ailedeki diğer modeller