All Articles

Browse all articles on this site by category.


Getting Started Roadmap

If you’re new here, we recommend reading in this order:

  1. Why z-image-turbo? — Why we chose this model
  2. Getting Started with ConoHa AI Canvas — Setting up your environment
  3. Prompt Fundamentals — Learn word order and emphasis syntax through experiments
  4. Prompt Design Thinking — How to design your own prompts
  5. Pose Prompt Collection — Practice with examples

Tutorials

ArticleLevelDescription
Getting Started with ConoHa AI CanvasBeginnerStart AI image generation with just a browser
ComfyUI Workflow DownloadBeginner–Intermediatez-image-turbo workflow (negative prompt pre-configured)
RunPod Serverless Setup GuideAdvancedBuild an API environment from scratch

Reviews & Comparisons

ArticleDescription
Why z-image-turbo?Model lineage (SD/Flux/z-image), feature comparison, licensing & NSFW support
Cloud GPU ComparisonRunPod vs Vast.ai vs ConoHa AI Canvas — pricing & performance

Tips & Techniques

Fundamentals

ArticleDescription
Prompt FundamentalsWord order rules, emphasis syntax, negative prompts (with experiment images)
Prompt Design ThinkingFrom copying to creating — the 5-element decomposition approach
Complete Negative Prompt GuideCategory-by-category standard phrases and customization methods

Hands-On Experiments

ArticleDescription
Improving a Beach Bikini Prompt in 8 StepsAdding one element at a time from a minimal prompt
Reproducing NB2 Prompts with z-image-turboConverting 10 Nano Banana 2 themes to SD-based generation
Prompt Refinement: Comparing “Trim” vs “Add” on 10 ThemesFinding unnecessary words and effective additions

Prompt Improvement Series

Starting from simple prompts and iterating to achieve the intended image. 3 images generated at each step to confirm stability.

ArticleThemeKey Point
Café SnapshotWindow-side café snapshotSucceeded immediately. Long descriptions for amateur feel worked well
Retro IzakayaRetro izakaya (Japanese pub)Flash, grain, and paper lanterns all reproduced
Night BeachNight beach cinematicBacklight + moonlight + dress transparency
Library EmoAfter-school library emo compositionResolved fluorescent vs sunset lighting conflict
Mirror SelfieMirror selfieHardest theme. 60-70% success rate
Rooftop FashionRooftop fashion magazine styleSolved the “magazine” keyword causing logo artifacts
Tokyo Neon StreetRainy Tokyo neon streetwet pavement reflections was the key
Morning Bed IntimateMorning bed intimate portraitSoft morning light through curtains with white sheets
Summer Festival PolaroidSummer festival polaroid stylecotton candy in hand stabilized hand rendering

Body Expression Experiments

ArticleDescription
The Ultimate Breast Size Prompt GuideConclusions from 30+ conditions. Max 11 words, efficient 5 words, plus what doesn’t work
Maximizing Breast Size: A Comprehensive Prompt ComparisonSummary of 30+ conditions tested. Best descriptors and key findings
Maximizing Butt Size: A Comprehensive Prompt Comparison27 conditions tested at front/rear angles. Stacking and body modifier effects
Prompts for “Saggy Breasts”31 conditions, 186 images. Only size words + spatial descriptions partially effective
Prompts for “Flat Chest”35 conditions, 192 images. None worked. Breast size reduction not possible with z-image-turbo

Effective Descriptors

ArticleDescription
Size Adjective Step-by-Step Comparisonlarge → huge → gigantic effect. Plateaus at huge
Stacking Effect and Token Volume HypothesisStacking adjectives adds size. Even repeating the same word works
“oversized” is the Strongest AdjectiveAchieve near-maximum with just 5 words — best cost-performance
Hiding Nipples Through Outfit ChangesEffects of padded bikini top and tight crop top

Ineffective Descriptors

ArticleResult
Cup Size NotationG-cup, K-cup, double D → No effect
cm Notation120cm, 150cm → No effect
Weight Syntax(huge breasts:1.6) → No effect
Slang & MetaphorsBombs, watermelons, and cows appeared
Japanese Romaji “oppai”No effect
super/strong PrefixesNo effect alone
Unified “breasts” TestCannot hide nipples (N=13)

Prompt Element Experiments

ArticleDescription
CLIP Chunk Splitting ExperimentWhat happens beyond 75 tokens. Experimentally testing 2nd chunk influence
Does “coherent anatomy” Work?Tested with 24 images → No effect on z-image-turbo. Recommend removing (saves 7 tokens)
Does “actress” Change Faces? Profession Comparison7 professions × 3 images. actress/model tend to produce more refined faces
God Prompt Ablation StudyRemoving elements one by one to verify necessity. Up to 50% token reduction achieved

Curated Prompts

ArticleDescription
3 “God Prompts” That Never MissAblation-verified minimal versions. Café 42 tokens, Summer Festival 48 tokens, Morning Bed 79 tokens

Prompt Collections

ArticleImagesNSFWDescription
Portrait Prompt CollectionSFWBasic examples by situation
Pose Prompt Collection12SFWArms crossed, reading, walking, yoga, etc.
Lighting Prompt Collection8SFWSame subject, different lighting comparison
Expression Prompt Collection8SFWSmile, serious, surprised — 8 patterns
Gravure Prompt Collection12NSFWSwimsuit, lingerie, body line

Paper Explanations

ArticlePaperRelevance to AI Image Generation
CLIP Paper ExplainedRadford et al., 2021Prompt vectorization, origin of the 75-token limit
LDM Paper ExplainedRombach et al., 2022Stable Diffusion architecture
CFG Paper ExplainedHo & Salimans, 2022Theoretical basis of negative prompts