Deep dives on Midjourney, Stable Diffusion, Flux, LoRA, ControlNet, AI video and licensing.
Midjourney --stylize controls how literally the model follows your prompt. 0 is the most literal, 1000 the most creative. This article gives recommended values per scenario with samples.
Read full articleMidjourney --chaos controls how different the four candidate images are. 0 is near-identical, 100 is wildly divergent. Practical workflow tips inside.
Read full articleMidjourney --weird (0–3000) pushes the model toward unusual interpretations. This article explains four reference values and how --weird differs from --chaos and --stylize.
Read full articleMidjourney --ar controls aspect ratio. This article covers 6 common ratios — square, e-commerce, portrait, illustration, banner, vertical — and how the ratio shapes composition.
Read full articleMidjourney --seed is the key to reproducing images and building consistent series. This article explains how to capture a seed, reuse it across prompts, and combine it with --cref.
Read full articleMidjourney --cref fixes a character with a reference image. This article covers --cw weight, combination with seed and --sref, and the full mini-series workflow.
Read full articleMidjourney --sref uses an image to lock the visual style. This article explains --sw weight ranges, multi-image stacking, and combination with --cref.
Read full articleNiji 6 is Midjourney's anime model. This article explains the four --style sub-modes, recommended --s values, and how to write prompts for Japanese, Korean and Chinese animation looks.
Read full articleSDXL and Flux Dev are the two main open-source image models. This article compares them on quality, speed, LoRA ecosystem, prompt friendliness and commercial licensing.
Read full articleFlux Dev and Schnell are Black Forest Labs' two open models. Compare on quality, speed, VRAM and licensing — and learn the recommended hybrid workflow.
Read full articleStable Diffusion offers many samplers. This article compares four production-grade options — Euler a, DPM++ 2M Karras, UniPC, DDIM — with recommended step counts.
Read full articleCFG scale controls how literally Stable Diffusion follows the prompt. This article compares four reference values and per-model sweet spots for SD 1.5, SDXL and Flux.
Read full articleSampling steps control how many denoising iterations Stable Diffusion runs. This article gives per-sampler sweet spots and explains when to push higher.
Read full articleHires fix is Stable Diffusion's key way to push resolution. This article covers upscale ratio, denoising strength, and choosing between Latent, 4x-UltraSharp and ESRGAN.
Read full articleThere are hundreds of SDXL checkpoints. This article picks the most reliable ones across realistic portrait, anime, landscape and illustration, with selection notes.
Read full articleLoRA is a small add-on model that gives Stable Diffusion a specific style, character or concept. This article explains weight, trigger words, multi-LoRA stacking and common pitfalls.
Read full articleControlNet uses a reference image to control composition, pose, outline or depth. This article explains four core preprocessors — OpenPose, Canny, Depth, Lineart.
Read full articleOpenPose locks AI character poses using skeleton points. This article covers OpenPose, OpenPose Full and DWPose differences, weight settings and standard workflow.
Read full articleCanny extracts edges; Depth extracts spatial relationships. This article gives four scenarios with a clear pick.
Read full articleDifferent platforms use different weight syntax. This article compares Automatic1111, ComfyUI, Midjourney and Flux.
Read full articleBREAK is Automatic1111's prompt grouping keyword. This article explains how to use BREAK to bypass the 75 token limit and structure long prompts.
Read full articleKeeping two characters consistent in the same image is hard. This article covers regional prompting, multi-LoRA stacking and the compositing workflow.
Read full articleRunway Gen-3 is highly sensitive to camera motion and action description. This article covers camera vocabulary, single-action phrasing, stability cues and image-to-video tricks.
Read full articleKling is a Chinese AI video model that handles Chinese prompts exceptionally well. This article covers camera vocabulary, action phrasing and image-to-video tips for English users too.
Read full articleA horizontal comparison of major AI video models: Runway, Pika, Kling, Sora, Seedance, Hailuo and Veo, with scenario-based recommendations.
Read full articleOpenAI Sora supports natural-language long descriptions and 10–60 second shots. This article covers narrative-style writing, multi-shot sequences and pitfalls.
Read full articleLens language is the vocabulary of cinematography in AI prompts. This article catalogs shot size, camera motion, focal length and pacing terms.
Read full articleCan AI images be used commercially? This article covers licensing for Midjourney, SDXL, Flux Dev/Schnell and LoRAs, plus four major risk areas.
Read full articleAI-generated hands are notorious for distortion. This article covers three solutions: negative prompts, ControlNet, and inpaint repainting.
Read full articleAI-generated text is notoriously unreliable. This article compares Flux, Ideogram and DALL-E 3 on text rendering and explains the overlay workflow.
Read full articleHex codes and Pantone numbers do not work in AI prompts. This article explains how to write descriptive color phrases and bring brand colors closer.
Read full articleColor vocabulary for traditional stylized illustration prompts: ink wash, mural, celadon, gongbi, seasonal palettes.
Read full articleAI portraits tend toward waxwork faces. This article gives 8 skin-texture keywords, age-band variations and LoRA recommendations.
Read full articleVocabulary for portrait photography prompts: focal length, depth of field, composition and light position.
Read full articleAI product shots often invent gibberish labels. This article shows the suppression syntax for Midjourney, SDXL and Flux.
Read full article