The strongest look for the least effort
Midjourney has taste built in. Short prompts come back styled, and the flags handle the frame. You trade fine control for that polish. Pick it when mood matters more than precision.
Tools
The toolbox
Build the prompt here, then take it to your generator. Each page below covers one tool family: its flags, its settings, or its dialect. The words themselves live on the walls and work everywhere.
Pick your generator
flag driven
🌀
Midjourney
The parameter flags that matter: --ar, --stylize, --chaos, and the character reference. Ends every prompt.
Open the page → settings driven 🧪Stable Diffusion
Samplers, steps, CFG, pixel sizes, and the negative prompt. The most control and the most dials.
Open the page → sentence driven 💬Chat image models
Nano Banana, GPT-Image, and their kin read plain sentences and edit in conversation. No flags at all.
Open the guide → prompt writing ✍️LLM Prompting
Use a chat model as your prompt writer: templates, system prompts, and guides for the writing side.
Open the hub →Which one should I use?
Match the tool to the job. Every family reads the same wall words. They differ in how much control you get and in how you hand the words over. The prompt skeleton from Start Here works in all of them.
Full control on your own machine
Stable Diffusion exposes every dial: samplers, steps, CFG, word weights, negative prompts. It runs on your hardware for free. Pick it for repeatable series and fine adjustments.
Edits, readable text, conversation
Nano Banana, GPT-Image, and the other chat image models follow long instructions, render legible text, and revise on request. Pick one when you would rather ask again than reroll.
A writing partner for the prompt itself
Any chat model can draft your keyword stacks. Hand it wall words as a vocabulary and let it write variations. Pick this route when the blank prompt box is the problem.
One picture, three dialects
The wall words survive the move between tools. The packaging around them changes. Here is the same fox in each dialect.
Midjourneyred fox, alpine meadow, golden hour, telephoto lens, shallow depth of field --ar 3:2 --stylize 250
Stable Diffusionred fox, alpine meadow, (golden hour:1.2), telephoto lens, bokeh
Negative prompt: blurry, watermark, extra limbs
Negative prompt: blurry, watermark, extra limbs
Chat image modelsA red fox stands in an alpine meadow at golden hour. Telephoto framing, shallow depth of field, the fox sharp against soft grass.
Midjourney and Stable Diffusion read a stack. Order sets priority, commas separate ideas, and flags or weights set the frame. Front-load what matters and keep the stack short.
The chat image models read a request. Full sentences work, and the second message counts as much as the first. Instead of a reroll you say what to change: "make it dusk, add ground mist, keep the fox." The chat image models guide covers that edit loop.
Two habits transfer everywhere. Change one thing at a time so you can tell what did what. And when a result goes wrong, name the symptom before you rewrite: Fix the Picture maps symptoms to cures.
Guides that pair with the tools
Start Here: Your First Good Prompt
The skeleton every tool on this page accepts: subject, style, lighting, and one or two details.
15 Tips to Not Suck at MidJourney
A checklist for when a Midjourney image misses: subject first, then style, color, viewpoint, and iteration.
Prompt Engineering in Stable Diffusion
The long version of the Stable Diffusion page: from subject choice to weights, one clear job per prompt.
Prompting Chat Image Models
Nano Banana and GPT-Image: the sentence dialect, the edit loop, and where wall words fit.
Using ChatGPT to Write Image Prompts
The bridge between the chat side and the walls: a chat model that writes your keyword stacks.
Fix the Picture
A symptom-to-cure guide for any generator: ignored subjects, mangled hands, muddy color, garbled text.