Fill the blanks, pick from the menus, copy a finished prompt for your tool. Built on the Ohneis system: goal first, context second, finish last.
--raw, --s and --sref.Pick a lighting setup for a complete rig, like three-point product, clamshell beauty or Rembrandt portrait. Or build the light by hand from source, direction, quality, color and ratio. Two or three sources at most.
Set a ratio to control contrast: 2:1 for beauty, 3:1 for portraits, 4:1 for drama, 8:1 for noir.
References get one job each. Use the "only" roles to borrow one thing: composition, lighting, color or atmosphere. Say what each one must not change.
Character traits that never change go in their own field: face, age, skin, hair, marks. Clothes, pose and light stay free.
Variations run in packs: one change at a time, a five-shot product catalog, the five-prompt rule, safe, recommended and bold, or ad coverage. Re-attach the original reference for every shot.
Alphaprompt turns your choices into a style engine. Paste it into a fresh chat, one chat per style, then send short ideas. Save your look in the style library and apply it to a new subject in one click.
Generate four to eight options. Reject structural failures first. Run the Review checklist on the best two. If it's mid, it's out.
Score the image out of ten on intent, composition, lighting, material and style. The lowest score picks your repair lane. Tick the fixes you want and copy the repair prompt. Low intent means pivot: rewrite the prompt instead of patching it.
The helper prompts go to Claude or ChatGPT with your image: critique, gap diagnosis against a reference, A/B test, reference analysis and a vocabulary builder.
Save every prompt to the Log with a note on what changed. Your winners become master prompts.
Import a brief sits at the top of the form. Paste the brief or upload a PDF, text file or screenshot, then add reference images in order. Claude fills the menus the brief supports, assigns each reference a role and lists what to ask the client.
Critique an output sits on the Review tab. Upload what you generated. Claude scores it against your brief and prompt, ticks the quality check, picks the repair lane and writes the repair prompt.
Inside Claude, these calls use your Claude plan. In the standalone file, they use the model you pick in AI settings: your Anthropic API key, billed separately in the Anthropic Console, or a free Ollama model on your computer.
Pick a lighting setup or framing and a diagram shows it: a top-down light plot, the framing in a 4:5 frame and the camera height.
In Midjourney mode, the Style tag menu holds the 30 tags from the Style Tag Library. One look, every prompt, change the subject words, never the tag.
The Library tab holds every prompt from your prompt bundle. Search it, copy a prompt, or load one into the builder as menu choices.
Built from the Ohneis AI Systems Classroom: Basics Prompt Mastery, the Visual Production System (foundation, signature style, prompt engineering, precision control, visual architecture, lighting, color, refinement, quality control, video, motion control, character consistency, tools), the Templates & Prompts library and the video breakdowns.
Saved styles and the log live in this browser only.
Model names and limits match the course as of late October 2026. Tools change fast. A prompt sets intent. It does not guarantee exact products, faces, text or anatomy. Review every output.